APG analysis
Analysis walkthrough for “Tuning in to non-adjacencies: Exposure to learnable patterns supports discovering otherwise difficult structures” (Zettersten, Potter & Saffran).
For details on the rationale for individual analyses, see the corresponding paper.
Experiment 1
Test Accuracy
Accuracy Plot
#Familiar X Test
p1 <- ggplot(subset(test_type_exp,testType=="familiarX"&exp=="exp1"),aes(condition,accuracy,fill=condition, color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="familiarX"&exp=="exp1"),aes(y=acc), width = 0.05, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, size=0.8)+
theme_classic(base_size=16)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Non-Learnable Pre-Exposure","Learnable Pre-Exposure"),
labels=c("Non-Learnable\nPre-Exposure","Learnable\nPre-Exposure"),
limits=c("Non-Learnable Pre-Exposure","Learnable Pre-Exposure"))+
scale_fill_manual(values=c("#4DAF4A","#E41A1C"))+
scale_color_manual(values=c("#4DAF4A","#E41A1C"))+
theme(legend.position="none")+
ylab("Accuracy - Familiar X Test")+
xlab("Condition")
#Novel X
p2 <- ggplot(subset(test_type_exp,testType=="novelX"&exp=="exp1"),aes(condition,accuracy,fill=condition,color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="novelX"&exp=="exp1"),aes(y=acc), width = 0.05, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, size=0.8)+
theme_classic(base_size=16)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Non-Learnable Pre-Exposure","Learnable Pre-Exposure"),
labels=c("Non-Learnable\nPre-Exposure","Learnable\nPre-Exposure"),
limits=c("Non-Learnable Pre-Exposure","Learnable Pre-Exposure"))+
scale_fill_manual(values=c("#4DAF4A","#E41A1C"))+
scale_color_manual(values=c("#4DAF4A","#E41A1C"))+
theme(legend.position="none")+
ylab("Accuracy - Novel X Test")+
xlab("Condition")
plot_grid(p1,p2, labels=c("A","B"),label_size=18)
Overall Accuracy Table
overall_exp %>%
filter(exp=="exp1") %>%
select(-c(se, accuracy_ci,d_prime_ci,c_bias_ci)) %>%
formattable()| exp | condition | N | accuracy | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|---|
| exp1 | Learnable Pre-Exposure | 32 | 0.6197917 | 0.5503856 | 0.6891977 | 0.8180681 | 0.33183941 | 1.3042967 | -0.1721116 | -0.2965946 | -0.0476286 |
| exp1 | Non-Learnable Pre-Exposure | 35 | 0.5293651 | 0.4935915 | 0.5651386 | 0.2026126 | -0.04537869 | 0.4506038 | -0.3028788 | -0.4757694 | -0.1299883 |
Accuracy Table By Test Type
test_type_exp %>%
filter(exp=="exp1") %>%
select(-c(se, accuracy_ci,d_prime_ci,c_bias_ci)) %>%
formattable()| exp | condition | testType | N | accuracy | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| exp1 | Learnable Pre-Exposure | familiarX | 32 | 0.6197917 | 0.5505723 | 0.6890110 | 0.7511226 | 0.32051674 | 1.1817285 | -0.24820588 | -0.4281889 | -0.06822284 |
| exp1 | Learnable Pre-Exposure | novelX | 32 | 0.6197917 | 0.5436288 | 0.6959545 | 0.7517021 | 0.28191789 | 1.2214863 | -0.08924026 | -0.2107732 | 0.03229271 |
| exp1 | Non-Learnable Pre-Exposure | familiarX | 35 | 0.5269841 | 0.4794718 | 0.5744965 | 0.1608595 | -0.12636554 | 0.4480845 | -0.47197810 | -0.6295813 | -0.31437490 |
| exp1 | Non-Learnable Pre-Exposure | novelX | 35 | 0.5317460 | 0.4921224 | 0.5713697 | 0.1901106 | -0.04401717 | 0.4242384 | -0.11014830 | -0.3209422 | 0.10064563 |
Correlation
Correlation between Familiar X and Novel X Trials: Learnable Pre-Exposure
##Correlations between Familiar X and Novel X
c <- corr.test(subset(subj_accuracy_wide, condition=="Learnable Pre-Exposure" & exp=="exp1")[,c("novelX","familiarX")])
c## Call:corr.test(x = subset(subj_accuracy_wide, condition == "Learnable Pre-Exposure" &
## exp == "exp1")[, c("novelX", "familiarX")])
## Correlation matrix
## novelX familiarX
## novelX 1.00 0.82
## familiarX 0.82 1.00
## Sample Size
## [1] 32
## Probability values (Entries above the diagonal are adjusted for multiple tests.)
## novelX familiarX
## novelX 0 0
## familiarX 0 0
##
## To see confidence intervals of the correlations, print with the short=FALSE optionc$p## novelX familiarX
## novelX 0.000000e+00 7.454645e-09
## familiarX 7.454645e-09 0.000000e+00Correlation between Familiar X and Novel X Trials: Non-Learnable Pre-Exposure
c <- corr.test(subset(subj_accuracy_wide, condition=="Non-Learnable Pre-Exposure" & exp=="exp1")[,c("novelX","familiarX")])
c## Call:corr.test(x = subset(subj_accuracy_wide, condition == "Non-Learnable Pre-Exposure" &
## exp == "exp1")[, c("novelX", "familiarX")])
## Correlation matrix
## novelX familiarX
## novelX 1.00 0.34
## familiarX 0.34 1.00
## Sample Size
## [1] 35
## Probability values (Entries above the diagonal are adjusted for multiple tests.)
## novelX familiarX
## novelX 0.00 0.04
## familiarX 0.04 0.00
##
## To see confidence intervals of the correlations, print with the short=FALSE optionc$p## novelX familiarX
## novelX 0.00000000 0.04366703
## familiarX 0.04366703 0.00000000Compute correlation without outlier participant with perfect Familiar X and Novel X accuracy
#check outlier in Non-Learnable Pre-Exposure Condition
outlierTest(lm(novelX~familiarX,subset(subj_accuracy_wide,exp=="exp1"&condition=="Non-Learnable Pre-Exposure")))## rstudent unadjusted p-value Bonferroni p
## 34 5.040898 1.7644e-05 0.00061755#correlation without outlier
c <- corr.test(subset(subj_accuracy_wide, condition=="Non-Learnable Pre-Exposure" & exp=="exp1"&participantCode!="apg_exp1_p7")[,c("novelX","familiarX")])
c## Call:corr.test(x = subset(subj_accuracy_wide, condition == "Non-Learnable Pre-Exposure" &
## exp == "exp1" & participantCode != "apg_exp1_p7")[, c("novelX",
## "familiarX")])
## Correlation matrix
## novelX familiarX
## novelX 1.00 -0.14
## familiarX -0.14 1.00
## Sample Size
## [1] 34
## Probability values (Entries above the diagonal are adjusted for multiple tests.)
## novelX familiarX
## novelX 0.00 0.44
## familiarX 0.44 0.00
##
## To see confidence intervals of the correlations, print with the short=FALSE optionc$p## novelX familiarX
## novelX 0.0000000 0.4434494
## familiarX 0.4434494 0.0000000Interaction between Test Type and Condition
m <- lm(novelX~familiarX*condition, subset(subj_accuracy_wide,exp=="exp1"))
summary(m)##
## Call:
## lm(formula = novelX ~ familiarX * condition, data = subset(subj_accuracy_wide,
## exp == "exp1"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.26809 -0.06266 -0.01132 0.04742 0.33776
##
## Coefficients:
## Estimate Std. Error t value
## (Intercept) 0.05859 0.07023 0.834
## familiarX 0.90546 0.10839 8.354
## conditionNon-Learnable Pre-Exposure 0.32240 0.10511 3.067
## familiarX:conditionNon-Learnable Pre-Exposure -0.61940 0.17996 -3.442
## Pr(>|t|)
## (Intercept) 0.40727
## familiarX 8.6e-12 ***
## conditionNon-Learnable Pre-Exposure 0.00318 **
## familiarX:conditionNon-Learnable Pre-Exposure 0.00103 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.1159 on 63 degrees of freedom
## Multiple R-squared: 0.5697, Adjusted R-squared: 0.5492
## F-statistic: 27.8 on 3 and 63 DF, p-value: 1.425e-11Correlation Plot
ggplot(subset(subj_accuracy_wide,exp=="exp1"),aes(familiarX,novelX, color=condition,linetype=condition,shape=condition))+
geom_jitter(width=0.02,height=0.02,size=3)+
geom_smooth(method=lm,se =F,size=1.5)+
scale_color_manual(name="Condition",values=c("#4DAF4A","#E41A1C"))+
scale_linetype_discrete(name="Condition")+
scale_shape_discrete(name="Condition")+
ylab("Accuracy - Novel X")+
xlab("Accuracy - Familiar X")+
theme_classic(base_size=18)+
theme(legend.position=c(0.3,0.9))
Logistic Mixed-Effects Model
Overall
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
##all data
#recode condition
d$conditionC <- ifelse(d$condition=="Learnable Pre-Exposure",0.5,
ifelse(d$condition=="Non-Learnable Pre-Exposure",-0.5,NA))
m <- glmer(isRight~conditionC+(1|participantCode)+(0+conditionC|stimulus),data=subset(d,exp=="exp1"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (0 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp1")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3187.2 3210.4 -1589.6 3179.2 2408
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.8145 -1.0146 0.3718 0.9347 1.3215
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.45292 0.6730
## stimulus conditionC 0.03135 0.1771
## Number of obs: 2412, groups: participantCode, 67; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.36233 0.09396 3.856 0.000115 ***
## conditionC 0.45319 0.18959 2.390 0.016831 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.070confint(m, method="Wald")[3:4,]## 2.5 % 97.5 %
## (Intercept) 0.17817515 0.5464922
## conditionC 0.08160272 0.8247797Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
##all data
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp1"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp1")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3191.2 3225.9 -1589.6 3179.2 2406
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.8161 -1.0141 0.3721 0.9345 1.3206
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 4.530e-01 0.673034
## stimulus (Intercept) 5.008e-06 0.002238
## conditionC 3.157e-02 0.177674 1.00
## Number of obs: 2412, groups: participantCode, 67; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.36236 0.09397 3.856 0.000115 ***
## conditionC 0.45326 0.18963 2.390 0.016836 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.071
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.17818361 0.5465339
## conditionC 0.08159645 0.8249155Familiar X Trials
Condition Effect
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
# excluding the random effect with an estimated covariance of zero (identical results)
m <- glmer(isRight~conditionC+(1|participantCode)+(0+conditionC|stimulus),data=subset(d,exp=="exp1"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (0 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp1" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1621.4 1641.8 -806.7 1613.4 1202
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.8397 -1.0270 0.5291 0.8899 1.2857
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.30588 0.5531
## stimulus conditionC 0.03155 0.1776
## Number of obs: 1206, groups: participantCode, 67; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.32844 0.09148 3.590 0.00033 ***
## conditionC 0.42159 0.18715 2.253 0.02428 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.069confint(m, method="Wald")[3:4,]## 2.5 % 97.5 %
## (Intercept) 0.14914321 0.5077313
## conditionC 0.05478651 0.7884028Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
#maximal model
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp1"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa")) #convergence warning## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp1" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1625.4 1656.0 -806.7 1613.4 1200
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.8397 -1.0270 0.5291 0.8899 1.2857
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.30588 0.5531
## stimulus (Intercept) 0.00000 0.0000
## conditionC 0.03155 0.1776 NaN
## Number of obs: 1206, groups: participantCode, 67; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.32844 0.09148 3.590 0.00033 ***
## conditionC 0.42159 0.18715 2.253 0.02428 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.069
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.14914303 0.5077312
## conditionC 0.05478489 0.7884044Testing against chance: Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp1"&testType=="familiarX"&conditionC==0.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data:
## subset(d, exp == "exp1" & testType == "familiarX" & conditionC ==
## 0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 740.7 749.5 -368.4 736.7 574
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1056 -1.0740 0.3629 0.8553 1.1999
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.6889 0.83
## Number of obs: 576, groups: participantCode, 32
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.5848 0.1761 3.321 0.000897 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.2396826 0.9299226Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp1"&testType=="familiarX"&conditionC==0.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data:
## subset(d, exp == "exp1" & testType == "familiarX" & conditionC ==
## 0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 742.7 755.8 -368.4 736.7 573
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1056 -1.0740 0.3629 0.8553 1.1999
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.6889 0.83
## stimulus (Intercept) 0.0000 0.00
## Number of obs: 576, groups: participantCode, 32; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.5848 0.1761 3.321 0.000897 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.239681 0.9299249Testing against chance: Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp1"&testType=="familiarX"&conditionC==-0.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data:
## subset(d, exp == "exp1" & testType == "familiarX" & conditionC ==
## -0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 873.8 882.7 -434.9 869.8 628
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.2102 -1.0412 0.7315 0.9604 1.0828
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.08204 0.2864
## Number of obs: 630, groups: participantCode, 35
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.11052 0.09406 1.175 0.24confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) -0.07383629 0.2948815Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp1"&testType=="familiarX"&conditionC==-0.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data:
## subset(d, exp == "exp1" & testType == "familiarX" & conditionC ==
## -0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 875.8 889.1 -434.9 869.8 627
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.2102 -1.0412 0.7315 0.9604 1.0828
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.08204 0.2864
## stimulus (Intercept) 0.00000 0.0000
## Number of obs: 630, groups: participantCode, 35; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.11052 0.09406 1.175 0.24
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) -0.07383649 0.2948815Novel X Trials
Condition Effect
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~conditionC+(1|participantCode)+(0+conditionC|stimulus),data=subset(d,exp=="exp1"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (0 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp1" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1618.4 1638.8 -805.2 1610.4 1202
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.0476 -1.0466 0.4978 0.8995 1.2835
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.33311 0.5772
## stimulus conditionC 0.02415 0.1554
## Number of obs: 1206, groups: participantCode, 67; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.34347 0.09399 3.654 0.000258 ***
## conditionC 0.41292 0.19094 2.163 0.030574 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.076confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.15925784 0.5276890
## conditionC 0.03868726 0.7871542Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp1"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp1" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1622.4 1652.9 -805.2 1610.4 1200
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.0476 -1.0466 0.4978 0.8995 1.2835
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.33311 0.5772
## stimulus (Intercept) 0.00000 0.0000
## conditionC 0.02415 0.1554 NaN
## Number of obs: 1206, groups: participantCode, 67; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.34347 0.09399 3.654 0.000258 ***
## conditionC 0.41292 0.19094 2.163 0.030574 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.076
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.1592572 0.527690
## conditionC 0.0386855 0.787156Testing against chance: Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp1"&testType=="novelX"&conditionC==0.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp1" & testType == "novelX" & conditionC ==
## 0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 727.2 740.2 -360.6 721.2 573
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.6207 -0.9700 0.3824 0.8683 1.3699
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.98658 0.99327
## stimulus (Intercept) 0.00439 0.06626
## Number of obs: 576, groups: participantCode, 32; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.6291 0.2038 3.086 0.00203 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.2295743 1.028565Testing against chance: Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glm(isRight~1,data=subset(d,exp=="exp1"&testType=="novelX"&conditionC==-0.5),family=binomial)
summary(m)##
## Call:
## glm(formula = isRight ~ 1, family = binomial, data = subset(d,
## exp == "exp1" & testType == "novelX" & conditionC == -0.5))
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.232 -1.232 1.124 1.124 1.124
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.12716 0.07984 1.593 0.111
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 870.82 on 629 degrees of freedom
## Residual deviance: 870.82 on 629 degrees of freedom
## AIC: 872.82
##
## Number of Fisher Scoring iterations: 3confint(m, method="Wald")## Waiting for profiling to be done...## 2.5 % 97.5 %
## -0.02915486 0.28398462Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp1"&testType=="novelX"&conditionC==-0.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp1" & testType == "novelX" & conditionC ==
## -0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 876.8 890.2 -435.4 870.8 627
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.0656 -1.0656 0.9384 0.9384 0.9384
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0 0
## stimulus (Intercept) 0 0
## Number of obs: 630, groups: participantCode, 35; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.12716 0.07984 1.593 0.111
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[3,]## 2.5 % 97.5 %
## -0.02933406 0.28364441Interaction between test type and condition
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#recode test type
d$testTypeC <- ifelse(d$testType=="novelX",0.5,
ifelse(d$testType=="familiarX",-0.5,NA))
##all data
m <- glmer(isRight~conditionC*testTypeC+(1|participantCode)+(0+conditionC|stimulus),data=subset(d,exp=="exp1"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC * testTypeC + (1 | participantCode) + (0 +
## conditionC | stimulus)
## Data: subset(d, exp == "exp1")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3191.2 3225.9 -1589.6 3179.2 2406
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.8151 -1.0158 0.3718 0.9372 1.3230
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.45292 0.6730
## stimulus conditionC 0.03123 0.1767
## Number of obs: 2412, groups: participantCode, 67; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.36233 0.09396 3.856 0.000115 ***
## conditionC 0.45319 0.18958 2.390 0.016827 *
## testTypeC 0.01004 0.08636 0.116 0.907448
## conditionC:testTypeC -0.02008 0.18248 -0.110 0.912371
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC tstTyC
## conditionC 0.070
## testTypeC 0.000 0.000
## cndtnC:tsTC 0.000 0.000 0.098confint(m, method="Wald")[3:6,]## 2.5 % 97.5 %
## (Intercept) 0.17817542 0.5464902
## conditionC 0.08161444 0.8247583
## testTypeC -0.15922356 0.1793038
## conditionC:testTypeC -0.37774439 0.3375804Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
##all data
m <- glmer(isRight~conditionC*testTypeC+(1+testTypeC|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp1"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionC * testTypeC + (1 + testTypeC | participantCode) +
## (1 + conditionC | stimulus)
## Data: subset(d, exp == "exp1")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3199.2 3257.0 -1589.6 3179.2 2402
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.8439 -1.0144 0.3759 0.9368 1.3191
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.4529594 0.67302
## testTypeC 0.0003477 0.01865 1.00
## stimulus (Intercept) 0.0000000 0.00000
## conditionC 0.0312275 0.17671 NaN
## Number of obs: 2412, groups: participantCode, 67; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.36238 0.09396 3.857 0.000115 ***
## conditionC 0.45328 0.18959 2.391 0.016810 *
## testTypeC 0.01336 0.08894 0.150 0.880571
## conditionC:testTypeC -0.01559 0.18477 -0.084 0.932777
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC tstTyC
## conditionC 0.070
## testTypeC 0.025 0.003
## cndtnC:tsTC 0.002 0.024 0.130
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[7:10,]## 2.5 % 97.5 %
## (Intercept) 0.17821292 0.5465478
## conditionC 0.08168851 0.8248784
## testTypeC -0.16095354 0.1876788
## conditionC:testTypeC -0.37772619 0.3465552Signal Detection Model
Sensitivity
Overall
#recode condition
subj_overall$conditionC <- ifelse(subj_overall$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_overall$condition=="Non-Learnable Pre-Exposure",-0.5,NA))
##all data
m <- lm(dprime~conditionC,data=subset(subj_overall,exp=="exp1"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_overall,
## exp == "exp1"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.5828 -0.6476 -0.2026 0.2241 3.6264
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.5103 0.1306 3.908 0.000224 ***
## conditionC 0.6155 0.2611 2.357 0.021460 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.068 on 65 degrees of freedom
## Multiple R-squared: 0.07872, Adjusted R-squared: 0.06455
## F-statistic: 5.554 on 1 and 65 DF, p-value: 0.02146Testing against chance: Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp1"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp1" &
## conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.5828 -0.9621 -0.6038 0.6306 3.0109
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.8181 0.2384 3.431 0.00172 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.349 on 31 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp1"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp1" &
## conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.1070 -0.2739 -0.2026 0.1443 3.6264
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.2026 0.1220 1.66 0.106
##
## Residual standard error: 0.7219 on 34 degrees of freedomFamiliar X Trials
Condition Effect
#recode condition
subj_testType$conditionC <- ifelse(subj_testType$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_testType$condition=="Non-Learnable Pre-Exposure",-0.5,NA))
m <- lm(dprime~conditionC,data=subset(subj_testType,exp=="exp1"&testType=="familiarX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## exp == "exp1" & testType == "familiarX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.6555 -0.7511 -0.1609 0.4542 3.0256
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.4560 0.1251 3.646 0.000531 ***
## conditionC 0.5903 0.2501 2.360 0.021305 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.023 on 65 degrees of freedom
## Multiple R-squared: 0.0789, Adjusted R-squared: 0.06473
## F-statistic: 5.568 on 1 and 65 DF, p-value: 0.02131Testing against chance: Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp1"&testType=="familiarX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp1" & testType == "familiarX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.6555 -0.7511 -0.4171 0.4855 2.4353
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.7511 0.2111 3.558 0.00123 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.194 on 31 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp1"&testType=="familiarX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp1" & testType == "familiarX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.5212 -0.4461 -0.1609 0.3379 3.0256
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.1609 0.1413 1.138 0.263
##
## Residual standard error: 0.8361 on 34 degrees of freedomNovel X Trials
Condition Effect
m <- lm(dprime~conditionC,data=subset(subj_testType,exp=="exp1"&testType=="novelX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## exp == "exp1" & testType == "novelX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.9471 -0.7517 -0.1901 0.3231 2.9963
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.4709 0.1255 3.753 0.000375 ***
## conditionC 0.5616 0.2509 2.238 0.028663 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.026 on 65 degrees of freedom
## Multiple R-squared: 0.07154, Adjusted R-squared: 0.05725
## F-statistic: 5.008 on 1 and 65 DF, p-value: 0.02866Testing against chance: Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp1"&testType=="novelX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp1" & testType == "novelX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.9471 -1.0311 -0.7517 0.9832 2.4347
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.7517 0.2303 3.263 0.00268 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.303 on 31 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp1"&testType=="novelX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp1" & testType == "novelX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.38555 -0.32982 0.08931 0.16317 2.99633
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.1901 0.1152 1.65 0.108
##
## Residual standard error: 0.6816 on 34 degrees of freedomInteraction between Novel X and Familiar X trials
#recode test type
subj_testType$testTypeC <- ifelse(subj_testType$testType=="novelX",0.5,
ifelse(subj_testType$testType=="familiarX",-0.5,NA))
m <- lm(dprime~conditionC*testTypeC,data=subset(subj_testType,exp=="exp1"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC * testTypeC, data = subset(subj_testType,
## exp == "exp1"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.9471 -0.7511 -0.1609 0.3803 3.0256
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.46345 0.08858 5.232 6.54e-07 ***
## conditionC 0.57593 0.17716 3.251 0.00147 **
## testTypeC 0.01492 0.17716 0.084 0.93304
## conditionC:testTypeC -0.02867 0.35433 -0.081 0.93563
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.024 on 130 degrees of freedom
## Multiple R-squared: 0.07527, Adjusted R-squared: 0.05393
## F-statistic: 3.527 on 3 and 130 DF, p-value: 0.01685Response Bias
Overall
#recode condition
subj_overall$conditionC <- ifelse(subj_overall$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_overall$condition=="Non-Learnable Pre-Exposure",-0.5,NA))
##all data
m <- lm(c~conditionC,data=subset(subj_overall,exp=="exp1"))
summary(m)##
## Call:
## lm(formula = c ~ conditionC, data = subset(subj_overall, exp ==
## "exp1"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.61163 -0.18436 0.01147 0.30288 1.26614
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.23750 0.05322 -4.463 3.29e-05 ***
## conditionC 0.13077 0.10643 1.229 0.224
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4352 on 65 degrees of freedom
## Multiple R-squared: 0.0227, Adjusted R-squared: 0.007662
## F-statistic: 1.51 on 1 and 65 DF, p-value: 0.2236Testing against chance: Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_overall,exp=="exp1"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp1" &
## conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.52696 -0.19691 -0.04105 0.18958 1.26614
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.17211 0.06104 -2.82 0.0083 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3453 on 31 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_overall,exp=="exp1"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp1" &
## conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.61163 -0.09478 0.02066 0.30288 0.73361
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.30288 0.08507 -3.56 0.00112 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5033 on 34 degrees of freedomFamiliar X Trials
Condition Effect
#recode condition
subj_testType$conditionC <- ifelse(subj_testType$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_testType$condition=="Non-Learnable Pre-Exposure",-0.5,NA))
m <- lm(c~conditionC,data=subset(subj_testType,exp=="exp1"&testType=="familiarX"))
summary(m)##
## Call:
## lm(formula = c ~ conditionC, data = subset(subj_testType, exp ==
## "exp1" & testType == "familiarX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.15872 -0.29250 0.04125 0.24821 1.65514
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.36009 0.05852 -6.154 5.26e-08 ***
## conditionC 0.22377 0.11703 1.912 0.0603 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4785 on 65 degrees of freedom
## Multiple R-squared: 0.05325, Adjusted R-squared: 0.03868
## F-statistic: 3.656 on 1 and 65 DF, p-value: 0.06028Testing against chance: Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp1"&testType=="familiarX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp1" &
## testType == "familiarX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.15872 -0.30657 -0.03701 0.24821 1.65514
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.24821 0.08825 -2.813 0.00845 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4992 on 31 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp1"&testType=="familiarX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp1" &
## testType == "familiarX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.12124 -0.20924 0.04125 0.25951 0.90271
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.47198 0.07755 -6.086 6.66e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4588 on 34 degrees of freedomNovel X Trials
Condition Effect
m <- lm(c~conditionC,data=subset(subj_testType,exp=="exp1"&testType=="novelX"))
summary(m)##
## Call:
## lm(formula = c ~ conditionC, data = subset(subj_testType, exp ==
## "exp1" & testType == "novelX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.48307 -0.19598 -0.03536 0.18000 1.70337
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.09969 0.06129 -1.627 0.109
## conditionC 0.02091 0.12258 0.171 0.865
##
## Residual standard error: 0.5012 on 65 degrees of freedom
## Multiple R-squared: 0.0004474, Adjusted R-squared: -0.01493
## F-statistic: 0.02909 on 1 and 65 DF, p-value: 0.8651Testing against chance: Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp1"&testType=="novelX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp1" &
## testType == "novelX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.67547 -0.19598 -0.05047 0.08924 0.85395
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.08924 0.05959 -1.498 0.144
##
## Residual standard error: 0.3371 on 31 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp1"&testType=="novelX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp1" &
## testType == "novelX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.4831 -0.1751 0.1101 0.2528 1.7034
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.1101 0.1037 -1.062 0.296
##
## Residual standard error: 0.6136 on 34 degrees of freedomInteraction between Novel X and Familiar X trials
m <- lm(c~conditionC*testTypeC,data=subset(subj_testType,exp=="exp1"))
summary(m)##
## Call:
## lm(formula = c ~ conditionC * testTypeC, data = subset(subj_testType,
## exp == "exp1"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.48307 -0.19598 0.04125 0.24821 1.70337
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.22989 0.04237 -5.426 2.72e-07 ***
## conditionC 0.12234 0.08474 1.444 0.15121
## testTypeC 0.26040 0.08474 3.073 0.00258 **
## conditionC:testTypeC -0.20286 0.16947 -1.197 0.23348
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.49 on 130 degrees of freedom
## Multiple R-squared: 0.09289, Adjusted R-squared: 0.07196
## F-statistic: 4.438 on 3 and 130 DF, p-value: 0.005277Experiment 2
Test Accuracy
Plot
#Familiar X Test
p1 <- ggplot(subset(test_type_exp,testType=="familiarX"&exp=="exp2"),aes(condition,accuracy,fill=condition, color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="familiarX"&exp=="exp2"),aes(y=acc), width = 0.05, height=0.02, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, size=0.8)+
theme_classic(base_size=12)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Learnable Pre-Exposure"),
labels=c("Non-Learnable\nPre-Exposure","No\nPre-Exposure","Learnable\nPre-Exposure"),
limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Learnable Pre-Exposure"))+
scale_fill_manual(values=c("#E41A1C","#377EB8","#4DAF4A"),limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Learnable Pre-Exposure"))+
scale_color_manual(values=c("#E41A1C","#377EB8","#4DAF4A"),limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Learnable Pre-Exposure"))+
theme(legend.position="none")+
ylab("Accuracy - Familiar X Test")+
xlab("Condition")
#Novel X
p2 <- ggplot(subset(test_type_exp,testType=="novelX"&exp=="exp2"),aes(condition,accuracy,fill=condition, color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="novelX"&exp=="exp2"),aes(y=acc), width = 0.05, height=0.02, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, size=0.8)+
theme_classic(base_size=12)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Learnable Pre-Exposure"),
labels=c("Non-Learnable\nPre-Exposure","No\nPre-Exposure","Learnable\nPre-Exposure"),
limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Learnable Pre-Exposure"))+
scale_fill_manual(values=c("#E41A1C","#377EB8","#4DAF4A"),limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Learnable Pre-Exposure"))+
scale_color_manual(values=c("#E41A1C","#377EB8","#4DAF4A"),limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Learnable Pre-Exposure"))+
theme(legend.position="none")+
ylab("Accuracy - Novel X Test")+
xlab("Condition")
plot_grid(p1,p2, labels=c("A","B"),label_size=18)
Overall Accuracy Table
overall_exp %>%
filter(exp=="exp2") %>%
select(-c(se,accuracy_ci,d_prime_ci,c_bias_ci)) %>%
formattable()| exp | condition | N | accuracy | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|---|
| exp2 | Learnable Pre-Exposure | 83 | 0.5722301 | 0.5361125 | 0.6083476 | 0.5077750 | 0.25212450 | 0.7634256 | -0.1953715 | -0.2714524 | -0.1192905 |
| exp2 | No Pre-Exposure | 81 | 0.5445816 | 0.5165071 | 0.5726561 | 0.2917437 | 0.09294647 | 0.4905409 | -0.2817981 | -0.3880251 | -0.1755711 |
| exp2 | Non-Learnable Pre-Exposure | 79 | 0.5256681 | 0.5047038 | 0.5466323 | 0.1619213 | 0.03136162 | 0.2924810 | -0.3036137 | -0.3863990 | -0.2208284 |
Accuracy Table By Test Type
test_type_exp %>%
filter(exp=="exp2") %>%
select(-c(se,accuracy_ci,d_prime_ci,c_bias_ci)) %>%
formattable()| exp | condition | testType | N | accuracy | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| exp2 | Learnable Pre-Exposure | familiarX | 83 | 0.5789826 | 0.5386545 | 0.6193107 | 0.4870309 | 0.236126039 | 0.7379357 | -0.408382341 | -0.49734439 | -0.3194203 |
| exp2 | Learnable Pre-Exposure | novelX | 83 | 0.5653447 | 0.5256768 | 0.6050126 | 0.4027506 | 0.159998868 | 0.6455023 | 0.003715546 | -0.10535999 | 0.1127911 |
| exp2 | No Pre-Exposure | familiarX | 81 | 0.5480110 | 0.5164667 | 0.5795553 | 0.2995076 | 0.101032689 | 0.4979825 | -0.645538873 | -0.75491277 | -0.5361650 |
| exp2 | No Pre-Exposure | novelX | 81 | 0.5411523 | 0.5100279 | 0.5722766 | 0.2466637 | 0.058929824 | 0.4343976 | 0.086566193 | -0.07935325 | 0.2524856 |
| exp2 | Non-Learnable Pre-Exposure | familiarX | 79 | 0.5246132 | 0.4991932 | 0.5500332 | 0.1377382 | -0.017625810 | 0.2931023 | -0.573193547 | -0.67709443 | -0.4692927 |
| exp2 | Non-Learnable Pre-Exposure | novelX | 79 | 0.5267229 | 0.4975832 | 0.5558626 | 0.1657024 | -0.001404172 | 0.3328089 | -0.048268953 | -0.16192481 | 0.0653869 |
Correlation
Correlation between Familiar X and Novel X Trials: Learnable Pre-Exposure
##Correlations between Familiar X and Novel X
c <- corr.test(subset(subj_accuracy_wide, condition=="Learnable Pre-Exposure" & exp=="exp2")[,c("novelX","familiarX")])
c## Call:corr.test(x = subset(subj_accuracy_wide, condition == "Learnable Pre-Exposure" &
## exp == "exp2")[, c("novelX", "familiarX")])
## Correlation matrix
## novelX familiarX
## novelX 1.00 0.63
## familiarX 0.63 1.00
## Sample Size
## [1] 83
## Probability values (Entries above the diagonal are adjusted for multiple tests.)
## novelX familiarX
## novelX 0 0
## familiarX 0 0
##
## To see confidence intervals of the correlations, print with the short=FALSE optionc$p## novelX familiarX
## novelX 0.000000e+00 2.081808e-10
## familiarX 2.081808e-10 0.000000e+00Correlation between Familiar X and Novel X Trials: No Pre-Exposure
c <- corr.test(subset(subj_accuracy_wide, condition=="No Pre-Exposure" & exp=="exp2")[,c("novelX","familiarX")])
c## Call:corr.test(x = subset(subj_accuracy_wide, condition == "No Pre-Exposure" &
## exp == "exp2")[, c("novelX", "familiarX")])
## Correlation matrix
## novelX familiarX
## novelX 1.00 0.61
## familiarX 0.61 1.00
## Sample Size
## [1] 81
## Probability values (Entries above the diagonal are adjusted for multiple tests.)
## novelX familiarX
## novelX 0 0
## familiarX 0 0
##
## To see confidence intervals of the correlations, print with the short=FALSE optionc$p## novelX familiarX
## novelX 0.00000e+00 2.12128e-09
## familiarX 2.12128e-09 0.00000e+00Correlation between Familiar X and Novel X Trials: Non-Learnable Pre-Exposure
c <- corr.test(subset(subj_accuracy_wide, condition=="Non-Learnable Pre-Exposure" & exp=="exp2")[,c("novelX","familiarX")])
c## Call:corr.test(x = subset(subj_accuracy_wide, condition == "Non-Learnable Pre-Exposure" &
## exp == "exp2")[, c("novelX", "familiarX")])
## Correlation matrix
## novelX familiarX
## novelX 1.00 0.18
## familiarX 0.18 1.00
## Sample Size
## [1] 79
## Probability values (Entries above the diagonal are adjusted for multiple tests.)
## novelX familiarX
## novelX 0.00 0.12
## familiarX 0.12 0.00
##
## To see confidence intervals of the correlations, print with the short=FALSE optionc$p## novelX familiarX
## novelX 0.0000000 0.1179566
## familiarX 0.1179566 0.0000000Interaction
#interaction
subj_accuracy_wide$conditionC <- ifelse(subj_accuracy_wide$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_accuracy_wide$condition=="Non-Learnable Pre-Exposure",-0.5,
ifelse(subj_accuracy_wide$condition=="No Pre-Exposure",0,NA)))
m <- lm(novelX~familiarX*conditionC, subset(subj_accuracy_wide,exp=="exp2"))
summary(m)##
## Call:
## lm(formula = novelX ~ familiarX * conditionC, data = subset(subj_accuracy_wide,
## exp == "exp2"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.45402 -0.07543 0.00254 0.07862 0.29724
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.27859 0.03362 8.287 8.36e-15 ***
## familiarX 0.47707 0.06007 7.941 7.81e-14 ***
## conditionC -0.18070 0.08022 -2.253 0.0252 *
## familiarX:conditionC 0.35041 0.14285 2.453 0.0149 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.1287 on 239 degrees of freedom
## Multiple R-squared: 0.3023, Adjusted R-squared: 0.2936
## F-statistic: 34.52 on 3 and 239 DF, p-value: < 2.2e-16Correlation Plot
ggplot(subset(subj_accuracy_wide,exp=="exp2"),aes(familiarX,novelX, color=condition,linetype=condition,shape=condition))+
geom_jitter(width=0.02,height=0.02,size=3)+
geom_smooth(method=lm,se =F,size=1.5)+
scale_color_manual(name="Condition",values=c("#4DAF4A","#377EB8","#E41A1C"))+
scale_linetype_manual(name="Condition",values=c(1,4,2))+
scale_shape_manual(name="Condition",values=c(16,15,17))+
ylab("Accuracy - Novel X")+
xlab("Accuracy - Familiar X")+
theme_classic(base_size=18)+
theme(legend.position=c(0.3,0.9))
Logistic Mixed-Effects Model
Overall
Final Converging Models
The final models with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
Richter single-contrast approach.
#recode condition
d$conditionC <- ifelse(d$condition=="Learnable Pre-Exposure",0.5,
ifelse(d$condition=="Non-Learnable Pre-Exposure",-0.5,
ifelse(d$condition=="No Pre-Exposure",0,NA)))
d$conditionOrthContrast2 <- ifelse(d$condition=="Learnable Pre-Exposure",-1/3,
ifelse(d$condition=="Non-Learnable Pre-Exposure",-1/3,
ifelse(d$condition=="No Pre-Exposure",2/3,NA)))
#overall analysis (Richter single-contrast apporach)
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,exp=="exp2"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data: subset(d, exp == "exp2")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11872.8 11894.0 -5933.4 11866.8 8743
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1798 -1.0343 0.4833 0.9310 1.3423
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2357 0.4855
## Number of obs: 8746, groups: participantCode, 243
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.2104 0.0383 5.493 3.94e-08 ***
## conditionC 0.2165 0.0937 2.310 0.0209 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.008confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.13533554 0.2854744
## conditionC 0.03282819 0.4001370Abelson & Prentice approach.
##check same analysis with Abelson & Prentice approach
m <- glmer(isRight~conditionC+conditionOrthContrast2+(1|participantCode),data=subset(d,exp=="exp2"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionC + conditionOrthContrast2 + (1 | participantCode)
## Data: subset(d, exp == "exp2")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11874.7 11903.0 -5933.3 11866.7 8742
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1830 -1.0320 0.4826 0.9331 1.3406
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2356 0.4854
## Number of obs: 8746, groups: participantCode, 243
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21042 0.03830 5.494 3.92e-08 ***
## conditionC 0.21628 0.09372 2.308 0.021 *
## conditionOrthContrast2 -0.02106 0.08098 -0.260 0.795
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC
## conditionC -0.008
## cndtnOrthC2 -0.002 0.008confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.13535407 0.2854761
## conditionC 0.03259514 0.3999630
## conditionOrthContrast2 -0.17978510 0.1376648Maximal Random Effects Model
Note that these model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
Richter single-contrast approach.
#overall analysis (Richter single-contrast apporach)
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp2"),family=binomial,glmerControl(optimizer="bobyqa")) #convergence warning## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp2")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11878.8 11921.2 -5933.4 11866.8 8740
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1798 -1.0343 0.4833 0.9310 1.3423
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 2.357e-01 4.855e-01
## stimulus (Intercept) 7.043e-14 2.654e-07
## conditionC 5.117e-12 2.262e-06 1.00
## Number of obs: 8746, groups: participantCode, 243; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21041 0.03830 5.493 3.95e-08 ***
## conditionC 0.21648 0.09372 2.310 0.0209 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.008
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.1353330 0.2854778
## conditionC 0.0327907 0.4001751Abelson & Prentice approach.
##check same analysis with Abelson & Prentice approach
m=glmer(isRight~conditionC+conditionOrthContrast2+(1|participantCode)+(1+conditionC+conditionOrthContrast2|stimulus),data=subset(d,exp=="exp2"),family=binomial,glmerControl(optimizer="bobyqa")) #convergence warning## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionC + conditionOrthContrast2 + (1 | participantCode) +
## (1 + conditionC + conditionOrthContrast2 | stimulus)
## Data: subset(d, exp == "exp2")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11886.5 11957.3 -5933.3 11866.5 8736
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1902 -1.0241 0.4887 0.9259 1.3662
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.236011 0.48581
## stimulus (Intercept) 0.000000 0.00000
## conditionC 0.006386 0.07991 NaN
## conditionOrthContrast2 0.003532 0.05943 NaN -1.00
## Number of obs: 8746, groups: participantCode, 243; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21054 0.03832 5.494 3.93e-08 ***
## conditionC 0.21654 0.09472 2.286 0.0222 *
## conditionOrthContrast2 -0.02112 0.08164 -0.259 0.7959
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC
## conditionC -0.008
## cndtnOrthC2 -0.002 -0.009
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[8:10,]## 2.5 % 97.5 %
## (Intercept) 0.13543373 0.2856546
## conditionC 0.03090042 0.4021836
## conditionOrthContrast2 -0.18113716 0.1388980Pairwise condition comparisons
Learnable vs. Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,exp=="exp2"&conditionC!=0),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data: subset(d, exp == "exp2" & conditionC != 0)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 7895.9 7915.9 -3945.0 7889.9 5827
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.2363 -1.0167 0.4943 0.9153 1.3574
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2563 0.5063
## Number of obs: 5830, groups: participantCode, 162
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21900 0.04832 4.533 5.82e-06 ***
## conditionC 0.21853 0.09651 2.264 0.0236 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.010confint(m, method="Wald")[2:3,]## 2.5 % 97.5 %
## (Intercept) 0.12430384 0.3136995
## conditionC 0.02937304 0.4076837Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
#Learnable vs. Non-Learnable
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp2"&conditionC!=0),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp2" & conditionC != 0)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 7901.8 7941.8 -3944.9 7889.8 5824
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.2458 -1.0167 0.4940 0.9184 1.3877
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.256944 0.50690
## stimulus (Intercept) 0.001265 0.03556
## conditionC 0.005020 0.07086 1.00
## Number of obs: 5830, groups: participantCode, 162; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21921 0.04872 4.499 6.83e-06 ***
## conditionC 0.21892 0.09732 2.250 0.0245 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.005
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.12371158 0.3147080
## conditionC 0.02818059 0.4096622No Pre-Exposure vs. Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#Non-Learnable vs. No Pre-Exposure
d$conditionNonLearnableNoPreExp <- ifelse(d$conditionC==0.5,1.5,
ifelse(d$conditionC==0,0.5,d$conditionC))
m <- glmer(isRight~conditionNonLearnableNoPreExp+(1|participantCode),data=subset(d,exp=="exp2"&conditionNonLearnableNoPreExp!=1.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionNonLearnableNoPreExp + (1 | participantCode)
## Data: subset(d, exp == "exp2" & conditionNonLearnableNoPreExp != 1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 7925.0 7945.0 -3959.5 7919.0 5757
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6833 -1.0280 0.6637 0.9472 1.1175
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1019 0.3192
## Number of obs: 5760, groups: participantCode, 160
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.14609 0.03681 3.969 7.22e-05 ***
## conditionNonLearnableNoPreExp 0.08050 0.07357 1.094 0.274
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnNnLNPE -0.008confint(m, method="Wald")[2:3,]## 2.5 % 97.5 %
## (Intercept) 0.07394511 0.2182280
## conditionNonLearnableNoPreExp -0.06368933 0.2246826Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~conditionNonLearnableNoPreExp+(1|participantCode)+(1+conditionNonLearnableNoPreExp|stimulus),data=subset(d,exp=="exp2"&conditionNonLearnableNoPreExp!=1.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionNonLearnableNoPreExp + (1 | participantCode) +
## (1 + conditionNonLearnableNoPreExp | stimulus)
## Data: subset(d, exp == "exp2" & conditionNonLearnableNoPreExp != 1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 7931.0 7971.0 -3959.5 7919.0 5754
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6832 -1.0280 0.6637 0.9472 1.1175
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 1.019e-01 3.192e-01
## stimulus (Intercept) 0.000e+00 0.000e+00
## conditionNonLearnableNoPreExp 1.884e-13 4.340e-07 NaN
## Number of obs: 5760, groups: participantCode, 160; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.14609 0.03681 3.969 7.22e-05 ***
## conditionNonLearnableNoPreExp 0.08050 0.07357 1.094 0.274
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnNnLNPE -0.008
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.07394446 0.2182281
## conditionNonLearnableNoPreExp -0.06369568 0.2246870Learnable vs. No Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#Learnable vs. No Pre-Exposure
d$conditionLearnableNoPreExp <- ifelse(d$conditionC==-0.5,-1.5,
ifelse(d$conditionC==0,-0.5,d$conditionC))
m <- glmer(isRight~conditionLearnableNoPreExp+(1|participantCode)+(0+conditionLearnableNoPreExp|stimulus),data=subset(d,exp=="exp2"&conditionLearnableNoPreExp!=-1.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionLearnableNoPreExp + (1 | participantCode) +
## (0 + conditionLearnableNoPreExp | stimulus)
## Data: subset(d, exp == "exp2" & conditionLearnableNoPreExp != -1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 7912.0 7938.7 -3952.0 7904.0 5898
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5150 -1.0227 0.4116 0.9217 1.4431
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.380740 0.61704
## stimulus conditionLearnableNoPreExp 0.007125 0.08441
## Number of obs: 5902, groups: participantCode, 164; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.27389 0.05571 4.916 8.82e-07 ***
## conditionLearnableNoPreExp 0.13751 0.11189 1.229 0.219
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnLrnNPE -0.003confint(m, method="Wald")[3:4,]## 2.5 % 97.5 %
## (Intercept) 0.16470200 0.3830818
## conditionLearnableNoPreExp -0.08180022 0.3568143Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
#Learnable vs. No Pre-Exposure
m <- glmer(isRight~conditionLearnableNoPreExp+(1|participantCode)+(1+conditionLearnableNoPreExp|stimulus),data=subset(d,exp=="exp2"&conditionLearnableNoPreExp!=-1.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionLearnableNoPreExp + (1 | participantCode) +
## (1 + conditionLearnableNoPreExp | stimulus)
## Data: subset(d, exp == "exp2" & conditionLearnableNoPreExp != -1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 7915.6 7955.7 -3951.8 7903.6 5896
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5284 -1.0206 0.4115 0.9242 1.4803
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.3815973 0.61774
## stimulus (Intercept) 0.0008796 0.02966
## conditionLearnableNoPreExp 0.0151500 0.12309 1.00
## Number of obs: 5902, groups: participantCode, 164; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.27415 0.05598 4.897 9.72e-07 ***
## conditionLearnableNoPreExp 0.13800 0.11299 1.221 0.222
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnLrnNPE 0.014
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.16443145 0.3838737
## conditionLearnableNoPreExp -0.08345495 0.3594534Testing against chance: Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#Learnable
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp2"&conditionC==0.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp2" & conditionC == 0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3924.0 3942.1 -1959.0 3918.0 2983
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.8814 -1.0176 0.3731 0.9146 1.5554
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.600946 0.77521
## stimulus (Intercept) 0.007036 0.08388
## Number of obs: 2986, groups: participantCode, 83; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.36246 0.09558 3.792 0.000149 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.1751252 0.5497959Testing against chance: No Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#No Pre-Exposure
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp2"&conditionC==0),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data: subset(d, exp == "exp2" & conditionC == 0)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3980.2 3992.1 -1988.1 3976.2 2914
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.0079 -1.0357 0.4980 0.9318 1.1954
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1951 0.4417
## Number of obs: 2916, groups: participantCode, 81
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.19355 0.06224 3.11 0.00187 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.07156634 0.3155389Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
#No Pre-Exposure
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp2"&conditionC==0),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp2" & conditionC == 0)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3982.2 4000.1 -1988.1 3976.2 2913
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.0079 -1.0357 0.4980 0.9318 1.1954
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1951 0.4417
## stimulus (Intercept) 0.0000 0.0000
## Number of obs: 2916, groups: participantCode, 81; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.19355 0.06224 3.11 0.00187 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.07156761 0.3155373Testing against chance: Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#Non-Learnable
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp2"&conditionC==-0.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data: subset(d, exp == "exp2" & conditionC == -0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3937.0 3948.9 -1966.5 3933.0 2842
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.2512 -1.0420 0.8860 0.9488 1.0044
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.02864 0.1692
## Number of obs: 2844, groups: participantCode, 79
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.10356 0.04223 2.452 0.0142 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.02079655 0.1863324Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
#No Pre-Exposure
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp2"&conditionC==0),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp2" & conditionC == 0)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3982.2 4000.1 -1988.1 3976.2 2913
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.0079 -1.0357 0.4980 0.9318 1.1954
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1951 0.4417
## stimulus (Intercept) 0.0000 0.0000
## Number of obs: 2916, groups: participantCode, 81; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.19355 0.06224 3.11 0.00187 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.07156761 0.3155373Familiar X Trials
Condition Effect
Final Converging Model
The final model swith the maximal random effects structure that still allowed the models to converge.
Richter single-contrast approach.
#Familiar X test
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,exp=="exp2"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data: subset(d, exp == "exp2" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 5977.9 5997.0 -2985.9 5971.9 4371
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6370 -1.0474 0.6335 0.9101 1.2752
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1683 0.4102
## Number of obs: 4374, groups: participantCode, 243
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21374 0.04078 5.241 1.59e-07 ***
## conditionC 0.23421 0.09982 2.346 0.019 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.009confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.13381332 0.2936650
## conditionC 0.03856683 0.4298617Abelson & Prentice approach.
##check same analysis with Abelson & Prentice approach
m <- glmer(isRight~conditionC+conditionOrthContrast2+(1|participantCode),data=subset(d,exp=="exp2"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionC + conditionOrthContrast2 + (1 | participantCode)
## Data: subset(d, exp == "exp2" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 5979.8 6005.4 -2985.9 5971.8 4370
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6399 -1.0491 0.6358 0.9085 1.2729
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1682 0.4102
## Number of obs: 4374, groups: participantCode, 243
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21375 0.04078 5.242 1.59e-07 ***
## conditionC 0.23405 0.09983 2.345 0.019 *
## conditionOrthContrast2 -0.01739 0.08627 -0.202 0.840
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC
## conditionC -0.009
## cndtnOrthC2 -0.002 0.008Maximal Random Effects Model
Note that these models do not converge (singular fit), thus the parameter estimates should be interpreted with caution.
Richter single-contrast approach.
#Familiar X test
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp2"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp2" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 5983.9 6022.2 -2985.9 5971.9 4368
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6370 -1.0474 0.6335 0.9101 1.2752
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 1.683e-01 4.102e-01
## stimulus (Intercept) 0.000e+00 0.000e+00
## conditionC 9.816e-13 9.908e-07 NaN
## Number of obs: 4374, groups: participantCode, 243; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21374 0.04078 5.241 1.59e-07 ***
## conditionC 0.23421 0.09983 2.346 0.019 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.009
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## .sig03 NA NA
## .sig04 NA NA
## (Intercept) 0.13381304 0.2936653
## conditionC 0.03855837 0.4298705Abelson & Prentice approach.
##check same analysis with Abelson & Prentice approach
m <- glmer(isRight~conditionC+conditionOrthContrast2+(1|participantCode)+(1+conditionC+conditionOrthContrast2|stimulus),data=subset(d,exp=="exp2"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionC + conditionOrthContrast2 + (1 | participantCode) +
## (1 + conditionC + conditionOrthContrast2 | stimulus)
## Data: subset(d, exp == "exp2" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 5991.4 6055.2 -2985.7 5971.4 4364
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6593 -1.0338 0.6264 0.9200 1.2700
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.168977 0.41107
## stimulus (Intercept) 0.000000 0.00000
## conditionC 0.006392 0.07995 NaN
## conditionOrthContrast2 0.013478 0.11610 NaN -1.00
## Number of obs: 4374, groups: participantCode, 243; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21404 0.04083 5.242 1.59e-07 ***
## conditionC 0.23458 0.10171 2.306 0.0211 *
## conditionOrthContrast2 -0.01735 0.09062 -0.191 0.8481
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC
## conditionC -0.008
## cndtnOrthC2 -0.002 -0.049
## convergence code: 0
## boundary (singular) fit: see ?isSingularPairwise condition comparisons: Learnable vs. Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,exp=="exp2"&testType=="familiarX"&conditionC!=0),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data:
## subset(d, exp == "exp2" & testType == "familiarX" & conditionC != 0)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3976.3 3994.2 -1985.1 3970.3 2913
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6929 -1.0392 0.5907 0.9135 1.3185
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1952 0.4418
## Number of obs: 2916, groups: participantCode, 162
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.22128 0.05171 4.280 1.87e-05 ***
## conditionC 0.23636 0.10330 2.288 0.0221 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.010confint(m, method="Wald")[2:3,]## 2.5 % 97.5 %
## (Intercept) 0.11993445 0.3226191
## conditionC 0.03390341 0.4388136Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
#Learnable vs. Non-Learnable
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp2"&testType=="familiarX"&conditionC!=0),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data:
## subset(d, exp == "exp2" & testType == "familiarX" & conditionC != 0)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3982.1 4018.0 -1985.0 3970.1 2910
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.7313 -1.0305 0.6027 0.9196 1.3325
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.196003 0.44272
## stimulus (Intercept) 0.002680 0.05177
## conditionC 0.004735 0.06881 1.00
## Number of obs: 2916, groups: participantCode, 162; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.22160 0.05319 4.166 3.1e-05 ***
## conditionC 0.23697 0.10470 2.263 0.0236 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.026
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.11734423 0.3258622
## conditionC 0.03176701 0.4421646Pairwise condition comparisons: No Pre-Exposure vs. Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#Non-Learnable vs. No Pre-Exposure
d$conditionNonLearnableNoPreExp <- ifelse(d$conditionC==0.5,1.5,
ifelse(d$conditionC==0,0.5,d$conditionC))
m <- glmer(isRight~conditionNonLearnableNoPreExp+(1|participantCode),data=subset(d,exp=="exp2"&testType=="familiarX"&conditionNonLearnableNoPreExp!=1.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionNonLearnableNoPreExp + (1 | participantCode)
## Data:
## subset(d, exp == "exp2" & testType == "familiarX" & conditionNonLearnableNoPreExp !=
## 1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3978.9 3996.8 -1986.5 3972.9 2877
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.2603 -1.0620 0.8258 0.9242 1.0339
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.04497 0.2121
## Number of obs: 2880, groups: participantCode, 160
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.14732 0.04118 3.578 0.000347 ***
## conditionNonLearnableNoPreExp 0.09535 0.08232 1.158 0.246734
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnNnLNPE -0.009confint(m, method="Wald")[2:3,]## 2.5 % 97.5 %
## (Intercept) 0.06661052 0.2280203
## conditionNonLearnableNoPreExp -0.06599196 0.2566988Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~conditionNonLearnableNoPreExp+(1|participantCode)+(1+conditionNonLearnableNoPreExp|stimulus),data=subset(d,exp=="exp2"&testType=="familiarX"&conditionNonLearnableNoPreExp!=1.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionNonLearnableNoPreExp + (1 | participantCode) +
## (1 + conditionNonLearnableNoPreExp | stimulus)
## Data:
## subset(d, exp == "exp2" & testType == "familiarX" & conditionNonLearnableNoPreExp !=
## 1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3984.9 4020.7 -1986.5 3972.9 2874
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.2603 -1.0620 0.8258 0.9242 1.0339
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 4.497e-02 2.121e-01
## stimulus (Intercept) 0.000e+00 0.000e+00
## conditionNonLearnableNoPreExp 8.318e-14 2.884e-07 NaN
## Number of obs: 2880, groups: participantCode, 160; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.14732 0.04118 3.578 0.000347 ***
## conditionNonLearnableNoPreExp 0.09535 0.08232 1.158 0.246736
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnNnLNPE -0.009
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.06661060 0.2280204
## conditionNonLearnableNoPreExp -0.06599234 0.2566987Learnable vs. No Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#Learnable vs. No Pre-Exposure
d$conditionLearnableNoPreExp <- ifelse(d$conditionC==-0.5,-1.5,
ifelse(d$conditionC==0,-0.5,d$conditionC))
m <- glmer(isRight~conditionLearnableNoPreExp+(1|participantCode)+(0+conditionLearnableNoPreExp|stimulus),data=subset(d,exp=="exp2"&testType=="familiarX"&conditionLearnableNoPreExp!=-1.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionLearnableNoPreExp + (1 | participantCode) +
## (0 + conditionLearnableNoPreExp | stimulus)
## Data:
## subset(d, exp == "exp2" & testType == "familiarX" & conditionLearnableNoPreExp !=
## -1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3994.1 4018.1 -1993.0 3986.1 2948
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.8783 -1.0341 0.5433 0.8938 1.4495
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.29097 0.5394
## stimulus conditionLearnableNoPreExp 0.02787 0.1670
## Number of obs: 2952, groups: participantCode, 164; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.27937 0.05724 4.881 1.06e-06 ***
## conditionLearnableNoPreExp 0.14039 0.12069 1.163 0.245
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnLrnNPE -0.002confint(m, method="Wald")[3:4,]## 2.5 % 97.5 %
## (Intercept) 0.16719023 0.3915568
## conditionLearnableNoPreExp -0.09615829 0.3769318Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
#Learnable vs. No Pre-Exposure
m <- glmer(isRight~conditionLearnableNoPreExp+(1|participantCode)+(1+conditionLearnableNoPreExp|stimulus),data=subset(d,exp=="exp2"&testType=="familiarX"&conditionLearnableNoPreExp!=-1.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionLearnableNoPreExp + (1 | participantCode) +
## (1 + conditionLearnableNoPreExp | stimulus)
## Data:
## subset(d, exp == "exp2" & testType == "familiarX" & conditionLearnableNoPreExp !=
## -1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3998.1 4034.0 -1993.0 3986.1 2946
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.8783 -1.0341 0.5433 0.8938 1.4495
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.29097 0.5394
## stimulus (Intercept) 0.00000 0.0000
## conditionLearnableNoPreExp 0.02787 0.1670 NaN
## Number of obs: 2952, groups: participantCode, 164; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.27937 0.05723 4.881 1.05e-06 ***
## conditionLearnableNoPreExp 0.14039 0.12066 1.164 0.245
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnLrnNPE -0.002
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.16719848 0.3915488
## conditionLearnableNoPreExp -0.09609512 0.3768695Testing against chance: Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp2"&testType=="familiarX"&condition=="Learnable Pre-Exposure"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp2" & testType == "familiarX" & condition ==
## "Learnable Pre-Exposure")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1983.4 1999.3 -988.7 1977.4 1491
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1794 -1.0002 0.4687 0.8565 1.6969
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.50348 0.7096
## stimulus (Intercept) 0.01017 0.1008
## Number of obs: 1494, groups: participantCode, 83; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.36784 0.09913 3.711 0.000207 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")[3,]## 2.5 % 97.5 %
## 0.1735530 0.5621242Testing against chance: No Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp2"&testType=="familiarX"&condition=="No Pre-Exposure"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data: subset(d, exp == "exp2" & testType == "familiarX" & condition ==
## "No Pre-Exposure")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2004.7 2015.2 -1000.3 2000.7 1456
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.4595 -1.0674 0.6852 0.9015 1.0512
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1172 0.3424
## Number of obs: 1458, groups: participantCode, 81
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.19919 0.06564 3.035 0.00241 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")[2,]## 2.5 % 97.5 %
## 0.07053917 0.32783279Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp2"&testType=="familiarX"&condition=="No Pre-Exposure"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp2" & testType == "familiarX" & condition ==
## "No Pre-Exposure")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2006.7 2022.5 -1000.3 2000.7 1455
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.4595 -1.0674 0.6852 0.9015 1.0512
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1172 0.3424
## stimulus (Intercept) 0.0000 0.0000
## Number of obs: 1458, groups: participantCode, 81; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.19919 0.06564 3.035 0.00241 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[3,]## 2.5 % 97.5 %
## 0.07053982 0.32783222Testing against chance: Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glm(isRight~1,data=subset(d,exp=="exp2"&testType=="familiarX"&condition=="Non-Learnable Pre-Exposure"),family=binomial)
summary(m)##
## Call:
## glm(formula = isRight ~ 1, family = binomial, data = subset(d,
## exp == "exp2" & testType == "familiarX" & condition == "Non-Learnable Pre-Exposure"))
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.220 -1.220 1.136 1.136 1.136
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.09853 0.05310 1.856 0.0635 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1967.9 on 1421 degrees of freedom
## Residual deviance: 1967.9 on 1421 degrees of freedom
## AIC: 1969.9
##
## Number of Fisher Scoring iterations: 3confint(m, method="Wald")## Waiting for profiling to be done...## 2.5 % 97.5 %
## -0.005479146 0.202722259Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp2"&testType=="familiarX"&condition=="Non-Learnable Pre-Exposure"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp2" & testType == "familiarX" & condition ==
## "Non-Learnable Pre-Exposure")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1973.9 1989.6 -983.9 1967.9 1419
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.0505 -1.0505 0.9519 0.9519 0.9519
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0 0
## stimulus (Intercept) 0 0
## Number of obs: 1422, groups: participantCode, 79; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.09853 0.05310 1.856 0.0635 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[3,]## 2.5 % 97.5 %
## -0.005544124 0.202609200Novel X Trials
Condition Effect
Final Converging Model
The final model swith the maximal random effects structure that still allowed the models to converge.
Richter single-contrast approach.
#novel X test
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,exp=="exp2"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data: subset(d, exp == "exp2" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 5985.5 6004.7 -2989.8 5979.5 4369
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6300 -1.0533 0.6135 0.9244 1.2478
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1799 0.4242
## Number of obs: 4372, groups: participantCode, 243
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.18786 0.04135 4.544 5.53e-06 ***
## conditionC 0.16734 0.10124 1.653 0.0984 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.012confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.10682393 0.2689007
## conditionC -0.03109545 0.3657677Abelson & Prentice approach.
##check same analysis with Abelson & Prentice approach
m <- glmer(isRight~conditionC+conditionOrthContrast2+(1|participantCode),data=subset(d,exp=="exp2"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionC + conditionOrthContrast2 + (1 | participantCode)
## Data: subset(d, exp == "exp2" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 5987.5 6013.0 -2989.7 5979.5 4368
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6334 -1.0492 0.6122 0.9226 1.2451
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1798 0.4241
## Number of obs: 4372, groups: participantCode, 243
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.18788 0.04134 4.544 5.51e-06 ***
## conditionC 0.16709 0.10125 1.650 0.0989 .
## conditionOrthContrast2 -0.02148 0.08750 -0.245 0.8061
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC
## conditionC -0.012
## cndtnOrthC2 -0.003 0.009Maximal Random Effects Model
Note that these models do not converge (singular fit), thus the parameter estimates should be interpreted with caution.
Richter single-contrast approach.
#novel X test
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp2"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp2" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 5991.5 6029.8 -2989.8 5979.5 4366
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6300 -1.0533 0.6135 0.9244 1.2478
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 1.799e-01 4.242e-01
## stimulus (Intercept) 7.604e-48 2.758e-24
## conditionC 4.796e-11 6.926e-06 1.00
## Number of obs: 4372, groups: participantCode, 243; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.18786 0.04135 4.544 5.53e-06 ***
## conditionC 0.16734 0.10124 1.653 0.0984 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.012
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## .sig03 NA NA
## .sig04 NA NA
## (Intercept) 0.10682401 0.2689005
## conditionC -0.03110048 0.3657719Abelson & Prentice approach.
##check same analysis with Abelson & Prentice approach
m <- glmer(isRight~conditionC+conditionOrthContrast2+(1|participantCode)+(1+conditionC+conditionOrthContrast2|stimulus),data=subset(d,exp=="exp2"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionC + conditionOrthContrast2 + (1 | participantCode) +
## (1 + conditionC + conditionOrthContrast2 | stimulus)
## Data: subset(d, exp == "exp2" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 5999.4 6063.3 -2989.7 5979.4 4362
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6110 -1.0469 0.6151 0.9186 1.2489
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 1.800e-01 0.42431
## stimulus (Intercept) 3.249e-05 0.00570
## conditionC 4.784e-03 0.06916 1.00
## conditionOrthContrast2 8.576e-04 0.02928 -1.00 -1.00
## Number of obs: 4372, groups: participantCode, 243; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.18795 0.04138 4.542 5.57e-06 ***
## conditionC 0.16725 0.10259 1.630 0.103
## conditionOrthContrast2 -0.02156 0.08780 -0.246 0.806
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC
## conditionC -0.006
## cndtnOrthC2 -0.005 -0.003
## convergence code: 0
## boundary (singular) fit: see ?isSingularPairwise condition comparisons: Learnable vs. Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,exp=="exp2"&testType=="novelX"&conditionC!=0),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data: subset(d, exp == "exp2" & testType == "novelX" & conditionC !=
## 0)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3978.1 3996.0 -1986.0 3972.1 2911
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.7107 -1.0286 0.5846 0.9198 1.2856
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2208 0.4699
## Number of obs: 2914, groups: participantCode, 162
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.19742 0.05324 3.708 0.000209 ***
## conditionC 0.17020 0.10638 1.600 0.109609
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.014confint(m, method="Wald")[2:3,]## 2.5 % 97.5 %
## (Intercept) 0.09307004 0.3017737
## conditionC -0.03829648 0.3786892Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
#Learnable vs. Non-Learnable
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp2"&testType=="novelX"&conditionC!=0),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp2" & testType == "novelX" & conditionC !=
## 0)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3984.1 4020.0 -1986.0 3972.1 2908
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6985 -1.0278 0.5844 0.9194 1.2875
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.2209195 0.47002
## stimulus (Intercept) 0.0001521 0.01233
## conditionC 0.0022525 0.04746 1.00
## Number of obs: 2914, groups: participantCode, 162; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.19747 0.05334 3.702 0.000214 ***
## conditionC 0.17028 0.10699 1.592 0.111487
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.008
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.09293437 0.3020100
## conditionC -0.03941817 0.3799848Pairwise condition comparisons: No Pre-Exposure vs. Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#Non-Learnable vs. No Pre-Exposure
d$conditionNonLearnableNoPreExp <- ifelse(d$conditionC==0.5,1.5,
ifelse(d$conditionC==0,0.5,d$conditionC))
m <- glmer(isRight~conditionNonLearnableNoPreExp+(1|participantCode),data=subset(d,exp=="exp2"&testType=="novelX"&conditionNonLearnableNoPreExp!=1.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionNonLearnableNoPreExp + (1 | participantCode)
## Data:
## subset(d, exp == "exp2" & testType == "novelX" & conditionNonLearnableNoPreExp !=
## 1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3977.8 3995.7 -1985.9 3971.8 2877
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.3354 -1.0475 0.8108 0.9335 1.1046
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.07499 0.2739
## Number of obs: 2880, groups: participantCode, 160
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.13873 0.04350 3.189 0.00143 **
## conditionNonLearnableNoPreExp 0.05942 0.08696 0.683 0.49446
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnNnLNPE -0.010confint(m, method="Wald")[2:3,]## 2.5 % 97.5 %
## (Intercept) 0.05346753 0.2239883
## conditionNonLearnableNoPreExp -0.11102859 0.2298610Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~conditionNonLearnableNoPreExp+(1|participantCode)+(1+conditionNonLearnableNoPreExp|stimulus),data=subset(d,exp=="exp2"&testType=="novelX"&conditionNonLearnableNoPreExp!=1.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionNonLearnableNoPreExp + (1 | participantCode) +
## (1 + conditionNonLearnableNoPreExp | stimulus)
## Data:
## subset(d, exp == "exp2" & testType == "novelX" & conditionNonLearnableNoPreExp !=
## 1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3983.8 4019.6 -1985.9 3971.8 2874
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.3354 -1.0475 0.8108 0.9335 1.1046
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 7.499e-02 2.739e-01
## stimulus (Intercept) 0.000e+00 0.000e+00
## conditionNonLearnableNoPreExp 6.435e-13 8.022e-07 NaN
## Number of obs: 2880, groups: participantCode, 160; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.13873 0.04350 3.189 0.00143 **
## conditionNonLearnableNoPreExp 0.05942 0.08697 0.683 0.49447
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnNnLNPE -0.010
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.05346743 0.2239885
## conditionNonLearnableNoPreExp -0.11103239 0.2298644Learnable vs. No Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#Learnable vs. No Pre-Exposure
d$conditionLearnableNoPreExp <- ifelse(d$conditionC==-0.5,-1.5,
ifelse(d$conditionC==0,-0.5,d$conditionC))
m <- glmer(isRight~conditionLearnableNoPreExp+(1|participantCode),data=subset(d,exp=="exp2"&testType=="novelX"&conditionLearnableNoPreExp!=-1.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionLearnableNoPreExp + (1 | participantCode)
## Data:
## subset(d, exp == "exp2" & testType == "novelX" & conditionLearnableNoPreExp !=
## -1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 4012.6 4030.6 -2003.3 4006.6 2947
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.7797 -1.0416 0.5391 0.9040 1.2670
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2611 0.511
## Number of obs: 2950, groups: participantCode, 164
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.2318 0.0554 4.184 2.86e-05 ***
## conditionLearnableNoPreExp 0.1090 0.1105 0.986 0.324
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnLrnNPE -0.003confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.1232104 0.3403638
## conditionLearnableNoPreExp -0.1076433 0.3255717Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
#Learnable vs. No Pre-Exposure
m <- glmer(isRight~conditionLearnableNoPreExp+(1|participantCode)+(1+conditionLearnableNoPreExp|stimulus),data=subset(d,exp=="exp2"&testType=="novelX"&conditionLearnableNoPreExp!=-1.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionLearnableNoPreExp + (1 | participantCode) +
## (1 + conditionLearnableNoPreExp | stimulus)
## Data:
## subset(d, exp == "exp2" & testType == "novelX" & conditionLearnableNoPreExp !=
## -1.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 4018.6 4054.5 -2003.3 4006.6 2944
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.7669 -1.0416 0.5390 0.9039 1.2767
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.261329 0.51120
## stimulus (Intercept) 0.000271 0.01646
## conditionLearnableNoPreExp 0.002611 0.05110 1.00
## Number of obs: 2950, groups: participantCode, 164; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.23186 0.05555 4.174 3e-05 ***
## conditionLearnableNoPreExp 0.10908 0.11121 0.981 0.327
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## cndtnLrnNPE 0.004
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.1229754 0.340737
## conditionLearnableNoPreExp -0.1088836 0.327037Testing against chance: Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp2"&testType=="novelX"&condition=="Learnable Pre-Exposure"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp2" & testType == "novelX" & condition ==
## "Learnable Pre-Exposure")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1999.2 2015.1 -996.6 1993.2 1489
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.0315 -1.0447 0.4406 0.8819 1.4071
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.459444 0.67782
## stimulus (Intercept) 0.001974 0.04443
## Number of obs: 1492, groups: participantCode, 83; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.30295 0.09367 3.234 0.00122 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")[3,]## 2.5 % 97.5 %
## 0.1193661 0.4865296Testing against chance: No Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp2"&testType=="novelX"&condition=="No Pre-Exposure"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data: subset(d, exp == "exp2" & testType == "novelX" & condition ==
## "No Pre-Exposure")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2009.4 2020.0 -1002.7 2005.4 1456
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.4130 -1.0596 0.7077 0.9109 1.0877
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1042 0.3227
## Number of obs: 1458, groups: participantCode, 81
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.16987 0.06424 2.644 0.00818 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")[2,]## 2.5 % 97.5 %
## 0.04396547 0.29578076Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp2"&testType=="novelX"&condition=="No Pre-Exposure"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp2" & testType == "novelX" & condition ==
## "No Pre-Exposure")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2011.4 2027.3 -1002.7 2005.4 1455
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.4130 -1.0596 0.7077 0.9109 1.0877
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 1.042e-01 0.3227442
## stimulus (Intercept) 4.000e-14 0.0000002
## Number of obs: 1458, groups: participantCode, 81; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.16987 0.06424 2.644 0.00818 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[3,]## 2.5 % 97.5 %
## 0.04396522 0.29578106Testing against chance: Non-Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp2"&testType=="novelX"&condition=="Non-Learnable Pre-Exposure"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data: subset(d, exp == "exp2" & testType == "novelX" & condition ==
## "Non-Learnable Pre-Exposure")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1969.7 1980.2 -982.9 1965.7 1420
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.2242 -1.0456 0.8667 0.9378 1.0552
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.04776 0.2185
## Number of obs: 1422, groups: participantCode, 79
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.10830 0.05883 1.841 0.0656 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) -0.00700433 0.2235945Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp2"&testType=="novelX"&condition=="Non-Learnable Pre-Exposure"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp2" & testType == "novelX" & condition ==
## "Non-Learnable Pre-Exposure")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1971.7 1987.5 -982.9 1965.7 1419
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.2242 -1.0456 0.8667 0.9378 1.0552
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 4.776e-02 2.185e-01
## stimulus (Intercept) 2.467e-14 1.571e-07
## Number of obs: 1422, groups: participantCode, 79; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.10830 0.05883 1.841 0.0656 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[3,]## 2.5 % 97.5 %
## -0.007004624 0.223595045Interaction between test type and condition
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#recode test type
d$testTypeC <- ifelse(d$testType=="novelX",0.5,
ifelse(d$testType=="familiarX",-0.5,NA))
##all data
m <- glmer(isRight~conditionC*testTypeC+(1|participantCode),data=subset(d,exp=="exp2"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC * testTypeC + (1 | participantCode)
## Data: subset(d, exp == "exp2")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11876.0 11911.4 -5933.0 11866.0 8741
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.2135 -1.0274 0.4906 0.9289 1.3629
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2358 0.4856
## Number of obs: 8746, groups: participantCode, 243
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21043 0.03831 5.493 3.94e-08 ***
## conditionC 0.21653 0.09373 2.310 0.0209 *
## testTypeC -0.02662 0.04404 -0.604 0.5455
## conditionC:testTypeC -0.06837 0.10798 -0.633 0.5266
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC tstTyC
## conditionC -0.008
## testTypeC -0.001 -0.001
## cndtnC:tsTC -0.001 -0.001 0.000confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.13534795 0.2855030
## conditionC 0.03283099 0.4002266
## testTypeC -0.11292590 0.0596914
## conditionC:testTypeC -0.28001687 0.1432687Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
##all data
m <- glmer(isRight~conditionC*testTypeC+(1+testTypeC|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp2"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionC * testTypeC + (1 + testTypeC | participantCode) +
## (1 + conditionC | stimulus)
## Data: subset(d, exp == "exp2")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11886.0 11956.7 -5933.0 11866.0 8736
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1980 -1.0278 0.4873 0.9287 1.3661
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 2.358e-01 4.856e-01
## testTypeC 1.253e-04 1.119e-02 1.00
## stimulus (Intercept) 0.000e+00 0.000e+00
## conditionC 1.663e-11 4.079e-06 NaN
## Number of obs: 8746, groups: participantCode, 243; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.21041 0.03831 5.493 3.95e-08 ***
## conditionC 0.21650 0.09373 2.310 0.0209 *
## testTypeC -0.02562 0.04436 -0.578 0.5635
## conditionC:testTypeC -0.06697 0.10828 -0.619 0.5362
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC tstTyC
## conditionC -0.008
## testTypeC 0.013 -0.001
## cndtnC:tsTC -0.001 0.013 0.008
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[7:10,]## 2.5 % 97.5 %
## (Intercept) 0.13533088 0.28548410
## conditionC 0.03278973 0.40020560
## testTypeC -0.11257745 0.06132889
## conditionC:testTypeC -0.27918889 0.14524766Signal Detection Analysis
Sensitivity
Overall
Richter single-contrast approach.
#recode condition
subj_overall$conditionC <- ifelse(subj_overall$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_overall$condition=="Non-Learnable Pre-Exposure",-0.5,
ifelse(subj_overall$condition=="No Pre-Exposure",0,NA)))
subj_overall$conditionOrthContrast2 <- ifelse(subj_overall$condition=="Learnable Pre-Exposure",-1/3,
ifelse(subj_overall$condition=="Non-Learnable Pre-Exposure",-1/3,
ifelse(subj_overall$condition=="No Pre-Exposure",2/3,NA)))
#overall analysis (Richter single-contrast apporach)
m <- lm(dprime~conditionC,data=subset(subj_overall,exp=="exp2"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_overall,
## exp == "exp2"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8479 -0.4792 -0.1865 0.1336 3.5085
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.32047 0.05898 5.433 1.35e-07 ***
## conditionC 0.34656 0.14448 2.399 0.0172 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9193 on 241 degrees of freedom
## Multiple R-squared: 0.02332, Adjusted R-squared: 0.01926
## F-statistic: 5.754 on 1 and 241 DF, p-value: 0.01722Abelson & Prentice approach.
##check same analysis with Abelson & Prentice approach
m <- lm(dprime~conditionC+conditionOrthContrast2,data=subset(subj_overall,exp=="exp2"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC + conditionOrthContrast2, data = subset(subj_overall,
## exp == "exp2"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8619 -0.4670 -0.2005 0.1346 3.5373
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.32048 0.05909 5.423 1.43e-07 ***
## conditionC 0.34585 0.14476 2.389 0.0177 *
## conditionOrthContrast2 -0.04310 0.12534 -0.344 0.7312
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.921 on 240 degrees of freedom
## Multiple R-squared: 0.0238, Adjusted R-squared: 0.01566
## F-statistic: 2.925 on 2 and 240 DF, p-value: 0.05556Testing against chance: Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp2"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8619 -0.6489 -0.3325 -0.0675 3.3212
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.5078 0.1285 3.951 0.000164 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.171 on 82 degrees of freedomTesting against chance: No Pre-Exposure
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp2"&conditionC==0))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.1532 -0.4403 -0.2917 0.1390 3.5373
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.29174 0.09989 2.921 0.00454 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.8991 on 80 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp2"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.7869 -0.3238 -0.1619 0.2160 2.9732
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.16192 0.06558 2.469 0.0157 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5829 on 78 degrees of freedomPairwise condition comparisons
Pairwise comparison between conditions.
Learnable Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#compare conditions
#Learnable vs. Non-Learnable
m <- lm(dprime~conditionC,data=subset(subj_overall,exp=="exp2"&conditionC!=0))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_overall,
## exp == "exp2" & conditionC != 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8619 -0.5048 -0.1812 0.1203 3.3212
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.33485 0.07323 4.573 9.59e-06 ***
## conditionC 0.34585 0.14645 2.362 0.0194 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9317 on 160 degrees of freedom
## Multiple R-squared: 0.03368, Adjusted R-squared: 0.02764
## F-statistic: 5.577 on 1 and 160 DF, p-value: 0.0194No Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#Non-Learnable vs. No Pre-Exposure
subj_overall$conditionNonLearnableNoPreExp <- ifelse(subj_overall$conditionC==0.5,1.5,
ifelse(subj_overall$conditionC==0,0.5,subj_overall$conditionC))
m <- lm(dprime~conditionNonLearnableNoPreExp,data=subset(subj_overall,exp=="exp2"&conditionNonLearnableNoPreExp!=1.5))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionNonLearnableNoPreExp, data = subset(subj_overall,
## exp == "exp2" & conditionNonLearnableNoPreExp != 1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.1532 -0.4358 -0.1619 0.1580 3.5373
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.22683 0.06006 3.777 0.000224 ***
## conditionNonLearnableNoPreExp 0.12982 0.12011 1.081 0.281417
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7596 on 158 degrees of freedom
## Multiple R-squared: 0.007339, Adjusted R-squared: 0.001057
## F-statistic: 1.168 on 1 and 158 DF, p-value: 0.2814Learnable Pre-Exposure vs. No Pre-Exposure Condition
#No Pre-Exposure vs. Learnable
subj_overall$conditionLearnableNoPreExp <- ifelse(subj_overall$conditionC==-0.5,-1.5,
ifelse(subj_overall$conditionC==0,-0.5,subj_overall$conditionC))
m <- lm(dprime~conditionLearnableNoPreExp,data=subset(subj_overall,exp=="exp2"&conditionLearnableNoPreExp!=-1.5))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionLearnableNoPreExp, data = subset(subj_overall,
## exp == "exp2" & conditionLearnableNoPreExp != -1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8619 -0.5078 -0.2917 0.1302 3.5373
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.39976 0.08164 4.896 2.34e-06 ***
## conditionLearnableNoPreExp 0.21603 0.16329 1.323 0.188
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.045 on 162 degrees of freedom
## Multiple R-squared: 0.01069, Adjusted R-squared: 0.004582
## F-statistic: 1.75 on 1 and 162 DF, p-value: 0.1877Testing against chance: Learnable Pre-Exposure
#Learnable
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp2"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8619 -0.6489 -0.3325 -0.0675 3.3212
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.5078 0.1285 3.951 0.000164 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.171 on 82 degrees of freedomTesting against chance: No Pre-Exposure
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp2"&conditionC==0))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.1532 -0.4403 -0.2917 0.1390 3.5373
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.29174 0.09989 2.921 0.00454 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.8991 on 80 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp2"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.7869 -0.3238 -0.1619 0.2160 2.9732
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.16192 0.06558 2.469 0.0157 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5829 on 78 degrees of freedomFamiliar X Trials
Condition Effect
subj_testType$conditionC <- ifelse(subj_testType$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_testType$condition=="Non-Learnable Pre-Exposure",-0.5,
ifelse(subj_testType$condition=="No Pre-Exposure",0,NA)))
subj_testType$conditionOrthContrast2 <- ifelse(subj_testType$condition=="Learnable Pre-Exposure",-1/3,
ifelse(subj_testType$condition=="Non-Learnable Pre-Exposure",-1/3,
ifelse(subj_testType$condition=="No Pre-Exposure",2/3,NA)))
#Familiar X test
#overall analysis (Richter single-contrast apporach)
m <- lm(dprime~conditionC,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## exp == "exp2" & testType == "familiarX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.9241 -0.5875 -0.1333 0.3226 2.8784
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.30809 0.05993 5.141 5.66e-07 ***
## conditionC 0.34950 0.14680 2.381 0.0181 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.934 on 241 degrees of freedom
## Multiple R-squared: 0.02298, Adjusted R-squared: 0.01893
## F-statistic: 5.668 on 1 and 241 DF, p-value: 0.01805##check same analysis with Abelson & Prentice approach
m <- lm(dprime~conditionC+conditionOrthContrast2,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC + conditionOrthContrast2, data = subset(subj_testType,
## exp == "exp2" & testType == "familiarX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.9283 -0.5789 -0.1377 0.3255 2.8869
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.30809 0.06005 5.130 5.97e-07 ***
## conditionC 0.34929 0.14712 2.374 0.0184 *
## conditionOrthContrast2 -0.01288 0.12738 -0.101 0.9196
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.936 on 240 degrees of freedom
## Multiple R-squared: 0.02302, Adjusted R-squared: 0.01488
## F-statistic: 2.828 on 2 and 240 DF, p-value: 0.06113Pairwise comparisons: Learnable Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#compare conditions
#Learnable vs. Non-Learnable
m <- lm(dprime~conditionC,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionC!=0))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## exp == "exp2" & testType == "familiarX" & conditionC != 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.9283 -0.4870 -0.1377 0.3415 2.6994
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.31238 0.07502 4.164 5.1e-05 ***
## conditionC 0.34929 0.15004 2.328 0.0212 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9546 on 160 degrees of freedom
## Multiple R-squared: 0.03276, Adjusted R-squared: 0.02672
## F-statistic: 5.419 on 1 and 160 DF, p-value: 0.02117No Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#Non-Learnable vs. No Pre-Exposure
subj_testType$conditionNonLearnableNoPreExp <- ifelse(subj_testType$conditionC==0.5,1.5,
ifelse(subj_testType$conditionC==0,0.5,subj_testType$conditionC))
m <- lm(dprime~conditionNonLearnableNoPreExp,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionNonLearnableNoPreExp!=1.5))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionNonLearnableNoPreExp, data = subset(subj_testType,
## exp == "exp2" & testType == "familiarX" & conditionNonLearnableNoPreExp !=
## 1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.3804 -0.4395 -0.1377 0.3200 2.8869
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.21862 0.06352 3.442 0.00074 ***
## conditionNonLearnableNoPreExp 0.16177 0.12704 1.273 0.20475
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.8034 on 158 degrees of freedom
## Multiple R-squared: 0.01016, Adjusted R-squared: 0.003894
## F-statistic: 1.622 on 1 and 158 DF, p-value: 0.2048Learnable Pre-Exposure vs. No Pre-Exposure Condition
#No Pre-Exposure vs. Learnable
subj_testType$conditionLearnableNoPreExp <- ifelse(subj_testType$conditionC==-0.5,-1.5,
ifelse(subj_testType$conditionC==0,-0.5,subj_testType$conditionC))
m <- lm(dprime~conditionLearnableNoPreExp,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionLearnableNoPreExp!=-1.5))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionLearnableNoPreExp, data = subset(subj_testType,
## exp == "exp2" & testType == "familiarX" & conditionLearnableNoPreExp !=
## -1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.9283 -0.6335 -0.2478 0.3295 2.8869
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.39327 0.08064 4.877 2.55e-06 ***
## conditionLearnableNoPreExp 0.18752 0.16127 1.163 0.247
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.033 on 162 degrees of freedom
## Multiple R-squared: 0.008277, Adjusted R-squared: 0.002155
## F-statistic: 1.352 on 1 and 162 DF, p-value: 0.2466Testing against chance: Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp2" & testType == "familiarX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.9283 -0.7781 -0.1960 0.3415 2.6994
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.4870 0.1261 3.861 0.000224 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.149 on 82 degrees of freedomTesting against chance: No Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionC==0))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp2" & testType == "familiarX" & conditionC == 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.3804 -0.5905 -0.2995 0.3255 2.8869
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.29951 0.09973 3.003 0.00357 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.8976 on 80 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp2" & testType == "familiarX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.3332 -0.4288 -0.1377 0.3182 2.6761
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.13774 0.07804 1.765 0.0815 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6936 on 78 degrees of freedomNovel X Trials
Condition Effect
subj_testType$conditionC <- ifelse(subj_testType$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_testType$condition=="Non-Learnable Pre-Exposure",-0.5,
ifelse(subj_testType$condition=="No Pre-Exposure",0,NA)))
subj_testType$conditionOrthContrast2 <- ifelse(subj_testType$condition=="Learnable Pre-Exposure",-1/3,
ifelse(subj_testType$condition=="Non-Learnable Pre-Exposure",-1/3,
ifelse(subj_testType$condition=="No Pre-Exposure",2/3,NA)))
#Novel X test
#overall analysis (Richter single-contrast apporach)
m <- lm(dprime~conditionC,data=subset(subj_testType,exp=="exp2"&testType=="novelX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## exp == "exp2" & testType == "novelX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.1235 -0.4654 -0.1111 0.2987 2.9147
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.27170 0.05882 4.62 6.27e-06 ***
## conditionC 0.23767 0.14407 1.65 0.1
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9167 on 241 degrees of freedom
## Multiple R-squared: 0.01117, Adjusted R-squared: 0.007063
## F-statistic: 2.721 on 1 and 241 DF, p-value: 0.1003##check same analysis with Abelson & Prentice approach
m <- lm(dprime~conditionC+conditionOrthContrast2,data=subset(subj_testType,exp=="exp2"&testType=="novelX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC + conditionOrthContrast2, data = subset(subj_testType,
## exp == "exp2" & testType == "novelX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.1357 -0.4782 -0.1233 0.2902 2.9398
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.27171 0.05893 4.611 6.52e-06 ***
## conditionC 0.23705 0.14436 1.642 0.102
## conditionOrthContrast2 -0.03756 0.12499 -0.301 0.764
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9184 on 240 degrees of freedom
## Multiple R-squared: 0.01154, Adjusted R-squared: 0.003301
## F-statistic: 1.401 on 2 and 240 DF, p-value: 0.2484Pairwise comparisons: Learnable Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#compare conditions
#Learnable vs. Non-Learnable
m <- lm(dprime~conditionC,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionC!=0))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## exp == "exp2" & testType == "novelX" & conditionC != 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.1357 -0.4889 -0.1233 0.3761 2.7837
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.28423 0.07475 3.802 0.000204 ***
## conditionC 0.23705 0.14951 1.586 0.114827
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9512 on 160 degrees of freedom
## Multiple R-squared: 0.01547, Adjusted R-squared: 0.009315
## F-statistic: 2.514 on 1 and 160 DF, p-value: 0.1148No Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#Non-Learnable vs. No Pre-Exposure
subj_testType$conditionNonLearnableNoPreExp <- ifelse(subj_testType$conditionC==0.5,1.5,
ifelse(subj_testType$conditionC==0,0.5,subj_testType$conditionC))
m <- lm(dprime~conditionNonLearnableNoPreExp,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionNonLearnableNoPreExp!=1.5))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionNonLearnableNoPreExp, data = subset(subj_testType,
## exp == "exp2" & testType == "novelX" & conditionNonLearnableNoPreExp !=
## 1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8171 -0.4567 -0.1657 0.3238 2.9398
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.20618 0.06324 3.26 0.00136 **
## conditionNonLearnableNoPreExp 0.08096 0.12648 0.64 0.52302
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7998 on 158 degrees of freedom
## Multiple R-squared: 0.002587, Adjusted R-squared: -0.003726
## F-statistic: 0.4098 on 1 and 158 DF, p-value: 0.523Learnable Pre-Exposure vs. No Pre-Exposure Condition
#No Pre-Exposure vs. Learnable
subj_testType$conditionLearnableNoPreExp <- ifelse(subj_testType$conditionC==-0.5,-1.5,
ifelse(subj_testType$conditionC==0,-0.5,subj_testType$conditionC))
m <- lm(dprime~conditionLearnableNoPreExp,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionLearnableNoPreExp!=-1.5))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionLearnableNoPreExp, data = subset(subj_testType,
## exp == "exp2" & testType == "novelX" & conditionLearnableNoPreExp !=
## -1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.1357 -0.5290 -0.1233 0.1781 2.9398
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.32471 0.07737 4.197 4.45e-05 ***
## conditionLearnableNoPreExp 0.15609 0.15474 1.009 0.315
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9907 on 162 degrees of freedom
## Multiple R-squared: 0.006242, Adjusted R-squared: 0.0001073
## F-statistic: 1.017 on 1 and 162 DF, p-value: 0.3146Testing against chance: Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp2" & testType == "novelX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.1357 -0.6938 -0.1233 0.1950 2.7837
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.4028 0.1220 3.3 0.00143 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.112 on 82 degrees of freedomTesting against chance: No Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionC==0))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp2" & testType == "novelX" & conditionC == 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.4421 -0.2467 -0.2467 0.1259 2.9398
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.24666 0.09434 2.615 0.0107 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.849 on 80 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp2" & testType == "novelX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8171 -0.4567 0.1137 0.4047 2.6482
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.16570 0.08394 1.974 0.0519 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7461 on 78 degrees of freedomInteraction between Novel X and Familiar X trials
#recode test type
subj_testType$testTypeC <- ifelse(subj_testType$testType=="novelX",0.5,
ifelse(subj_testType$testType=="familiarX",-0.5,NA))
m <- lm(dprime~conditionC*testTypeC,data=subset(subj_testType,exp=="exp2"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC * testTypeC, data = subset(subj_testType,
## exp == "exp2"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.9241 -0.5011 -0.1333 0.3061 2.9147
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.28990 0.04199 6.905 1.59e-11 ***
## conditionC 0.29359 0.10284 2.855 0.00449 **
## testTypeC -0.03639 0.08397 -0.433 0.66494
## conditionC:testTypeC -0.11184 0.20568 -0.544 0.58688
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9254 on 482 degrees of freedom
## Multiple R-squared: 0.01761, Adjusted R-squared: 0.0115
## F-statistic: 2.881 on 3 and 482 DF, p-value: 0.03551Response Bias
Overall
Unlike with the sensitivty/ accuracy measures, we had no specific hypothesis (linear or otherwise) with respect to response bias. Hence, we test for an overall effect of condition (multi-df, omnibus effect) with condition dummy-coded.
subj_testType$conditionC <- ifelse(subj_testType$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_testType$condition=="Non-Learnable Pre-Exposure",-0.5,
ifelse(subj_testType$condition=="No Pre-Exposure",0,NA)))
#overall analysis (condition dummy-coded)
m <- lm(c~condition,data=subset(subj_overall,exp=="exp2"))
summary(m)##
## Call:
## lm(formula = c ~ condition, data = subset(subj_overall, exp ==
## "exp2"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.63271 -0.16921 0.05426 0.20904 1.02369
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.19537 0.04429 -4.411 1.55e-05
## conditionNo Pre-Exposure -0.08643 0.06302 -1.371 0.1715
## conditionNon-Learnable Pre-Exposure -0.10824 0.06342 -1.707 0.0892
##
## (Intercept) ***
## conditionNo Pre-Exposure
## conditionNon-Learnable Pre-Exposure .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4035 on 240 degrees of freedom
## Multiple R-squared: 0.01352, Adjusted R-squared: 0.005295
## F-statistic: 1.644 on 2 and 240 DF, p-value: 0.1954Anova(m, type="III")## Anova Table (Type III tests)
##
## Response: c
## Sum Sq Df F value Pr(>F)
## (Intercept) 3.168 1 19.4595 1.552e-05 ***
## condition 0.535 2 1.6441 0.1954
## Residuals 39.073 240
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1There is no significant overall effect of condition on response bias c - indicating that response bias does not differ overall across conditions.
Testing against chance: Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_overall,exp=="exp2"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.02527 -0.16921 0.05426 0.19537 1.01152
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.19537 0.03824 -5.108 2.08e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3484 on 82 degrees of freedomTesting against chance: No Pre-Exposure
m <- lm(c~1,data=subset(subj_overall,exp=="exp2"&conditionC==0))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.63271 -0.20191 -0.00042 0.28180 1.02369
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.28180 0.05338 -5.279 1.09e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4804 on 80 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_overall,exp=="exp2"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.61089 -0.14041 0.08825 0.22831 0.66820
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.30361 0.04158 -7.301 2.09e-10 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3696 on 78 degrees of freedomPairwise condition comparisons
Pairwise comparison between conditions.
Learnable Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#compare conditions
#Learnable vs. Non-Learnable
m <- lm(c~conditionC,data=subset(subj_overall,exp=="exp2"&conditionC!=0))
summary(m)##
## Call:
## lm(formula = c ~ conditionC, data = subset(subj_overall, exp ==
## "exp2" & conditionC != 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.61089 -0.16718 0.05496 0.19537 1.01152
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.24949 0.02821 -8.845 1.6e-15 ***
## conditionC 0.10824 0.05641 1.919 0.0568 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3589 on 160 degrees of freedom
## Multiple R-squared: 0.02249, Adjusted R-squared: 0.01638
## F-statistic: 3.682 on 1 and 160 DF, p-value: 0.0568No Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#Non-Learnable vs. No Pre-Exposure
subj_overall$conditionNonLearnableNoPreExp <- ifelse(subj_overall$conditionC==0.5,1.5,
ifelse(subj_overall$conditionC==0,0.5,subj_overall$conditionC))
m <- lm(c~conditionNonLearnableNoPreExp,data=subset(subj_overall,exp=="exp2"&conditionNonLearnableNoPreExp!=1.5))
summary(m)##
## Call:
## lm(formula = c ~ conditionNonLearnableNoPreExp, data = subset(subj_overall,
## exp == "exp2" & conditionNonLearnableNoPreExp != 1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.63271 -0.16055 0.01839 0.23236 1.02369
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.29271 0.03394 -8.624 6.44e-15 ***
## conditionNonLearnableNoPreExp 0.02182 0.06788 0.321 0.748
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4293 on 158 degrees of freedom
## Multiple R-squared: 0.0006532, Adjusted R-squared: -0.005672
## F-statistic: 0.1033 on 1 and 158 DF, p-value: 0.7484Learnable Pre-Exposure vs. No Pre-Exposure Condition
#No Pre-Exposure vs. Learnable
subj_overall$conditionLearnableNoPreExp <- ifelse(subj_overall$conditionC==-0.5,-1.5,
ifelse(subj_overall$conditionC==0,-0.5,subj_overall$conditionC))
m <- lm(c~conditionLearnableNoPreExp,data=subset(subj_overall,exp=="exp2"&conditionLearnableNoPreExp!=-1.5))
summary(m)##
## Call:
## lm(formula = c ~ conditionLearnableNoPreExp, data = subset(subj_overall,
## exp == "exp2" & conditionLearnableNoPreExp != -1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.63271 -0.16921 0.05426 0.20754 1.02369
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.23858 0.03271 -7.294 1.26e-11 ***
## conditionLearnableNoPreExp 0.08643 0.06542 1.321 0.188
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4188 on 162 degrees of freedom
## Multiple R-squared: 0.01066, Adjusted R-squared: 0.004553
## F-statistic: 1.746 on 1 and 162 DF, p-value: 0.1883Testing against chance: Learnable Pre-Exposure
#Learnable
m <- lm(c~1,data=subset(subj_overall,exp=="exp2"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.02527 -0.16921 0.05426 0.19537 1.01152
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.19537 0.03824 -5.108 2.08e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3484 on 82 degrees of freedomTesting against chance: No Pre-Exposure
m <- lm(c~1,data=subset(subj_overall,exp=="exp2"&conditionC==0))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.63271 -0.20191 -0.00042 0.28180 1.02369
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.28180 0.05338 -5.279 1.09e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4804 on 80 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_overall,exp=="exp2"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp2" &
## conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.61089 -0.14041 0.08825 0.22831 0.66820
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.30361 0.04158 -7.301 2.09e-10 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3696 on 78 degrees of freedomFamiliar X Trials
Condition Effect
#overall analysis (condition dummy-coded)
m <- lm(c~condition,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"))
summary(m)##
## Call:
## lm(formula = c ~ condition, data = subset(subj_testType, exp ==
## "exp2" & testType == "familiarX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.18484 -0.25249 0.09588 0.33304 1.09775
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.40838 0.05008 -8.154 1.96e-14
## conditionNo Pre-Exposure -0.23716 0.07127 -3.328 0.00101
## conditionNon-Learnable Pre-Exposure -0.16481 0.07172 -2.298 0.02243
##
## (Intercept) ***
## conditionNo Pre-Exposure **
## conditionNon-Learnable Pre-Exposure *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4563 on 240 degrees of freedom
## Multiple R-squared: 0.04632, Adjusted R-squared: 0.03837
## F-statistic: 5.828 on 2 and 240 DF, p-value: 0.003376Anova(m, type="III")## Anova Table (Type III tests)
##
## Response: c
## Sum Sq Df F value Pr(>F)
## (Intercept) 13.842 1 66.4859 1.964e-14 ***
## condition 2.427 2 5.8281 0.003376 **
## Residuals 49.968 240
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Response bias differs across conditions.
Pairwise comparisons: Learnable Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#compare conditions
#Learnable vs. Non-Learnable
m <- lm(c~conditionC,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionC!=0))
summary(m)##
## Call:
## lm(formula = c ~ conditionC, data = subset(subj_testType, exp ==
## "exp2" & testType == "familiarX" & conditionC != 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.1848 -0.2372 0.1232 0.2880 0.8226
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.49079 0.03425 -14.328 <2e-16 ***
## conditionC 0.16481 0.06851 2.406 0.0173 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4359 on 160 degrees of freedom
## Multiple R-squared: 0.03491, Adjusted R-squared: 0.02888
## F-statistic: 5.787 on 1 and 160 DF, p-value: 0.01728No Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#Non-Learnable vs. No Pre-Exposure
subj_testType$conditionNonLearnableNoPreExp <- ifelse(subj_testType$conditionC==0.5,1.5,
ifelse(subj_testType$conditionC==0,0.5,subj_testType$conditionC))
m <- lm(c~conditionNonLearnableNoPreExp,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionNonLearnableNoPreExp!=1.5))
summary(m)##
## Call:
## lm(formula = c ~ conditionNonLearnableNoPreExp, data = subset(subj_testType,
## exp == "exp2" & testType == "familiarX" & conditionNonLearnableNoPreExp !=
## 1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.02003 -0.25249 0.07645 0.36032 1.09775
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.60937 0.03793 -16.067 <2e-16 ***
## conditionNonLearnableNoPreExp -0.07235 0.07585 -0.954 0.342
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4797 on 158 degrees of freedom
## Multiple R-squared: 0.005724, Adjusted R-squared: -0.0005685
## F-statistic: 0.9097 on 1 and 158 DF, p-value: 0.3417Learnable Pre-Exposure vs. No Pre-Exposure Condition
#No Pre-Exposure vs. Learnable
subj_testType$conditionLearnableNoPreExp <- ifelse(subj_testType$conditionC==-0.5,-1.5,
ifelse(subj_testType$conditionC==0,-0.5,subj_testType$conditionC))
m <- lm(c~conditionLearnableNoPreExp,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionLearnableNoPreExp!=-1.5))
summary(m)##
## Call:
## lm(formula = c ~ conditionLearnableNoPreExp, data = subset(subj_testType,
## exp == "exp2" & testType == "familiarX" & conditionLearnableNoPreExp !=
## -1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.18484 -0.19723 0.04782 0.36032 1.09775
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.52696 0.03534 -14.909 < 2e-16 ***
## conditionLearnableNoPreExp 0.23716 0.07069 3.355 0.000989 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4526 on 162 degrees of freedom
## Multiple R-squared: 0.06497, Adjusted R-squared: 0.05919
## F-statistic: 11.26 on 1 and 162 DF, p-value: 0.0009888Testing against chance: Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp2" &
## testType == "familiarX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.18484 -0.18934 0.09588 0.26287 0.82264
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.40838 0.04472 -9.132 3.93e-14 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4074 on 82 degrees of freedomTesting against chance: No Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionC==0))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp2" &
## testType == "familiarX" & conditionC == 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.94768 -0.34714 0.04782 0.36032 1.09775
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.64554 0.05496 -11.75 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4946 on 80 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp2"&testType=="familiarX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp2" &
## testType == "familiarX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.0200 -0.2525 0.1425 0.2880 0.7595
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.57319 0.05219 -10.98 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4639 on 78 degrees of freedomNovel X Trials
Condition Effect
#overall analysis (condition dummy-coded)
m <- lm(c~condition,data=subset(subj_testType,exp=="exp2"&testType=="novelX"))
summary(m)##
## Call:
## lm(formula = c ~ condition, data = subset(subj_testType, exp ==
## "exp2" & testType == "novelX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.67979 -0.27658 -0.08657 0.19622 1.64149
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.003716 0.065549 0.057 0.955
## conditionNo Pre-Exposure 0.082851 0.093270 0.888 0.375
## conditionNon-Learnable Pre-Exposure -0.051984 0.093866 -0.554 0.580
##
## Residual standard error: 0.5972 on 240 degrees of freedom
## Multiple R-squared: 0.00859, Adjusted R-squared: 0.0003284
## F-statistic: 1.04 on 2 and 240 DF, p-value: 0.3551Anova(m, type="III")## Anova Table (Type III tests)
##
## Response: c
## Sum Sq Df F value Pr(>F)
## (Intercept) 0.001 1 0.0032 0.9548
## condition 0.742 2 1.0397 0.3551
## Residuals 85.588 240Pairwise comparisons: Learnable Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#compare conditions
#Learnable vs. Non-Learnable
m <- lm(c~conditionC,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionC!=0))
summary(m)##
## Call:
## lm(formula = c ~ conditionC, data = subset(subj_testType, exp ==
## "exp2" & testType == "novelX" & conditionC != 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.54495 -0.23695 -0.00372 0.19378 1.64149
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.02228 0.03956 -0.563 0.574
## conditionC 0.05198 0.07912 0.657 0.512
##
## Residual standard error: 0.5034 on 160 degrees of freedom
## Multiple R-squared: 0.002691, Adjusted R-squared: -0.003543
## F-statistic: 0.4316 on 1 and 160 DF, p-value: 0.5121No Pre-Exposure vs. Non-Learnable Pre-Exposure Condition
#Non-Learnable vs. No Pre-Exposure
subj_testType$conditionNonLearnableNoPreExp <- ifelse(subj_testType$conditionC==0.5,1.5,
ifelse(subj_testType$conditionC==0,0.5,subj_testType$conditionC))
m <- lm(c~conditionNonLearnableNoPreExp,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionNonLearnableNoPreExp!=1.5))
summary(m)##
## Call:
## lm(formula = c ~ conditionNonLearnableNoPreExp, data = subset(subj_testType,
## exp == "exp2" & testType == "novelX" & conditionNonLearnableNoPreExp !=
## 1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.67979 -0.28484 -0.08657 0.19378 1.64149
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.01915 0.05076 0.377 0.707
## conditionNonLearnableNoPreExp 0.13484 0.10152 1.328 0.186
##
## Residual standard error: 0.642 on 158 degrees of freedom
## Multiple R-squared: 0.01104, Adjusted R-squared: 0.004782
## F-statistic: 1.764 on 1 and 158 DF, p-value: 0.186Learnable Pre-Exposure vs. No Pre-Exposure Condition
#No Pre-Exposure vs. Learnable
subj_testType$conditionLearnableNoPreExp <- ifelse(subj_testType$conditionC==-0.5,-1.5,
ifelse(subj_testType$conditionC==0,-0.5,subj_testType$conditionC))
m <- lm(c~conditionLearnableNoPreExp,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionLearnableNoPreExp!=-1.5))
summary(m)##
## Call:
## lm(formula = c ~ conditionLearnableNoPreExp, data = subset(subj_testType,
## exp == "exp2" & testType == "novelX" & conditionLearnableNoPreExp !=
## -1.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.67979 -0.33011 -0.08657 0.20505 1.58950
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.04514 0.04966 0.909 0.365
## conditionLearnableNoPreExp -0.08285 0.09932 -0.834 0.405
##
## Residual standard error: 0.6359 on 162 degrees of freedom
## Multiple R-squared: 0.004277, Adjusted R-squared: -0.001869
## F-statistic: 0.6959 on 1 and 162 DF, p-value: 0.4054Testing against chance: Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp2" &
## testType == "novelX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.41065 -0.23947 -0.00372 0.22923 1.58950
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.003716 0.054831 0.068 0.946
##
## Residual standard error: 0.4995 on 82 degrees of freedomTesting against chance: No Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionC==0))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp2" &
## testType == "novelX" & conditionC == 0))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.67979 -0.37178 -0.08657 0.19865 1.50665
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.08657 0.08337 1.038 0.302
##
## Residual standard error: 0.7504 on 80 degrees of freedomTesting against chance: Non-Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp2"&testType=="novelX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp2" &
## testType == "novelX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.54495 -0.23695 -0.09144 0.19088 1.64149
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.04827 0.05709 -0.846 0.4
##
## Residual standard error: 0.5074 on 78 degrees of freedomInteraction between Novel X and Familiar X trials
#recode test type
subj_testType$testTypeC <- ifelse(subj_testType$testType=="novelX",0.5,
ifelse(subj_testType$testType=="familiarX",-0.5,NA))
m <- lm(c~condition*testTypeC,data=subset(subj_testType,exp=="exp2"))
summary(m)##
## Call:
## lm(formula = c ~ condition * testTypeC, data = subset(subj_testType,
## exp == "exp2"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.67979 -0.26156 -0.00372 0.28150 1.64149
##
## Coefficients:
## Estimate Std. Error t value
## (Intercept) -0.20233 0.04125 -4.905
## conditionNo Pre-Exposure -0.07715 0.05869 -1.315
## conditionNon-Learnable Pre-Exposure -0.10840 0.05906 -1.835
## testTypeC 0.41210 0.08249 4.996
## conditionNo Pre-Exposure:testTypeC 0.32001 0.11738 2.726
## conditionNon-Learnable Pre-Exposure:testTypeC 0.11283 0.11813 0.955
## Pr(>|t|)
## (Intercept) 1.28e-06 ***
## conditionNo Pre-Exposure 0.18928
## conditionNon-Learnable Pre-Exposure 0.06709 .
## testTypeC 8.23e-07 ***
## conditionNo Pre-Exposure:testTypeC 0.00664 **
## conditionNon-Learnable Pre-Exposure:testTypeC 0.34000
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5314 on 480 degrees of freedom
## Multiple R-squared: 0.2307, Adjusted R-squared: 0.2227
## F-statistic: 28.79 on 5 and 480 DF, p-value: < 2.2e-16Anova(m, type="III")## Anova Table (Type III tests)
##
## Response: c
## Sum Sq Df F value Pr(>F)
## (Intercept) 6.796 1 24.0638 1.278e-06 ***
## condition 1.015 2 1.7969 0.16693
## testTypeC 7.048 1 24.9557 8.230e-07 ***
## condition:testTypeC 2.154 2 3.8127 0.02276 *
## Residuals 135.557 480
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Experiment 3
Test Accuracy
Plot
#Familiar X test
p1 <- ggplot(subset(test_type_exp,testType=="familiarX"&exp=="exp3"),aes(condition,accuracy,fill=condition, color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="familiarX"&exp=="exp3"),aes(y=acc), width = 0.05, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, size=0.8)+
theme_classic(base_size=16)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Unstructured Pre-Exposure","Learnable Pre-Exposure"),
labels=c("Unstructured\nPre-Exposure","Learnable\nPre-Exposure"),
limits=c("Unstructured Pre-Exposure","Learnable Pre-Exposure"))+
scale_fill_manual(values=c("#4DAF4A","#54278f"))+
scale_color_manual(values=c("#4DAF4A","#54278f"))+
theme(legend.position="none")+
ylab("Accuracy - Familiar X Test")+
xlab("Condition")
#Novel X
p2 <- ggplot(subset(test_type_exp,testType=="novelX"&exp=="exp3"),aes(condition,accuracy,fill=condition,color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="novelX"&exp=="exp3"),aes(y=acc), width = 0.05, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, size=0.8)+
theme_classic(base_size=16)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Unstructured Pre-Exposure","Learnable Pre-Exposure"),
labels=c("Unstructured\nPre-Exposure","Learnable\nPre-Exposure"),
limits=c("Unstructured Pre-Exposure","Learnable Pre-Exposure"))+
scale_fill_manual(values=c("#4DAF4A","#54278f"))+
scale_color_manual(values=c("#4DAF4A","#54278f"))+
theme(legend.position="none")+
ylab("Accuracy - Novel X Test")+
xlab("Condition")
plot_grid(p1,p2, labels=c("A","B"),label_size=18)
Overall Accuracy Table
overall_exp %>%
filter(exp=="exp3") %>%
select(-c(se, accuracy_ci,d_prime_ci,c_bias_ci)) %>%
formattable()| exp | condition | N | accuracy | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|---|
| exp3 | Learnable Pre-Exposure | 90 | 0.6157407 | 0.5761732 | 0.6553082 | 0.7932834 | 0.5051798 | 1.0813870 | -0.1471900 | -0.2255496 | -0.06883044 |
| exp3 | Unstructured Pre-Exposure | 89 | 0.5355805 | 0.5085204 | 0.5626406 | 0.2401722 | 0.0604255 | 0.4199189 | -0.1638137 | -0.2497717 | -0.07785573 |
Accuracy Table By Test Type
test_type_exp %>%
filter(exp=="exp3") %>%
select(-c(se, accuracy_ci,d_prime_ci,c_bias_ci)) %>%
formattable()| exp | condition | testType | N | accuracy | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| exp3 | Learnable Pre-Exposure | familiarX | 90 | 0.6197531 | 0.5787377 | 0.6607685 | 0.7360229 | 0.48172951 | 0.9903163 | -0.3575601 | -0.45300126 | -0.2621189 |
| exp3 | Learnable Pre-Exposure | novelX | 90 | 0.6117284 | 0.5682435 | 0.6552133 | 0.6917028 | 0.41857251 | 0.9648331 | 0.0653921 | -0.04610090 | 0.1768851 |
| exp3 | Unstructured Pre-Exposure | familiarX | 89 | 0.5380774 | 0.5053266 | 0.5708282 | 0.2373254 | 0.03843109 | 0.4362197 | -0.4554699 | -0.55536408 | -0.3555757 |
| exp3 | Unstructured Pre-Exposure | novelX | 89 | 0.5330836 | 0.5043785 | 0.5617888 | 0.2000827 | 0.02878172 | 0.3713837 | 0.1334533 | -0.01092315 | 0.2778297 |
Correlation
Correlation between Familiar X and Novel X Trials
Learnable Pre-Exposure
##Correlations between Familiar X and Novel X
c <- corr.test(subset(subj_accuracy_wide, condition=="Learnable Pre-Exposure" & exp=="exp3")[,c("novelX","familiarX")])
c## Call:corr.test(x = subset(subj_accuracy_wide, condition == "Learnable Pre-Exposure" &
## exp == "exp3")[, c("novelX", "familiarX")])
## Correlation matrix
## novelX familiarX
## novelX 1.00 0.75
## familiarX 0.75 1.00
## Sample Size
## [1] 90
## Probability values (Entries above the diagonal are adjusted for multiple tests.)
## novelX familiarX
## novelX 0 0
## familiarX 0 0
##
## To see confidence intervals of the correlations, print with the short=FALSE optionc$p## novelX familiarX
## novelX 0.000000e+00 9.870517e-18
## familiarX 9.870517e-18 0.000000e+00Unstructured Pre-Exposure
c <- corr.test(subset(subj_accuracy_wide, condition=="Unstructured Pre-Exposure" & exp=="exp3")[,c("novelX","familiarX")])
c## Call:corr.test(x = subset(subj_accuracy_wide, condition == "Unstructured Pre-Exposure" &
## exp == "exp3")[, c("novelX", "familiarX")])
## Correlation matrix
## novelX familiarX
## novelX 1.00 0.55
## familiarX 0.55 1.00
## Sample Size
## [1] 89
## Probability values (Entries above the diagonal are adjusted for multiple tests.)
## novelX familiarX
## novelX 0 0
## familiarX 0 0
##
## To see confidence intervals of the correlations, print with the short=FALSE optionc$p## novelX familiarX
## novelX 0.000000e+00 2.524806e-08
## familiarX 2.524806e-08 0.000000e+00Interaction between Test Type and Condition
m <- lm(novelX~familiarX*condition, subset(subj_accuracy_wide,exp=="exp3"))
summary(m)##
## Call:
## lm(formula = novelX ~ familiarX * condition, data = subset(subj_accuracy_wide,
## exp == "exp3"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.33799 -0.08303 0.01198 0.07857 0.36172
##
## Coefficients:
## Estimate Std. Error t value
## (Intercept) 0.11638 0.04446 2.618
## familiarX 0.79926 0.06844 11.679
## conditionUnstructured Pre-Exposure 0.15775 0.06582 2.397
## familiarX:conditionUnstructured Pre-Exposure -0.31801 0.11045 -2.879
## Pr(>|t|)
## (Intercept) 0.00963 **
## familiarX < 2e-16 ***
## conditionUnstructured Pre-Exposure 0.01759 *
## familiarX:conditionUnstructured Pre-Exposure 0.00448 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.1264 on 175 degrees of freedom
## Multiple R-squared: 0.5132, Adjusted R-squared: 0.5049
## F-statistic: 61.51 on 3 and 175 DF, p-value: < 2.2e-16Correlation Plot
ggplot(subset(subj_accuracy_wide,exp=="exp3"),aes(familiarX,novelX, color=condition,linetype=condition,shape=condition))+
geom_jitter(width=0.02,height=0.02,size=3)+
geom_smooth(method=lm,se =F,size=1.5)+
scale_color_manual(name="Condition",values=c("#4DAF4A","#54278f"))+
scale_linetype_manual(name="Condition",values=c(1,5))+
scale_shape_manual(name="Condition",values=c(16,18))+
ylab("Accuracy - Novel X")+
xlab("Accuracy - Familiar X")+
theme_classic(base_size=18)+
theme(legend.position=c(0.3,0.9))
Logistic Mixed-Effects Model
Overall
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#recode condition
d$conditionC <- ifelse(d$condition=="Learnable Pre-Exposure",0.5,
ifelse(d$condition=="Unstructured Pre-Exposure",-0.5,NA))
##all data
m <- glmer(isRight~conditionC+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp3"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp3")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 8420.2 8447.2 -4206.1 8412.2 6440
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.0204 -1.0153 0.3568 0.9141 1.4900
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.600740 0.77507
## stimulus (Intercept) 0.004486 0.06697
## Number of obs: 6444, groups: participantCode, 179; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.3848 0.0655 5.876 4.21e-09 ***
## conditionC 0.4254 0.1285 3.309 0.000935 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.019confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.2564603 0.5131961
## conditionC 0.1734409 0.6772993Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
#maximal model
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp3"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp3")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 8423.4 8464.0 -4205.7 8411.4 6438
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.0103 -1.0155 0.3546 0.9121 1.5246
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.601095 0.77530
## stimulus (Intercept) 0.006028 0.07764
## conditionC 0.005060 0.07113 -1.00
## Number of obs: 6444, groups: participantCode, 179; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.38485 0.06584 5.845 5.06e-09 ***
## conditionC 0.42487 0.12913 3.290 0.001 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.001
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.2558063 0.5138941
## conditionC 0.1717902 0.6779561Familiar X Trials
Condition Effect
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,exp=="exp3"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data: subset(d, exp == "exp3" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 4271.3 4289.5 -2132.7 4265.3 3219
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1365 -1.0299 0.4681 0.9022 1.3070
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.4302 0.6559
## Number of obs: 3222, groups: participantCode, 179
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.36486 0.06222 5.864 4.52e-09 ***
## conditionC 0.38408 0.12387 3.101 0.00193 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.018confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.2429110 0.4868076
## conditionC 0.1412912 0.6268640Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
#maximal model
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp3"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa")) #convergence warning## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp3" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 4277.3 4313.8 -2132.7 4265.3 3216
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1365 -1.0299 0.4681 0.9022 1.3070
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 4.302e-01 6.559e-01
## stimulus (Intercept) 2.777e-12 1.666e-06
## conditionC 8.156e-12 2.856e-06 -1.00
## Number of obs: 3222, groups: participantCode, 179; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.36486 0.06222 5.864 4.52e-09 ***
## conditionC 0.38408 0.12387 3.101 0.00193 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.018
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.2429104 0.4868084
## conditionC 0.1412901 0.6268634Testing against chance: Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp3"&testType=="familiarX"&conditionC==0.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data:
## subset(d, exp == "exp3" & testType == "familiarX" & conditionC ==
## 0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2064.3 2075.1 -1030.2 2060.3 1618
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.4478 -0.9814 0.4085 0.7845 1.3262
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.76 0.8718
## Number of obs: 1620, groups: participantCode, 90
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.5939 0.1090 5.45 5.04e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.3803136 0.8074874Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp3"&testType=="familiarX"&conditionC==0.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data:
## subset(d, exp == "exp3" & testType == "familiarX" & conditionC ==
## 0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2066.3 2082.5 -1030.2 2060.3 1617
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.4478 -0.9814 0.4085 0.7845 1.3262
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.76 0.8718
## stimulus (Intercept) 0.00 0.0000
## Number of obs: 1620, groups: participantCode, 90; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.5939 0.1090 5.45 5.04e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.3803169 0.8074857Testing against chance: Unstructured Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp3"&testType=="familiarX"&conditionC==-0.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data:
## subset(d, exp == "exp3" & testType == "familiarX" & conditionC ==
## -0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2197.9 2208.7 -1097.0 2193.9 1600
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6037 -1.0443 0.6236 0.9092 1.1787
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1937 0.4401
## Number of obs: 1602, groups: participantCode, 89
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.16208 0.06942 2.335 0.0195 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.02602679 0.2981361Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp3"&testType=="familiarX"&conditionC==-0.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data:
## subset(d, exp == "exp3" & testType == "familiarX" & conditionC ==
## -0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2199.9 2216.1 -1097.0 2193.9 1599
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.6037 -1.0443 0.6236 0.9092 1.1787
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.1937 0.4401
## stimulus (Intercept) 0.0000 0.0000
## Number of obs: 1602, groups: participantCode, 89; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.16208 0.06942 2.335 0.0195 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.02602678 0.2981357Novel X Trials
Condition Effect
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~conditionC+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp3"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp3" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 4288.8 4313.1 -2140.4 4280.8 3218
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1169 -1.0178 0.4405 0.9043 1.6234
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.4370 0.6610
## stimulus (Intercept) 0.0136 0.1166
## Number of obs: 3222, groups: participantCode, 179; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.33919 0.06834 4.963 6.93e-07 ***
## conditionC 0.38311 0.12467 3.073 0.00212 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.022confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.2052465 0.4731307
## conditionC 0.1387632 0.6274558Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp3"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data: subset(d, exp == "exp3" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 4292.1 4328.5 -2140.0 4280.1 3216
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1229 -1.0235 0.4378 0.9032 1.6613
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 0.43693 0.66101
## stimulus (Intercept) 0.01427 0.11945
## conditionC 0.00701 0.08373 -1.00
## Number of obs: 3222, groups: participantCode, 179; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.33901 0.06862 4.941 7.78e-07 ***
## conditionC 0.38201 0.12623 3.026 0.00248 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC -0.043
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.2045242 0.4734924
## conditionC 0.1345932 0.6294225Testing against chance: Learnable Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="exp3"&testType=="novelX"&conditionC==0.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data: subset(d, exp == "exp3" & testType == "novelX" & conditionC ==
## 0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2046.6 2057.4 -1021.3 2042.6 1618
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.6331 -0.9616 0.3150 0.8649 1.3816
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 1.051 1.025
## Number of obs: 1620, groups: participantCode, 90
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.6025 0.1246 4.834 1.34e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.3582096 0.846745Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution (though the parameter estimates and test statistics are highly similar to the converging model).
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp3"&testType=="novelX"&conditionC==0.5),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp3" & testType == "novelX" & conditionC ==
## 0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2048.6 2064.8 -1021.3 2042.6 1617
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.6332 -0.9616 0.3150 0.8649 1.3816
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 1.051 1.025
## stimulus (Intercept) 0.000 0.000
## Number of obs: 1620, groups: participantCode, 90; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.6025 0.1246 4.834 1.34e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.3582252 0.8468066Testing against chance: Unstructured Pre-Exposure
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge, as reported in the manuscript.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp3"&testType=="novelX"&conditionC==-0.5),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "exp3" & testType == "novelX" & conditionC ==
## -0.5)
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2214.6 2230.8 -1104.3 2208.6 1599
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.3338 -1.0401 0.7710 0.9299 1.1929
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.08077 0.2842
## stimulus (Intercept) 0.01810 0.1346
## Number of obs: 1602, groups: participantCode, 89; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.13604 0.06698 2.031 0.0422 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.004761652 0.2673123Interaction between test type and condition
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#recode test type
d$testTypeC <- ifelse(d$testType=="novelX",0.5,
ifelse(d$testType=="familiarX",-0.5,NA))
##all data
m <- glmer(isRight~conditionC*testTypeC+(1|participantCode)+(1|stimulus),data=subset(d,exp=="exp3"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC * testTypeC + (1 | participantCode) + (1 |
## stimulus)
## Data: subset(d, exp == "exp3")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 8423.9 8464.5 -4205.9 8411.9 6438
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.0446 -1.0163 0.3558 0.9145 1.4950
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.60076 0.77509
## stimulus (Intercept) 0.00426 0.06527
## Number of obs: 6444, groups: participantCode, 179; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.38483 0.06545 5.880 4.11e-09 ***
## conditionC 0.42539 0.12854 3.309 0.000935 ***
## testTypeC -0.02990 0.05730 -0.522 0.601772
## conditionC:testTypeC -0.01681 0.10602 -0.159 0.874005
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC tstTyC
## conditionC 0.019
## testTypeC -0.001 -0.001
## cndtnC:tsTC -0.001 -0.001 0.041confint(m, method="Wald")[3:6,]## 2.5 % 97.5 %
## (Intercept) 0.2565559 0.51311242
## conditionC 0.1734545 0.67732689
## testTypeC -0.1422117 0.08240574
## conditionC:testTypeC -0.2246079 0.19098383Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
##all data
m <- glmer(isRight~conditionC*testTypeC+(1+testTypeC|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="exp3"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula:
## isRight ~ conditionC * testTypeC + (1 + testTypeC | participantCode) +
## (1 + conditionC | stimulus)
## Data: subset(d, exp == "exp3")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 8431.0 8498.8 -4205.5 8411.0 6434
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.0524 -1.0177 0.3537 0.9142 1.5229
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 6.011e-01 0.775307
## testTypeC 9.903e-05 0.009951 -1.00
## stimulus (Intercept) 5.873e-03 0.076637
## conditionC 5.111e-03 0.071495 -1.00
## Number of obs: 6444, groups: participantCode, 179; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.38486 0.06581 5.848 4.97e-09 ***
## conditionC 0.42483 0.12913 3.290 0.001 **
## testTypeC -0.03192 0.06093 -0.524 0.600
## conditionC:testTypeC -0.01920 0.11031 -0.174 0.862
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cndtnC tstTyC
## conditionC 0.001
## testTypeC -0.013 0.000
## cndtnC:tsTC -0.001 -0.013 -0.010
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[7:10,]## 2.5 % 97.5 %
## (Intercept) 0.2558836 0.51384564
## conditionC 0.1717295 0.67792630
## testTypeC -0.1513293 0.08749837
## conditionC:testTypeC -0.2354027 0.19701156Signal Detection Model
Sensitivity
Overall
#recode condition
subj_overall$conditionC <- ifelse(subj_overall$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_overall$condition=="Unstructured Pre-Exposure",-0.5,NA))
##all data
m <- lm(dprime~conditionC,data=subset(subj_overall,exp=="exp3"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_overall,
## exp == "exp3"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8135 -0.6835 -0.2402 0.1862 3.5888
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.51673 0.08566 6.032 9.21e-09 ***
## conditionC 0.55311 0.17132 3.229 0.00148 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.146 on 177 degrees of freedom
## Multiple R-squared: 0.05561, Adjusted R-squared: 0.05028
## F-statistic: 10.42 on 1 and 177 DF, p-value: 0.001483Testing against chance: Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp3"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp3" &
## conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8135 -0.7933 -0.6345 0.1980 3.0357
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.7933 0.1450 5.471 4.07e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.376 on 89 degrees of freedomTesting against chance: Unstructured Pre-Exposure
m <- lm(dprime~1,data=subset(subj_overall,exp=="exp3"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "exp3" &
## conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.4356 -0.3827 -0.1005 0.1818 3.5888
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.24017 0.09045 2.655 0.0094 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.8533 on 88 degrees of freedomFamiliar X Trials
Condition Effect
#recode condition
subj_testType$conditionC <- ifelse(subj_testType$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_testType$condition=="Unstructured Pre-Exposure",-0.5,NA))
m <- lm(dprime~conditionC,data=subset(subj_testType,exp=="exp3"&testType=="familiarX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## exp == "exp3" & testType == "familiarX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.9315 -0.7360 -0.2373 0.3877 2.9491
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.48667 0.08135 5.983 1.19e-08 ***
## conditionC 0.49870 0.16269 3.065 0.00252 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.088 on 177 degrees of freedom
## Multiple R-squared: 0.05041, Adjusted R-squared: 0.04504
## F-statistic: 9.396 on 1 and 177 DF, p-value: 0.002515Testing against chance: Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp3"&testType=="familiarX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp3" & testType == "familiarX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.9315 -0.7360 -0.3218 0.3867 2.4504
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.736 0.128 5.751 1.23e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.214 on 89 degrees of freedomTesting against chance: Unstructured Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp3"&testType=="familiarX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp3" & testType == "familiarX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.4328 -0.5713 -0.2373 0.3877 2.9491
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.2373 0.1001 2.371 0.0199 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.9442 on 88 degrees of freedomNovel X Trials
Condition Effect
m <- lm(dprime~conditionC,data=subset(subj_testType,exp=="exp3"&testType=="novelX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## exp == "exp3" & testType == "novelX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.1854 -0.6917 -0.2001 0.3704 2.9863
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.44589 0.08133 5.483 1.43e-07 ***
## conditionC 0.49162 0.16265 3.023 0.00288 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.088 on 177 degrees of freedom
## Multiple R-squared: 0.04908, Adjusted R-squared: 0.04371
## F-statistic: 9.136 on 1 and 177 DF, p-value: 0.002879Testing against chance: Learnable Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp3"&testType=="novelX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp3" & testType == "novelX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.0520 -0.9798 -0.5520 0.5037 2.4947
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.6917 0.1375 5.032 2.51e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.304 on 89 degrees of freedomTesting against chance: Unstructured Pre-Exposure
m <- lm(dprime~1,data=subset(subj_testType,exp=="exp3"&testType=="novelX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "exp3" & testType == "novelX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.1854 -0.4911 -0.2001 0.2559 2.9863
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.2001 0.0862 2.321 0.0226 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.8132 on 88 degrees of freedomInteraction between Novel X and Familiar X trials
#recode test type
subj_testType$testTypeC <- ifelse(subj_testType$testType=="novelX",0.5,
ifelse(subj_testType$testType=="familiarX",-0.5,NA))
m <- lm(dprime~conditionC*testTypeC,data=subset(subj_testType,exp=="exp3"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC * testTypeC, data = subset(subj_testType,
## exp == "exp3"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.1854 -0.7360 -0.2001 0.3704 2.9863
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.466283 0.057513 8.107 8.54e-15 ***
## conditionC 0.495159 0.115027 4.305 2.17e-05 ***
## testTypeC -0.040781 0.115027 -0.355 0.723
## conditionC:testTypeC -0.007077 0.230053 -0.031 0.975
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.088 on 354 degrees of freedom
## Multiple R-squared: 0.05007, Adjusted R-squared: 0.04202
## F-statistic: 6.219 on 3 and 354 DF, p-value: 0.0003985Response Bias
Overall
#recode condition
subj_overall$conditionC <- ifelse(subj_overall$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_overall$condition=="Unstructured Pre-Exposure",-0.5,NA))
##all data
m <- lm(c~conditionC,data=subset(subj_overall,exp=="exp3"))
summary(m)##
## Call:
## lm(formula = c ~ conditionC, data = subset(subj_overall, exp ==
## "exp3"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.7673 -0.1365 0.0241 0.2215 0.9152
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.15550 0.02925 -5.316 3.17e-07 ***
## conditionC 0.01662 0.05850 0.284 0.777
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3914 on 177 degrees of freedom
## Multiple R-squared: 0.0004559, Adjusted R-squared: -0.005191
## F-statistic: 0.08074 on 1 and 177 DF, p-value: 0.7766Testing against chance: Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_overall,exp=="exp3"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp3" &
## conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.76732 -0.13803 0.07258 0.21705 0.74491
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.14719 0.03944 -3.732 0.000334 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3741 on 89 degrees of freedomTesting against chance: Unstructured Pre-Exposure
m <- lm(c~1,data=subset(subj_overall,exp=="exp3"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_overall, exp == "exp3" &
## conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.59005 -0.13091 0.02271 0.23507 0.91524
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.16381 0.04325 -3.787 0.000278 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4081 on 88 degrees of freedomFamiliar X Trials
Condition Effect
#recode condition
subj_testType$conditionC <- ifelse(subj_testType$condition=="Learnable Pre-Exposure",0.5,
ifelse(subj_testType$condition=="Unstructured Pre-Exposure",-0.5,NA))
m <- lm(c~conditionC,data=subset(subj_testType,exp=="exp3"&testType=="familiarX"))
summary(m)##
## Call:
## lm(formula = c ~ conditionC, data = subset(subj_testType, exp ==
## "exp3" & testType == "familiarX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.23566 -0.22471 0.07234 0.35756 0.78829
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.40651 0.03476 -11.696 <2e-16 ***
## conditionC 0.09791 0.06951 1.409 0.161
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.465 on 177 degrees of freedom
## Multiple R-squared: 0.01108, Adjusted R-squared: 0.005498
## F-statistic: 1.984 on 1 and 177 DF, p-value: 0.1607Testing against chance: Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp3"&testType=="familiarX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp3" &
## testType == "familiarX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.23566 -0.19143 0.07234 0.35756 0.78829
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.35756 0.04803 -7.444 5.92e-11 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4557 on 89 degrees of freedomTesting against chance: Unstructured Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp3"&testType=="familiarX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp3" &
## testType == "familiarX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.1378 -0.2247 0.1430 0.3158 0.6834
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.45547 0.05027 -9.061 3.07e-14 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4742 on 88 degrees of freedomNovel X Trials
Condition Effect
m <- lm(c~conditionC,data=subset(subj_testType,exp=="exp3"&testType=="novelX"))
summary(m)##
## Call:
## lm(formula = c ~ conditionC, data = subset(subj_testType, exp ==
## "exp3" & testType == "novelX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.65861 -0.29336 -0.06539 0.21983 1.52783
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.09942 0.04583 2.169 0.0314 *
## conditionC -0.06806 0.09167 -0.742 0.4588
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6132 on 177 degrees of freedom
## Multiple R-squared: 0.003105, Adjusted R-squared: -0.002527
## F-statistic: 0.5513 on 1 and 177 DF, p-value: 0.4588Testing against chance: Learnable Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp3"&testType=="novelX"&conditionC==0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp3" &
## testType == "novelX" & conditionC == 0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.65861 -0.23238 -0.06539 0.21983 1.52783
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.06539 0.05611 1.165 0.247
##
## Residual standard error: 0.5323 on 89 degrees of freedomTesting against chance: Unstructured Pre-Exposure
m <- lm(c~1,data=subset(subj_testType,exp=="exp3"&testType=="novelX"&conditionC==-0.5))
summary(m)##
## Call:
## lm(formula = c ~ 1, data = subset(subj_testType, exp == "exp3" &
## testType == "novelX" & conditionC == -0.5))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.5404 -0.4187 -0.1335 0.1518 1.4598
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.13345 0.07265 1.837 0.0696 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6854 on 88 degrees of freedomInteraction between Novel X and Familiar X trials
m <- lm(c~conditionC*testTypeC,data=subset(subj_testType,exp=="exp3"))
summary(m)##
## Call:
## lm(formula = c ~ conditionC * testTypeC, data = subset(subj_testType,
## exp == "exp3"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.65861 -0.27316 0.00626 0.31431 1.52783
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.15355 0.02876 -5.339 1.68e-07 ***
## conditionC 0.01492 0.05752 0.259 0.795
## testTypeC 0.50594 0.05752 8.796 < 2e-16 ***
## conditionC:testTypeC -0.16597 0.11504 -1.443 0.150
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5442 on 354 degrees of freedom
## Multiple R-squared: 0.1832, Adjusted R-squared: 0.1762
## F-statistic: 26.46 on 3 and 354 DF, p-value: 1.833e-15Much larger response bias for familiar X trials compared to novel X trials.
Overall Analysis: Experiments 1 - 3
Plot all data from all conditions across Experiments 1 - 3
Plot
#Familiar X test
p1 <- ggplot(subset(test_type,testType=="familiarX"),aes(condition,accuracy,fill=condition, color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="familiarX"),aes(y=acc), width = 0.05, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, height=0.02, size=0.8)+
theme_classic(base_size=11)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Unstructured Pre-Exposure","Learnable Pre-Exposure"),
labels=c("Non-Learnable\nPre-Exposure","No\n Pre-Exposure","Unstructured\nPre-Exposure","Learnable\nPre-Exposure"),
limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Unstructured Pre-Exposure","Learnable Pre-Exposure"))+
scale_fill_manual(values=c("#E41A1C","#377EB8","#54278f","#4DAF4A"),limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Unstructured Pre-Exposure","Learnable Pre-Exposure"))+
scale_color_manual(values=c("#E41A1C","#377EB8","#54278f","#4DAF4A"),limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Unstructured Pre-Exposure","Learnable Pre-Exposure"))+
theme(legend.position="none")+
ylab("Accuracy - Familiar X Test")+
xlab("Condition")
#Novel X
p2 <- ggplot(subset(test_type,testType=="novelX"),aes(condition,accuracy,fill=condition,color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="novelX"),aes(y=acc), width = 0.05, height=0.02, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, size=0.8)+
theme_classic(base_size=11)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Unstructured Pre-Exposure","Learnable Pre-Exposure"),
labels=c("Non-Learnable\nPre-Exposure","No\n Pre-Exposure","Unstructured\nPre-Exposure","Learnable\nPre-Exposure"),
limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Unstructured Pre-Exposure","Learnable Pre-Exposure"))+
scale_fill_manual(values=c("#E41A1C","#377EB8","#54278f","#4DAF4A"),limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Unstructured Pre-Exposure","Learnable Pre-Exposure"))+
scale_color_manual(values=c("#E41A1C","#377EB8","#54278f","#4DAF4A"),limits=c("Non-Learnable Pre-Exposure","No Pre-Exposure","Unstructured Pre-Exposure","Learnable Pre-Exposure"))+
theme(legend.position="none")+
ylab("Accuracy - Novel X Test")+
xlab("Condition")
plot_grid(p1,p2, labels=c("A","B"),label_size=18)
Overall Accuracy Table
overall %>%
select(-c(se, accuracy_ci,d_prime_ci,c_bias_ci)) %>%
formattable()| condition | N | accuracy | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|
| Learnable Pre-Exposure | 205 | 0.5987566 | 0.5738617 | 0.6236514 | 0.6815562 | 0.50340886 | 0.8597035 | -0.1705878 | -0.2199460 | -0.12122966 |
| Learnable Pre-Exposure Only | 48 | 0.6464120 | 0.5870300 | 0.7057941 | 1.0119951 | 0.57692063 | 1.4470695 | -0.2476500 | -0.3558765 | -0.13942359 |
| No Pre-Exposure | 81 | 0.5445816 | 0.5165071 | 0.5726561 | 0.2917437 | 0.09294647 | 0.4905409 | -0.2817981 | -0.3880251 | -0.17557112 |
| Non-Learnable Pre-Exposure | 114 | 0.5268031 | 0.5088965 | 0.5447098 | 0.1744142 | 0.05828405 | 0.2905444 | -0.3033881 | -0.3800089 | -0.22676727 |
| Unstructured Pre-Exposure | 89 | 0.5355805 | 0.5085204 | 0.5626406 | 0.2401722 | 0.06042550 | 0.4199189 | -0.1638137 | -0.2497717 | -0.07785573 |
Accuracy Table By Test Type
test_type %>%
select(-c(se, accuracy_ci,d_prime_ci,c_bias_ci)) %>%
formattable()| condition | testType | N | accuracy | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Learnable Pre-Exposure | familiarX | 205 | 0.6032520 | 0.5769629 | 0.6295412 | 0.6375685 | 0.474275807 | 0.8008612 | -0.36106697 | -0.42235755 | -0.299776379 |
| Learnable Pre-Exposure | novelX | 205 | 0.5942073 | 0.5668629 | 0.6215517 | 0.5840781 | 0.414323524 | 0.7538328 | 0.01628288 | -0.05174899 | 0.084314749 |
| Learnable Pre-Exposure Only | familiarX | 48 | 0.6412037 | 0.5770248 | 0.7053826 | 0.8763520 | 0.476210917 | 1.2764931 | -0.37304689 | -0.49147685 | -0.254616937 |
| Learnable Pre-Exposure Only | novelX | 48 | 0.6516204 | 0.5906462 | 0.7125946 | 0.9315712 | 0.553606822 | 1.3095356 | -0.11367134 | -0.22191307 | -0.005429604 |
| No Pre-Exposure | familiarX | 81 | 0.5480110 | 0.5164667 | 0.5795553 | 0.2995076 | 0.101032689 | 0.4979825 | -0.64553887 | -0.75491277 | -0.536164973 |
| No Pre-Exposure | novelX | 81 | 0.5411523 | 0.5100279 | 0.5722766 | 0.2466637 | 0.058929824 | 0.4343976 | 0.08656619 | -0.07935325 | 0.252485632 |
| Non-Learnable Pre-Exposure | familiarX | 114 | 0.5253411 | 0.5028839 | 0.5477983 | 0.1448369 | 0.008158158 | 0.2815156 | -0.54211863 | -0.62796874 | -0.456268514 |
| Non-Learnable Pre-Exposure | novelX | 114 | 0.5282651 | 0.5050216 | 0.5515086 | 0.1731961 | 0.038864646 | 0.3075276 | -0.06726700 | -0.16750917 | 0.032975169 |
| Unstructured Pre-Exposure | familiarX | 89 | 0.5380774 | 0.5053266 | 0.5708282 | 0.2373254 | 0.038431089 | 0.4362197 | -0.45546991 | -0.55536408 | -0.355575746 |
| Unstructured Pre-Exposure | novelX | 89 | 0.5330836 | 0.5043785 | 0.5617888 | 0.2000827 | 0.028781716 | 0.3713837 | 0.13345328 | -0.01092315 | 0.277829719 |
Overall - All pre-exposure conditions
All Test Trials
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
Condition dummy-coded - Learnable Pre-Exposure condition is the reference level.
m <- glmer(isRight~condition+(1|participantCode),data=filter(d, condition!="Learnable Pre-Exposure Only"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ condition + (1 | participantCode)
## Data: filter(d, condition != "Learnable Pre-Exposure Only")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 23491.8 23530.6 -11740.9 23481.8 17597
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5857 -1.0185 0.4159 0.9080 1.4127
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.3866 0.6218
## Number of obs: 17602, groups: participantCode, 489
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.47568 0.05065 9.392 < 2e-16
## conditionNo Pre-Exposure -0.27030 0.09387 -2.880 0.003983
## conditionNon-Learnable Pre-Exposure -0.35704 0.08351 -4.276 1.91e-05
## conditionUnstructured Pre-Exposure -0.31147 0.09075 -3.432 0.000599
##
## (Intercept) ***
## conditionNo Pre-Exposure **
## conditionNon-Learnable Pre-Exposure ***
## conditionUnstructured Pre-Exposure ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cnNP-E cN-LP-
## cndtnNPr-Ex -0.537
## cndtnN-LP-E -0.605 0.325
## cndtnUnPr-E -0.556 0.299 0.337Anova(m,type="III")## Analysis of Deviance Table (Type III Wald chisquare tests)
##
## Response: isRight
## Chisq Df Pr(>Chisq)
## (Intercept) 88.203 1 < 2.2e-16 ***
## condition 24.080 3 2.403e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
##all data
#does not converge
m <- glmer(isRight~condition+(1|participantCode)+(1+condition|stimulus),data=filter(d, condition!="Learnable Pre-Exposure Only"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ condition + (1 | participantCode) + (1 + condition |
## stimulus)
## Data: filter(d, condition != "Learnable Pre-Exposure Only")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 23510.7 23627.3 -11740.3 23480.7 17587
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5927 -1.0172 0.4152 0.9123 1.4523
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 3.871e-01 0.622135
## stimulus (Intercept) 6.622e-05 0.008138
## conditionNo Pre-Exposure 7.993e-05 0.008940
## conditionNon-Learnable Pre-Exposure 3.363e-04 0.018338
## conditionUnstructured Pre-Exposure 9.815e-03 0.099071
## Corr
##
##
## 1.00
## 1.00 1.00
## 1.00 1.00 1.00
## Number of obs: 17602, groups: participantCode, 489; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.47575 0.05070 9.383 < 2e-16
## conditionNo Pre-Exposure -0.27033 0.09392 -2.878 0.003996
## conditionNon-Learnable Pre-Exposure -0.35708 0.08361 -4.271 1.95e-05
## conditionUnstructured Pre-Exposure -0.31115 0.09234 -3.370 0.000752
##
## (Intercept) ***
## conditionNo Pre-Exposure **
## conditionNon-Learnable Pre-Exposure ***
## conditionUnstructured Pre-Exposure ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cnNP-E cN-LP-
## cndtnNPr-Ex -0.537
## cndtnN-LP-E -0.604 0.326
## cndtnUnPr-E -0.542 0.297 0.338
## convergence code: 0
## boundary (singular) fit: see ?isSingularAnova(m,type="III")## Analysis of Deviance Table (Type III Wald chisquare tests)
##
## Response: isRight
## Chisq Df Pr(>Chisq)
## (Intercept) 88.040 1 < 2.2e-16 ***
## condition 23.799 3 2.752e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Familiar X
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
Condition dummy-coded - Learnable Pre-Exposure condition is the reference level.
m <- glmer(isRight~condition+(1|participantCode)+(1|stimulus),data=filter(d, testType=="familiarX"&condition!="Learnable Pre-Exposure Only"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ condition + (1 | participantCode) + (1 | stimulus)
## Data:
## filter(d, testType == "familiarX" & condition != "Learnable Pre-Exposure Only")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11875.7 11918.2 -5931.8 11863.7 8796
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.8964 -1.0379 0.5335 0.9006 1.4035
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2762752 0.52562
## stimulus (Intercept) 0.0007265 0.02695
## Number of obs: 8802, groups: participantCode, 489; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.45832 0.05136 8.924 < 2e-16
## conditionNo Pre-Exposure -0.25018 0.09445 -2.649 0.00808
## conditionNon-Learnable Pre-Exposure -0.34996 0.08405 -4.164 3.13e-05
## conditionUnstructured Pre-Exposure -0.29221 0.09141 -3.197 0.00139
##
## (Intercept) ***
## conditionNo Pre-Exposure **
## conditionNon-Learnable Pre-Exposure ***
## conditionUnstructured Pre-Exposure **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cnNP-E cN-LP-
## cndtnNPr-Ex -0.533
## cndtnN-LP-E -0.601 0.326
## cndtnUnPr-E -0.551 0.299 0.337Anova(m,type="III")## Analysis of Deviance Table (Type III Wald chisquare tests)
##
## Response: isRight
## Chisq Df Pr(>Chisq)
## (Intercept) 79.644 1 < 2.2e-16 ***
## condition 21.940 3 6.714e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
##all data
#does not converge
m <- glmer(isRight~condition+(1|participantCode)+(1+condition|stimulus),data=filter(d, testType=="familiarX"&condition!="Learnable Pre-Exposure Only"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ condition + (1 | participantCode) + (1 + condition |
## stimulus)
## Data:
## filter(d, testType == "familiarX" & condition != "Learnable Pre-Exposure Only")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11893.3 11999.5 -5931.6 11863.3 8787
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.8953 -1.0359 0.5368 0.9006 1.4034
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2765631 0.52589
## stimulus (Intercept) 0.0002882 0.01698
## conditionNo Pre-Exposure 0.0028056 0.05297
## conditionNon-Learnable Pre-Exposure 0.0002372 0.01540
## conditionUnstructured Pre-Exposure 0.0033356 0.05776
## Corr
##
##
## 1.00
## 1.00 1.00
## 1.00 1.00 1.00
## Number of obs: 8802, groups: participantCode, 489; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.45831 0.05113 8.963 < 2e-16
## conditionNo Pre-Exposure -0.24999 0.09532 -2.623 0.00872
## conditionNon-Learnable Pre-Exposure -0.34994 0.08415 -4.158 3.2e-05
## conditionUnstructured Pre-Exposure -0.29203 0.09245 -3.159 0.00159
##
## (Intercept) ***
## conditionNo Pre-Exposure **
## conditionNon-Learnable Pre-Exposure ***
## conditionUnstructured Pre-Exposure **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cnNP-E cN-LP-
## cndtnNPr-Ex -0.521
## cndtnN-LP-E -0.600 0.328
## cndtnUnPr-E -0.536 0.312 0.339
## convergence code: 0
## boundary (singular) fit: see ?isSingularAnova(m,type="III")## Analysis of Deviance Table (Type III Wald chisquare tests)
##
## Response: isRight
## Chisq Df Pr(>Chisq)
## (Intercept) 80.344 1 < 2.2e-16 ***
## condition 21.625 3 7.807e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Novel X
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
Condition dummy-coded - Learnable Pre-Exposure condition is the reference level.
m <- glmer(isRight~condition+(1|participantCode),data=filter(d, testType=="novelX"&condition!="Learnable Pre-Exposure Only"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ condition + (1 | participantCode)
## Data:
## filter(d, testType == "novelX" & condition != "Learnable Pre-Exposure Only")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11896.2 11931.7 -5943.1 11886.2 8795
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.8833 -1.0317 0.5301 0.9116 1.4208
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2853 0.5341
## Number of obs: 8800, groups: participantCode, 489
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.42193 0.05142 8.206 2.28e-16
## conditionNo Pre-Exposure -0.24356 0.09525 -2.557 0.010558
## conditionNon-Learnable Pre-Exposure -0.30163 0.08483 -3.556 0.000377
## conditionUnstructured Pre-Exposure -0.27919 0.09212 -3.031 0.002439
##
## (Intercept) ***
## conditionNo Pre-Exposure *
## conditionNon-Learnable Pre-Exposure ***
## conditionUnstructured Pre-Exposure **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cnNP-E cN-LP-
## cndtnNPr-Ex -0.538
## cndtnN-LP-E -0.605 0.326
## cndtnUnPr-E -0.557 0.300 0.337Anova(m,type="III")## Analysis of Deviance Table (Type III Wald chisquare tests)
##
## Response: isRight
## Chisq Df Pr(>Chisq)
## (Intercept) 67.341 1 2.284e-16 ***
## condition 17.616 3 0.0005278 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
##all data
#does not converge
m <- glmer(isRight~condition+(1|participantCode)+(1+condition|stimulus),data=filter(d, testType=="novelX"&condition!="Learnable Pre-Exposure Only"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ condition + (1 | participantCode) + (1 + condition |
## stimulus)
## Data:
## filter(d, testType == "novelX" & condition != "Learnable Pre-Exposure Only")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 11914.8 12021.0 -5942.4 11884.8 8785
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.8845 -1.0302 0.5306 0.9117 1.5052
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.2860806 0.53487
## stimulus (Intercept) 0.0000000 0.00000
## conditionNo Pre-Exposure 0.0001722 0.01312
## conditionNon-Learnable Pre-Exposure 0.0002414 0.01554
## conditionUnstructured Pre-Exposure 0.0219312 0.14809
## Corr
##
##
## NaN
## NaN -1.00
## NaN -1.00 1.00
## Number of obs: 8800, groups: participantCode, 489; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.42204 0.05145 8.202 2.36e-16
## conditionNo Pre-Exposure -0.24363 0.09538 -2.554 0.010639
## conditionNon-Learnable Pre-Exposure -0.30171 0.08497 -3.551 0.000384
## conditionUnstructured Pre-Exposure -0.27862 0.09864 -2.825 0.004734
##
## (Intercept) ***
## conditionNo Pre-Exposure *
## conditionNon-Learnable Pre-Exposure ***
## conditionUnstructured Pre-Exposure **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) cnNP-E cN-LP-
## cndtnNPr-Ex -0.538
## cndtnN-LP-E -0.605 0.324
## cndtnUnPr-E -0.520 0.269 0.330
## convergence code: 0
## boundary (singular) fit: see ?isSingularAnova(m,type="III")## Analysis of Deviance Table (Type III Wald chisquare tests)
##
## Response: isRight
## Chisq Df Pr(>Chisq)
## (Intercept) 67.275 1 2.362e-16 ***
## condition 17.082 3 0.0006797 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1d-prime
All Test Trials
Condition dummy-coded - Learnable Pre-Exposure condition is the reference level.
m <- lm(dprime~condition,data=filter(subj_overall, condition!="Learnable Pre-Exposure Only"))
summary(m)##
## Call:
## lm(formula = dprime ~ condition, data = filter(subj_overall,
## condition != "Learnable Pre-Exposure Only"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.0357 -0.5418 -0.2402 0.1580 3.6546
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.68156 0.07193 9.475 < 2e-16
## conditionNo Pre-Exposure -0.38981 0.13516 -2.884 0.004100
## conditionNon-Learnable Pre-Exposure -0.50714 0.12032 -4.215 2.98e-05
## conditionUnstructured Pre-Exposure -0.44138 0.13073 -3.376 0.000794
##
## (Intercept) ***
## conditionNo Pre-Exposure **
## conditionNon-Learnable Pre-Exposure ***
## conditionUnstructured Pre-Exposure ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.03 on 485 degrees of freedom
## Multiple R-squared: 0.04654, Adjusted R-squared: 0.04064
## F-statistic: 7.891 on 3 and 485 DF, p-value: 3.78e-05Anova(m, type="III")## Anova Table (Type III tests)
##
## Response: dprime
## Sum Sq Df F value Pr(>F)
## (Intercept) 95.23 1 89.7822 < 2.2e-16 ***
## condition 25.11 3 7.8913 3.78e-05 ***
## Residuals 514.41 485
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Familiar X Test Trials
Condition dummy-coded - Learnable Pre-Exposure condition is the reference level.
m <- lm(dprime~condition,data=filter(subj_testType, testType=="familiarX"&condition!="Learnable Pre-Exposure Only"))
summary(m)##
## Call:
## lm(formula = dprime ~ condition, data = filter(subj_testType,
## testType == "familiarX" & condition != "Learnable Pre-Exposure Only"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0789 -0.6376 -0.1448 0.3255 3.0416
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.63757 0.07028 9.072 < 2e-16
## conditionNo Pre-Exposure -0.33806 0.13206 -2.560 0.01077
## conditionNon-Learnable Pre-Exposure -0.49273 0.11756 -4.191 3.3e-05
## conditionUnstructured Pre-Exposure -0.40024 0.12773 -3.133 0.00183
##
## (Intercept) ***
## conditionNo Pre-Exposure *
## conditionNon-Learnable Pre-Exposure ***
## conditionUnstructured Pre-Exposure **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.006 on 485 degrees of freedom
## Multiple R-squared: 0.04315, Adjusted R-squared: 0.03724
## F-statistic: 7.291 on 3 and 485 DF, p-value: 8.625e-05Anova(m, type="III")## Anova Table (Type III tests)
##
## Response: dprime
## Sum Sq Df F value Pr(>F)
## (Intercept) 83.33 1 82.3021 < 2.2e-16 ***
## condition 22.15 3 7.2912 8.625e-05 ***
## Residuals 491.06 485
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Novel X Test Trials
Condition dummy-coded - Learnable Pre-Exposure condition is the reference level.
m <- lm(dprime~condition,data=filter(subj_testType, testType=="novelX"&condition!="Learnable Pre-Exposure Only"))
summary(m)##
## Call:
## lm(formula = dprime ~ condition, data = filter(subj_testType,
## testType == "novelX" & condition != "Learnable Pre-Exposure Only"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.3170 -0.5841 -0.2001 0.2827 3.0132
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.58408 0.06985 8.362 6.59e-16
## conditionNo Pre-Exposure -0.33741 0.13125 -2.571 0.010445
## conditionNon-Learnable Pre-Exposure -0.41088 0.11684 -3.517 0.000478
## conditionUnstructured Pre-Exposure -0.38400 0.12695 -3.025 0.002621
##
## (Intercept) ***
## conditionNo Pre-Exposure *
## conditionNon-Learnable Pre-Exposure ***
## conditionUnstructured Pre-Exposure **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1 on 485 degrees of freedom
## Multiple R-squared: 0.03498, Adjusted R-squared: 0.02901
## F-statistic: 5.861 on 3 and 485 DF, p-value: 0.0006165Anova(m, type="III")## Anova Table (Type III tests)
##
## Response: dprime
## Sum Sq Df F value Pr(>F)
## (Intercept) 69.94 1 69.9231 6.585e-16 ***
## condition 17.58 3 5.8606 0.0006165 ***
## Residuals 485.08 485
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1S1: Learnable Pre-Exposure Only
Test Accuracy
Plot
#Familiar X test
p1 <- ggplot(subset(test_type_exp,testType=="familiarX"&exp=="s1"),aes(condition,accuracy,fill=condition, color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="familiarX"&exp=="s1"),aes(y=acc), width = 0.05, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, size=0.8)+
theme_classic(base_size=18)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Learnable Pre-Exposure Only"),
labels=c("Learnable\nPre-Exposure\nOnly"))+
scale_fill_manual(values=c("#265725"))+
scale_color_manual(values=c("#265725"))+
theme(legend.position="none")+
ylab("Accuracy - Familiar X test")+
xlab("Condition")
#Novel X
p2 <- ggplot(subset(test_type_exp,testType=="novelX"&exp=="s1"),aes(condition,accuracy,fill=condition, color=condition))+
geom_bar(position=position_dodge(.9), stat="identity", size=1.2,alpha=0.3, width=0.7)+
geom_jitter(data=subset(subj_testType,testType=="novelX"&exp=="s1"),aes(y=acc), width = 0.05, alpha=0.6,shape=21)+
geom_errorbar(aes(ymin=accuracy-se,ymax=accuracy+se),color="black",position=position_dodge(.9),width=0.05, size=0.8)+
theme_classic(base_size=18)+
geom_hline(yintercept=0.5, linetype="dotted")+
scale_x_discrete(name="Condition",
breaks=c("Learnable Pre-Exposure Only"),
labels=c("Learnable\nPre-Exposure\nOnly"))+
scale_fill_manual(values=c("#265725"))+
scale_color_manual(values=c("#265725"))+
theme(legend.position="none")+
ylab("Accuracy - Novel X Test")+
xlab("Condition")
plot_grid(p1,p2, labels=c("A","B"),label_size=20)
Overall Accuracy Table
overall_exp %>%
filter(exp=="s1") %>%
#select(-c(accuracy_ci,d_prime_ci,c_bias,c_bias_ci,c_bias_lower_ci,c_bias_upper_ci)) %>%
formattable()| exp | condition | N | accuracy | se | accuracy_ci | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_ci | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_ci | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| s1 | Learnable Pre-Exposure Only | 48 | 0.646412 | 0.02951775 | 0.05938206 | 0.58703 | 0.7057941 | 1.011995 | 0.4350744 | 0.5769206 | 1.447069 | -0.24765 | 0.1082264 | -0.3558765 | -0.1394236 |
Accuracy Table By Test Type
test_type_exp %>%
filter(exp=="s1") %>%
#select(-c(accuracy_ci,d_prime_ci,c_bias,c_bias_ci,c_bias_lower_ci,c_bias_upper_ci)) %>%
formattable()| exp | condition | testType | N | accuracy | se | accuracy_ci | accuracy_lower_ci | accuracy_upper_ci | d_prime | d_prime_ci | d_prime_lower_ci | d_prime_upper_ci | c_bias | c_bias_ci | c_bias_lower_ci | c_bias_upper_ci |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| s1 | Learnable Pre-Exposure Only | familiarX | 48 | 0.6412037 | 0.03190216 | 0.06417887 | 0.5770248 | 0.7053826 | 0.8763520 | 0.4001411 | 0.4762109 | 1.276493 | -0.3730469 | 0.1184300 | -0.4914768 | -0.254616937 |
| s1 | Learnable Pre-Exposure Only | novelX | 48 | 0.6516204 | 0.03030918 | 0.06097421 | 0.5906462 | 0.7125946 | 0.9315712 | 0.3779644 | 0.5536068 | 1.309536 | -0.1136713 | 0.1082417 | -0.2219131 | -0.005429604 |
Correlation
Correlation between Familiar X and Novel X Trials
##Correlations between Familiar X and Novel X
c <- corr.test(subset(subj_accuracy_wide, exp=="s1")[,c("novelX","familiarX")])
c## Call:corr.test(x = subset(subj_accuracy_wide, exp == "s1")[, c("novelX",
## "familiarX")])
## Correlation matrix
## novelX familiarX
## novelX 1.0 0.8
## familiarX 0.8 1.0
## Sample Size
## [1] 48
## Probability values (Entries above the diagonal are adjusted for multiple tests.)
## novelX familiarX
## novelX 0 0
## familiarX 0 0
##
## To see confidence intervals of the correlations, print with the short=FALSE optionc$p## novelX familiarX
## novelX 0.000000e+00 8.227912e-12
## familiarX 8.227912e-12 0.000000e+00Correlation Plot
ggplot(subset(subj_accuracy_wide,exp=="s1"),aes(familiarX,novelX, color=condition,linetype=condition,shape=condition))+
geom_jitter(width=0.02,height=0.02,size=3)+
geom_smooth(method=lm,se =F,size=1.5)+
scale_color_manual(name="Condition",values=c("#265725"))+
scale_linetype_manual(name="Condition",values=c(1))+
scale_shape_manual(name="Condition",values=c(16))+
ylab("Accuracy - Novel X")+
xlab("Accuracy - Familiar X")+
theme_classic(base_size=18)+
theme(legend.position=c(0.25,0.9))
Logistic Mixed-Effects Model
Overall
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="s1"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data: subset(d, exp == "s1")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2025.6 2036.5 -1010.8 2021.6 1726
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.0537 -0.9745 0.2467 0.8765 1.2702
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 1.782 1.335
## Number of obs: 1728, groups: participantCode, 48
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.9017 0.2065 4.367 1.26e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.4970538 1.306441Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="s1"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "s1")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 2027.6 2044.0 -1010.8 2021.6 1725
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.0537 -0.9745 0.2467 0.8765 1.2702
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 1.782 1.335
## stimulus (Intercept) 0.000 0.000
## Number of obs: 1728, groups: participantCode, 48; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.9017 0.2065 4.367 1.26e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.497068 1.306398Familiar X Trials
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="s1"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data: subset(d, exp == "s1" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1045.4 1054.9 -520.7 1041.4 862
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.8641 -0.9603 0.2777 0.7782 1.5632
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 1.373 1.172
## Number of obs: 864, groups: participantCode, 48
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.7916 0.1923 4.116 3.85e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.4146819 1.168537Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="s1"&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "s1" & testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1047.4 1061.7 -520.7 1041.4 861
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.8640 -0.9603 0.2777 0.7782 1.5632
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 1.373 1.172
## stimulus (Intercept) 0.000 0.000
## Number of obs: 864, groups: participantCode, 48; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.7916 0.1923 4.116 3.85e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.4146746 1.168538Novel X Trials
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
m <- glmer(isRight~1+(1|participantCode),data=subset(d,exp=="s1"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode)
## Data: subset(d, exp == "s1" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1048.9 1058.4 -522.4 1044.9 862
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.7843 -0.9724 0.3592 0.7737 1.2429
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 1.166 1.08
## Number of obs: 864, groups: participantCode, 48
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.8163 0.1797 4.543 5.56e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.4640702 1.168449Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~1+(1|participantCode)+(1|stimulus),data=subset(d,exp=="s1"&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ 1 + (1 | participantCode) + (1 | stimulus)
## Data: subset(d, exp == "s1" & testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 1050.9 1065.1 -522.4 1044.9 861
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.7843 -0.9724 0.3592 0.7737 1.2429
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 1.166e+00 1.080e+00
## stimulus (Intercept) 1.257e-76 1.121e-38
## Number of obs: 864, groups: participantCode, 48; stimulus, 18
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.8163 0.1797 4.543 5.56e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## .sig02 NA NA
## (Intercept) 0.4640675 1.168448Comparison to No Pre-Exposure condition from Experiment 2
All Test Trials
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#compare to No Pre-Exposure condition from Experiment 2
#recode condition
d$conditionC <- ifelse(d$condition=="Learnable Pre-Exposure Only",0.5,
ifelse(d$condition=="No Pre-Exposure",-0.5,NA))
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,exp=="s1"|(exp=="exp2"&condition=="No Pre-Exposure")),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data:
## subset(d, exp == "s1" | (exp == "exp2" & condition == "No Pre-Exposure"))
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 6036.3 6055.6 -3015.1 6030.3 4641
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.0895 -1.0167 0.3237 0.8949 1.3116
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.6149 0.7841
## Number of obs: 4644, groups: participantCode, 129
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.49378 0.08032 6.147 7.87e-10 ***
## conditionC 0.55590 0.15961 3.483 0.000496 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.287confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.3363528 0.6512145
## conditionC 0.2430805 0.8687295Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,exp=="s1"|(exp=="exp2"&condition=="No Pre-Exposure")),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data:
## subset(d, exp == "s1" | (exp == "exp2" & condition == "No Pre-Exposure"))
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 6042.3 6080.9 -3015.1 6030.3 4638
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.0895 -1.0167 0.3237 0.8949 1.3116
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 6.149e-01 7.841e-01
## stimulus (Intercept) 0.000e+00 0.000e+00
## conditionC 5.426e-13 7.366e-07 NaN
## Number of obs: 4644, groups: participantCode, 129; stimulus, 72
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.49378 0.08033 6.147 7.89e-10 ***
## conditionC 0.55590 0.15961 3.483 0.000496 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.287
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.3363456 0.6512215
## conditionC 0.2430712 0.8687369Familiar X Test Trials
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#compare to No Pre-Exposure condition from Experiment 2
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,(exp=="s1"|(exp=="exp2"&condition=="No Pre-Exposure"))&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data:
## subset(d, (exp == "s1" | (exp == "exp2" & condition == "No Pre-Exposure")) &
## testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3072.1 3089.3 -1533.0 3066.1 2319
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.2045 -1.0370 0.4558 0.8925 1.2954
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.4379 0.6617
## Number of obs: 2322, groups: participantCode, 129
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.44492 0.07719 5.764 8.2e-09 ***
## conditionC 0.45795 0.15337 2.986 0.00283 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.289confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.2936397 0.5962060
## conditionC 0.1573446 0.7585483Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,(exp=="s1"|(exp=="exp2"&condition=="No Pre-Exposure"))&testType=="familiarX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data:
## subset(d, (exp == "s1" | (exp == "exp2" & condition == "No Pre-Exposure")) &
## testType == "familiarX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3078.1 3112.6 -1533.0 3066.1 2316
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.2045 -1.0370 0.4558 0.8925 1.2954
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 4.379e-01 6.617e-01
## stimulus (Intercept) 0.000e+00 0.000e+00
## conditionC 5.111e-14 2.261e-07 NaN
## Number of obs: 2322, groups: participantCode, 129; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.44492 0.07719 5.764 8.2e-09 ***
## conditionC 0.45795 0.15337 2.986 0.00283 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.289
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.2936385 0.5962072
## conditionC 0.1573528 0.7585392Novel X Trials
Final Converging Model
The final model with the maximal random effects structure that still allowed the model to converge.
#compare to No Pre-Exposure condition from Experiment 2
m <- glmer(isRight~conditionC+(1|participantCode),data=subset(d,(exp=="s1"|(exp=="exp2"&condition=="No Pre-Exposure"))&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))
summary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode)
## Data:
## subset(d, (exp == "s1" | (exp == "exp2" & condition == "No Pre-Exposure")) &
## testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3077.6 3094.9 -1535.8 3071.6 2319
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1515 -1.0345 0.4827 0.9013 1.2818
##
## Random effects:
## Groups Name Variance Std.Dev.
## participantCode (Intercept) 0.3747 0.6122
## Number of obs: 2322, groups: participantCode, 129
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.44545 0.07344 6.066 1.31e-09 ***
## conditionC 0.52673 0.14605 3.606 0.00031 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.292confint(m, method="Wald")## 2.5 % 97.5 %
## .sig01 NA NA
## (Intercept) 0.3015137 0.5893828
## conditionC 0.2404672 0.8129891Maximal Random Effects Model
Note that this model does not converge (singular fit), thus the parameter estimates should be interpreted with caution.
m <- glmer(isRight~conditionC+(1|participantCode)+(1+conditionC|stimulus),data=subset(d,(exp=="s1"|(exp=="exp2"&condition=="No Pre-Exposure"))&testType=="novelX"),family=binomial,glmerControl(optimizer="bobyqa"))## boundary (singular) fit: see ?isSingularsummary(m)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ conditionC + (1 | participantCode) + (1 + conditionC |
## stimulus)
## Data:
## subset(d, (exp == "s1" | (exp == "exp2" & condition == "No Pre-Exposure")) &
## testType == "novelX")
## Control: glmerControl(optimizer = "bobyqa")
##
## AIC BIC logLik deviance df.resid
## 3083.6 3118.1 -1535.8 3071.6 2316
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.1515 -1.0345 0.4827 0.9013 1.2818
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## participantCode (Intercept) 3.747e-01 6.122e-01
## stimulus (Intercept) 0.000e+00 0.000e+00
## conditionC 9.772e-14 3.126e-07 NaN
## Number of obs: 2322, groups: participantCode, 129; stimulus, 36
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.44545 0.07344 6.066 1.31e-09 ***
## conditionC 0.52673 0.14606 3.606 0.000311 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## conditionC 0.292
## convergence code: 0
## boundary (singular) fit: see ?isSingularconfint(m, method="Wald")[5:6,]## 2.5 % 97.5 %
## (Intercept) 0.3015126 0.5893824
## conditionC 0.2404509 0.8130027Signal Detection Analysis
Overall
##all data
m <- lm(dprime~1,data=subset(subj_overall,exp=="s1"))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_overall, exp == "s1"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.7249 -1.0120 -0.6694 0.5681 2.8170
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 1.0120 0.2163 4.679 2.47e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.498 on 47 degrees of freedomFamiliar X Trials
m <- lm(dprime~1,data=subset(subj_testType,exp=="s1"&testType=="familiarX"))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "s1" & testType == "familiarX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.5277 -0.8763 -0.2514 0.8585 2.3101
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.8764 0.1989 4.406 6.06e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.378 on 47 degrees of freedomNovel X Trials
m <- lm(dprime~1,data=subset(subj_testType,exp=="s1"&testType=="novelX"))
summary(m)##
## Call:
## lm(formula = dprime ~ 1, data = subset(subj_testType, exp ==
## "s1" & testType == "novelX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8360 -0.9316 -0.3611 0.7118 2.2549
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.9316 0.1879 4.958 9.68e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.302 on 47 degrees of freedomComparison to No Pre-Exposure condition from Experiment 2
Overall
#recode condition
subj_overall$conditionC <- ifelse(subj_overall$condition=="Learnable Pre-Exposure Only",0.5,
ifelse(subj_overall$condition=="No Pre-Exposure",-0.5,NA))
m <- lm(dprime~conditionC,data=subset(subj_overall,exp=="s1"|(exp=="exp2"&condition=="No Pre-Exposure")))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_overall,
## exp == "s1" | (exp == "exp2" & condition == "No Pre-Exposure")))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.7249 -0.6257 -0.2917 0.1580 3.5373
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.6519 0.1054 6.183 7.93e-09 ***
## conditionC 0.7203 0.2109 3.416 0.000854 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.158 on 127 degrees of freedom
## Multiple R-squared: 0.08414, Adjusted R-squared: 0.07693
## F-statistic: 11.67 on 1 and 127 DF, p-value: 0.0008544Familiar X Trials
#recode condition
subj_testType$conditionC <- ifelse(subj_testType$condition=="Learnable Pre-Exposure Only",0.5,
ifelse(subj_testType$condition=="No Pre-Exposure",-0.5,NA))
m <- lm(dprime~conditionC,data=subset(subj_testType,(exp=="s1"|(exp=="exp2"&condition=="No Pre-Exposure"))&testType=="familiarX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## (exp == "s1" | (exp == "exp2" & condition == "No Pre-Exposure")) &
## testType == "familiarX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.5277 -0.7554 -0.2995 0.4840 2.8869
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.5879 0.1002 5.868 3.61e-08 ***
## conditionC 0.5768 0.2004 2.879 0.00469 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.1 on 127 degrees of freedom
## Multiple R-squared: 0.06125, Adjusted R-squared: 0.05386
## F-statistic: 8.286 on 1 and 127 DF, p-value: 0.004688Novel X Trials
m <- lm(dprime~conditionC,data=subset(subj_testType,(exp=="s1"|(exp=="exp2"&condition=="No Pre-Exposure"))&testType=="novelX"))
summary(m)##
## Call:
## lm(formula = dprime ~ conditionC, data = subset(subj_testType,
## (exp == "s1" | (exp == "exp2" & condition == "No Pre-Exposure")) &
## testType == "novelX"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8360 -0.6405 -0.2467 0.3238 2.9398
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.5891 0.0947 6.221 6.59e-09 ***
## conditionC 0.6849 0.1894 3.616 0.00043 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.04 on 127 degrees of freedom
## Multiple R-squared: 0.09336, Adjusted R-squared: 0.08622
## F-statistic: 13.08 on 1 and 127 DF, p-value: 0.0004296