TME2 analyses
Subject-level graph of overall accuracy and naming frequency
ggplot(tme_subj, aes(NamingFreq_Study_num, testAcc))+
geom_point()+
geom_smooth(method=lm)+
theme_classic()
What factors predict accuracy? (typicality, nameability, judgment, naming frequency during study)
accuracy_randomSlope <- glmer(isRight ~ kinds_c + avgFam_c + typ_c*judgment_c + simpson_c*NamingFreq_Study_c +
(1|category) + (typ_c*judgment_c|subjCode),
data=tme2_full, family=binomial)## Warning in checkConv(attr(opt, "derivs"), opt$par, ctrl =
## control$checkConv, : Model failed to converge with max|grad| = 0.00830015
## (tol = 0.001, component 1)accuracy_randomInt <- glmer(isRight ~ kinds_c+ avgFam_c + typ_c*judgment_c + simpson_c*NamingFreq_Study_c +
(1|category) + (1|subjCode), data=tme2_full, family=binomial)
anova(accuracy_randomSlope, accuracy_randomInt) #random slope model didn't converge but is a better fit...## Data: tme2_full
## Models:
## accuracy_randomInt: isRight ~ kinds_c + avgFam_c + typ_c * judgment_c + simpson_c *
## accuracy_randomInt: NamingFreq_Study_c + (1 | category) + (1 | subjCode)
## accuracy_randomSlope: isRight ~ kinds_c + avgFam_c + typ_c * judgment_c + simpson_c *
## accuracy_randomSlope: NamingFreq_Study_c + (1 | category) + (typ_c * judgment_c |
## accuracy_randomSlope: subjCode)
## Df AIC BIC logLik deviance Chisq Chi Df
## accuracy_randomInt 11 2432.8 2494.6 -1205.4 2410.8
## accuracy_randomSlope 20 2098.9 2211.1 -1029.4 2058.9 351.99 9
## Pr(>Chisq)
## accuracy_randomInt
## accuracy_randomSlope < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1summary(accuracy_randomInt)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ kinds_c + avgFam_c + typ_c * judgment_c + simpson_c *
## NamingFreq_Study_c + (1 | category) + (1 | subjCode)
## Data: tme2_full
##
## AIC BIC logLik deviance df.resid
## 2432.8 2494.6 -1205.4 2410.8 2014
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.4171 -0.8709 0.4271 0.6953 3.1317
##
## Random effects:
## Groups Name Variance Std.Dev.
## category (Intercept) 0.1575 0.3969
## subjCode (Intercept) 0.8379 0.9154
## Number of obs: 2025, groups: category, 45; subjCode, 45
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.59605 0.15785 3.776 0.000159 ***
## kinds_c 0.28680 0.12336 2.325 0.020073 *
## avgFam_c -0.20244 0.37410 -0.541 0.588408
## typ_c -0.01048 0.02070 -0.506 0.612745
## judgment_c 0.59903 0.10702 5.598 2.17e-08 ***
## simpson_c -0.35214 0.26208 -1.344 0.179063
## NamingFreq_Study_c -0.06906 0.12224 -0.565 0.572064
## typ_c:judgment_c 0.04128 0.04227 0.976 0.328832
## simpson_c:NamingFreq_Study_c -0.21845 0.12892 -1.694 0.090181 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) knds_c avgFm_ typ_c jdgmn_ smpsn_ NmF_S_ typ_:_
## kinds_c 0.010
## avgFam_c -0.002 -0.363
## typ_c 0.001 -0.005 -0.010
## judgment_c -0.003 0.001 0.015 -0.008
## simpson_c 0.001 0.182 -0.394 0.003 -0.001
## NmngFrq_St_ -0.005 -0.001 0.000 0.000 0.006 -0.003
## typ_c:jdgm_ 0.000 0.042 0.015 -0.032 0.004 -0.008 0.004
## smps_:NF_S_ -0.005 -0.003 0.000 0.000 -0.036 -0.009 -0.005 -0.027summary(accuracy_randomSlope)## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: isRight ~ kinds_c + avgFam_c + typ_c * judgment_c + simpson_c *
## NamingFreq_Study_c + (1 | category) + (typ_c * judgment_c |
## subjCode)
## Data: tme2_full
##
## AIC BIC logLik deviance df.resid
## 2098.9 2211.1 -1029.4 2058.9 2005
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.5545 -0.6264 0.2190 0.5116 2.7549
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## category (Intercept) 0.055007 0.23454
## subjCode (Intercept) 1.029561 1.01467
## typ_c 0.005629 0.07503 0.34
## judgment_c 8.256849 2.87347 0.01 0.38
## typ_c:judgment_c 0.063433 0.25186 -0.14 0.78 0.78
## Number of obs: 2025, groups: category, 45; subjCode, 45
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.95064 0.17311 5.492 3.99e-08 ***
## kinds_c 0.12690 0.10719 1.184 0.2365
## avgFam_c -0.21463 0.32467 -0.661 0.5086
## typ_c 0.01497 0.03085 0.485 0.6274
## judgment_c 0.78243 0.45510 1.719 0.0856 .
## simpson_c -0.25077 0.22656 -1.107 0.2683
## NamingFreq_Study_c -0.08261 0.14082 -0.587 0.5574
## typ_c:judgment_c 0.05578 0.06900 0.808 0.4188
## simpson_c:NamingFreq_Study_c -0.15931 0.14578 -1.093 0.2745
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) knds_c avgFm_ typ_c jdgmn_ smpsn_ NmF_S_ typ_:_
## kinds_c 0.006
## avgFam_c -0.004 -0.356
## typ_c 0.171 -0.001 -0.028
## judgment_c 0.023 -0.003 0.004 0.142
## simpson_c -0.001 0.171 -0.389 0.014 0.005
## NmngFrq_St_ 0.004 0.009 0.013 0.009 0.010 -0.013
## typ_c:jdgm_ -0.052 0.046 -0.004 0.258 0.440 0.005 -0.041
## smps_:NF_S_ -0.005 -0.013 0.002 -0.007 -0.005 -0.035 -0.005 -0.025
## convergence code: 0
## Model failed to converge with max|grad| = 0.00830015 (tol = 0.001, component 1)Findings from random-intercepts model:
Significant effect of kinds: accuracy increases as category members are more conceptually distinct
Significant effect of judgment: accuracy is higher for ‘remember’ judgments
Marginal interaction between nameability and naming frequency: people with higher naming frequency are more negatively affected by nameability
Simpson vs. naming frequency (raw)
ggplot(filter(tme2_full, NamingFreq_Study_num %in% c(1,2,3,4,5)), aes(x=simpson, y=isRight, color=as.factor(NamingFreq_Study_num)))+
geom_point()+
geom_smooth(method="glm", formula=y~x, alpha=.2)+
theme_classic()
Simpson vs. naming frequency (median split)
tme2_medsplit <- tme2_full %>%
mutate(naming_median=median(NamingFreq_Study_num)) %>%
mutate(naming_medSplit=ifelse(NamingFreq_Study_num>naming_median,1,0))
ggplot(tme2_medsplit, aes(x=simpson, y=isRight, color=as.factor(naming_medSplit)))+
geom_point()+
geom_smooth(method="glm", formula=y~x)+
theme_classic()
Graph subject-level interaction between typ, overall naming frequency, and judgment
Each person contributes 6 data points: their mean accuracy for typ = 2, 5, or 8, and judgment = remember or know 
How do naming frequency during study, nameability, typicality, and judgment affect typicality shift?
Absolute value (random slope model has singular fit)
abs_typ_shift_randomSlope <- lmer(abs_typ_adv_choice ~kinds_c + avgFam_c + judgment_c*typ_c + simpson_c*NamingFreq_Study_c +
(1|category) + (judgment_c*typ_c|subjCode),
data=filter(tme2_full, isRight==0))## boundary (singular) fit: see ?isSingular## Warning: Model failed to converge with 1 negative eigenvalue: -1.5e+02abs_typ_shift_randomInt <- lmer(abs_typ_adv_choice ~kinds_c + avgFam_c + judgment_c*typ_c + simpson_c*NamingFreq_Study_c +
(1|subjCode)+(1|category),
data=filter(tme2_full, isRight==0))
anova(abs_typ_shift_randomInt,abs_typ_shift_randomSlope) #random slope does not improve model fit and has singular fit## refitting model(s) with ML (instead of REML)## Data: filter(tme2_full, isRight == 0)
## Models:
## abs_typ_shift_randomInt: abs_typ_adv_choice ~ kinds_c + avgFam_c + judgment_c * typ_c +
## abs_typ_shift_randomInt: simpson_c * NamingFreq_Study_c + (1 | subjCode) + (1 | category)
## abs_typ_shift_randomSlope: abs_typ_adv_choice ~ kinds_c + avgFam_c + judgment_c * typ_c +
## abs_typ_shift_randomSlope: simpson_c * NamingFreq_Study_c + (1 | category) + (judgment_c *
## abs_typ_shift_randomSlope: typ_c | subjCode)
## Df AIC BIC logLik deviance Chisq Chi Df
## abs_typ_shift_randomInt 12 3157.9 3213.6 -1567.0 3133.9
## abs_typ_shift_randomSlope 21 3173.0 3270.4 -1565.5 3131.0 2.9138 9
## Pr(>Chisq)
## abs_typ_shift_randomInt
## abs_typ_shift_randomSlope 0.9676summary(abs_typ_shift_randomInt)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: abs_typ_adv_choice ~ kinds_c + avgFam_c + judgment_c * typ_c +
## simpson_c * NamingFreq_Study_c + (1 | subjCode) + (1 | category)
## Data: filter(tme2_full, isRight == 0)
##
## REML criterion at convergence: 3157.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.7562 -0.7494 -0.1686 0.6374 2.6530
##
## Random effects:
## Groups Name Variance Std.Dev.
## subjCode (Intercept) 7.447e-02 0.2728970
## category (Intercept) 1.088e-07 0.0003299
## Residual 3.539e+00 1.8813068
## Number of obs: 763, groups: subjCode, 45; category, 45
##
## Fixed effects:
## Estimate Std. Error df t value
## (Intercept) 3.15018 0.08221 43.62511 38.320
## kinds_c -0.21608 0.11084 744.92154 -1.949
## avgFam_c 0.25240 0.33072 744.27049 0.763
## judgment_c -0.11021 0.14940 217.29292 -0.738
## typ_c 0.25854 0.02806 745.23225 9.214
## simpson_c -0.24992 0.23125 749.10282 -1.081
## NamingFreq_Study_c 0.06780 0.06861 46.94793 0.988
## judgment_c:typ_c 0.01290 0.05684 751.36464 0.227
## simpson_c:NamingFreq_Study_c 0.15642 0.18284 748.09943 0.855
## Pr(>|t|)
## (Intercept) <2e-16 ***
## kinds_c 0.0516 .
## avgFam_c 0.4456
## judgment_c 0.4615
## typ_c <2e-16 ***
## simpson_c 0.2802
## NamingFreq_Study_c 0.3281
## judgment_c:typ_c 0.8205
## simpson_c:NamingFreq_Study_c 0.3926
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) knds_c avgFm_ jdgmn_ typ_c smpsn_ NmF_S_ jdg_:_
## kinds_c 0.087
## avgFam_c -0.017 -0.335
## judgment_c 0.086 -0.051 0.081
## typ_c -0.002 0.026 0.017 0.020
## simpson_c -0.043 0.144 -0.371 -0.022 0.008
## NmngFrq_St_ -0.039 -0.001 0.032 -0.093 0.002 -0.043
## jdgmnt_c:t_ 0.014 0.094 0.075 0.001 0.081 -0.032 0.021
## smps_:NF_S_ -0.036 -0.036 -0.006 0.018 -0.022 -0.044 -0.077 -0.013Raw value (random slope model has singular fit)
typ_shift_randomSlope <- lmer(typ_adv_choice ~kinds_c+avgFam_c+judgment_c*typ_c+simpson_c*NamingFreq_Study_c+
(1|category)+(typ_c*judgment_c|subjCode),
data=filter(tme2_full, isRight==0))## boundary (singular) fit: see ?isSingular## Warning: Model failed to converge with 1 negative eigenvalue: -8.2e+01typ_shift_randomInt <- lmer(typ_adv_choice ~kinds_c+avgFam_c+judgment_c*typ_c+simpson_c*NamingFreq_Study_c+
(1|subjCode)+(1|category),
data=filter(tme2_full, isRight==0))
anova(typ_shift_randomInt,typ_shift_randomSlope) #random slope has singular fit and does not improve fit## refitting model(s) with ML (instead of REML)## Data: filter(tme2_full, isRight == 0)
## Models:
## typ_shift_randomInt: typ_adv_choice ~ kinds_c + avgFam_c + judgment_c * typ_c + simpson_c *
## typ_shift_randomInt: NamingFreq_Study_c + (1 | subjCode) + (1 | category)
## typ_shift_randomSlope: typ_adv_choice ~ kinds_c + avgFam_c + judgment_c * typ_c + simpson_c *
## typ_shift_randomSlope: NamingFreq_Study_c + (1 | category) + (typ_c * judgment_c |
## typ_shift_randomSlope: subjCode)
## Df AIC BIC logLik deviance Chisq Chi Df
## typ_shift_randomInt 12 3563.9 3619.5 -1769.9 3539.9
## typ_shift_randomSlope 21 3568.1 3665.5 -1763.0 3526.1 13.79 9
## Pr(>Chisq)
## typ_shift_randomInt
## typ_shift_randomSlope 0.13summary(typ_shift_randomInt)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula:
## typ_adv_choice ~ kinds_c + avgFam_c + judgment_c * typ_c + simpson_c *
## NamingFreq_Study_c + (1 | subjCode) + (1 | category)
## Data: filter(tme2_full, isRight == 0)
##
## REML criterion at convergence: 3556.9
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5773 -0.7167 0.1930 0.7877 1.8076
##
## Random effects:
## Groups Name Variance Std.Dev.
## subjCode (Intercept) 0.1025 0.3201
## category (Intercept) 0.3206 0.5662
## Residual 5.8166 2.4118
## Number of obs: 763, groups: subjCode, 45; category, 45
##
## Fixed effects:
## Estimate Std. Error df t value
## (Intercept) 1.07377 0.13436 37.14709 7.991
## kinds_c -0.14067 0.19650 40.30703 -0.716
## avgFam_c -1.49139 0.59351 38.58857 -2.513
## judgment_c -0.08893 0.19145 198.85191 -0.464
## typ_c 1.04934 0.03631 730.89230 28.898
## simpson_c 1.17047 0.41478 38.22704 2.822
## NamingFreq_Study_c 0.16027 0.08644 45.96493 1.854
## judgment_c:typ_c -0.13626 0.07521 746.56308 -1.812
## simpson_c:NamingFreq_Study_c -0.05472 0.23589 727.01733 -0.232
## Pr(>|t|)
## (Intercept) 1.37e-09 ***
## kinds_c 0.47819
## avgFam_c 0.01627 *
## judgment_c 0.64281
## typ_c < 2e-16 ***
## simpson_c 0.00753 **
## NamingFreq_Study_c 0.07013 .
## judgment_c:typ_c 0.07042 .
## simpson_c:NamingFreq_Study_c 0.81661
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) knds_c avgFm_ jdgmn_ typ_c smpsn_ NmF_S_ jdg_:_
## kinds_c 0.051
## avgFam_c -0.010 -0.353
## judgment_c 0.067 -0.035 0.051
## typ_c -0.001 0.019 0.010 0.023
## simpson_c -0.017 0.168 -0.382 -0.010 0.010
## NmngFrq_St_ -0.031 -0.003 0.027 -0.094 0.000 -0.030
## jdgmnt_c:t_ 0.024 0.071 0.046 0.010 0.084 -0.029 0.013
## smps_:NF_S_ -0.024 -0.026 -0.005 0.015 -0.020 -0.031 -0.078 -0.020Compare absolute value random-intercept models
#abs_model1 <- lmer(abs_typ_adv_choice ~ avgFam_c + (1|subjCode)+(1|category), data=filter(tme2_full, isRight==0))
abs_model1 <- lmer(abs_typ_adv_choice ~ kinds_c + avgFam_c + (1|subjCode)+(1|category),
data=filter(tme2_full, isRight==0)) #singular fit## boundary (singular) fit: see ?isSingularabs_model2 <- lmer(abs_typ_adv_choice ~ kinds_c + avgFam_c + judgment_c*typ_c+simpson_c*NamingFreq_Study_c+
(1|subjCode)+(1|category),
data=filter(tme2_full, isRight==0))
anova(abs_model1, abs_model2)## refitting model(s) with ML (instead of REML)## Data: filter(tme2_full, isRight == 0)
## Models:
## abs_model1: abs_typ_adv_choice ~ kinds_c + avgFam_c + (1 | subjCode) + (1 |
## abs_model1: category)
## abs_model2: abs_typ_adv_choice ~ kinds_c + avgFam_c + judgment_c * typ_c +
## abs_model2: simpson_c * NamingFreq_Study_c + (1 | subjCode) + (1 | category)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## abs_model1 6 3231.1 3258.9 -1609.5 3219.1
## abs_model2 12 3157.9 3213.6 -1567.0 3133.9 85.138 6 3.091e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Typicality shift model summaries
Absolute value:
Significant effect of kinds: typicality shift is smaller for categories that are more conceptually distinct
Significant effect of typ: typicality shift is greater when studied object was less typical (driven by typ=8) Absolute value model fits better when including all predictors (compared to model with only kind similarity and familiarity)
Raw value:
Significant effect of familiarity: for more familiar object categories, when people are wrong, they choose a less typical item than the one they studied
Significant effect of typ: typicality shift is greater when studied object was less typical
Significant effect of nameability: incorrect responses for more-nameable objects tend to be more typical than studied object Marginal effect of naming frequency: people who name more during study tend to choose more-typical objects when they’re wrong Marginal interaction between typicality and judgment: effect of studied typ on shift is smaller for ‘remember’ judgments than ‘know’ judgments (specifically from typ == 8)
Density plots for typicality shift depending on studied typ and judgment
Absolute value
meanshift2 <- filter(tme2_full2, isRight==0) %>%
group_by(typ,judgment) %>%
summarise(mean=mean(abs_typ_adv_choice))
ggplot(filter(tme2_full2, isRight==0), aes(x=abs_typ_adv_choice, color=as.factor(typ), fill=as.factor(typ)))+
geom_density(alpha=.05)+
# geom_histogram(aes(y=..density..), color="black", fill="white", alpha=.03)+
geom_vline(data=meanshift2, aes(xintercept=mean, color=as.factor(typ)), linetype="dashed")+
facet_grid(typ~judgment)+
scale_x_continuous(name="Absolute value of typicality shift", limits=c(0,8),breaks=c(1,2,3,4,5,6,7,8))+
theme_classic()
Raw value
meanshift_raw2 <- filter(tme2_full2, isRight==0) %>%
group_by(typ,judgment) %>%
summarise(mean=mean(abs_typ_adv_choice))
ggplot(filter(tme2_full2, isRight==0), aes(x=typ_adv_choice, color=as.factor(typ), fill=as.factor(typ)))+
geom_density(alpha=.05)+
# geom_histogram(aes(y=..density..), color="black", fill="white", alpha=.03)+
geom_vline(data=meanshift_raw2, aes(xintercept=mean, color=as.factor(typ)), linetype="dashed")+
facet_grid(typ~judgment)+
scale_x_continuous(name="Raw value of typicality shift", limits=c(-8,8),breaks=c(-7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6,7,8))+
theme_classic()
Histograms for typicality shift depending on studied typ and judgment
Absolute value
meanshift2 <- filter(tme2_full2, isRight==0) %>%
group_by(typ,judgment) %>%
summarise(mean=mean(abs_typ_adv_choice))
ggplot(filter(tme2_full2, isRight==0), aes(x=abs_typ_adv_choice, color=as.factor(typ), fill=as.factor(typ)))+
geom_histogram(alpha=.1)+
geom_vline(data=meanshift2, aes(xintercept=mean, color=as.factor(typ)), linetype="dashed")+
facet_grid(typ~judgment)+
scale_x_continuous(name="Absolute value of typicality shift", limits=c(0,8),breaks=c(1,2,3,4,5,6,7,8))+
theme_classic()## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.## Warning: Removed 12 rows containing missing values (geom_bar).
Raw value
meanshift_raw2 <- filter(tme2_full2, isRight==0) %>%
group_by(typ,judgment) %>%
summarise(mean=mean(typ_adv_choice))
ggplot(filter(tme2_full2, isRight==0), aes(x=typ_adv_choice, color=as.factor(typ), fill=as.factor(typ)))+
geom_histogram(alpha=.1)+
geom_vline(data=meanshift_raw2, aes(xintercept=mean, color=as.factor(typ)), linetype="dashed")+
facet_grid(typ~judgment)+
scale_x_continuous(name="Raw value of typicality shift", limits=c(-8,8),breaks=c(-7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6,7,8))+
theme_classic()## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.## Warning: Removed 12 rows containing missing values (geom_bar).