ManyBabies Simulated Analysis
Preliminaries
options(dplyr.width = Inf)
knitr::opts_chunk$set(message = FALSE, warning = FALSE, cache=TRUE)
library(ggplot2)
library(dplyr)##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, unionlibrary(lme4)## Loading required package: Matrixlibrary(tidyr)##
## Attaching package: 'tidyr'## The following object is masked from 'package:Matrix':
##
## expandlibrary(magrittr)##
## Attaching package: 'magrittr'## The following object is masked from 'package:tidyr':
##
## extractlibrary(lsmeans)## Warning: package 'lsmeans' was built under R version 3.2.4## Loading required package: estimabilitylibrary(langcog)##
## Attaching package: 'langcog'## The following object is masked from 'package:base':
##
## scaletheme_manylabs <- theme_bw() +
theme(axis.text = element_text(size = 14),
axis.title = element_text(size = 16, family="Arial"),
text = element_text(family="Arial"),
legend.key = element_rect(fill = "navy"),
legend.background = element_rect(fill = "white"),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
panel.border = element_blank(),
axis.line.y = element_line(),
axis.line.x = element_line(),
strip.background = element_rect(fill = "white", colour = NA),
strip.text.x = element_text(size = 14, face = "bold", colour = "black"))Simulate Data
Create basics of data.
data <- expand.grid(Lab = factor(paste0('Lab',1:20)),
Subject = 1:30,
Trial = 1:16)Simulation sequence:
- assign block
- randomly assign conditions within blocks
- randomly generate looking times within constraints
- do it lab-by-lab and then add some subject attention diffs
- for (1) realism and (2) convergence issues
- max 20s looking per trial
- randomly generate ages
removed this from first batch and created a second batch varyng those two factors + randomly generate method and participant vars (session, language, bilingual)
blocks <- factor(c('IDS','IDS','ADS','ADS'))
langs <- factor(c('American English','French','Spanish','German','Japanese', 'British English'))
methods <- factor(c('HPP','SingleScreen','AnotherMethod'))
sessions <- factor(c('First','Second'))
bilingual <- c(TRUE, FALSE)
lt_max <- 20 # max LT in a trial
data %<>%
mutate(Subject = factor(toupper(paste0(Lab,'-',Subject))),
Block = ((Trial-1) %/% 4) + 1) %>%
group_by(Subject,Block) %>%
mutate(Condition = sample(blocks)) %>%
group_by(Lab) %>%
mutate(.LabDiff = runif(1, 0, .5)) %>%
ungroup() %>%
mutate(LT = ifelse(Condition == 'IDS',
rlnorm(n(), 1.5 + .LabDiff, .7),
rlnorm(n(), 1.5, .7))) %>%
group_by(Subject) %>%
mutate(LT = LT + runif(1,0,.5)) %>%
ungroup() %>%
mutate(LT = ifelse(LT > lt_max, lt_max, LT)) %>%
group_by(Lab) %>%
mutate(.LabMeanAge = round(runif(1, 3, 12))) %>%
ungroup() %>%
group_by(Subject) %>%
mutate(Age = round(runif(1, .LabMeanAge-.5, .LabMeanAge+.5),2)) %>%
ungroup() %>%
group_by(Lab) %>%
mutate(Method = sample(methods,1)) %>%
ungroup() %>%
group_by(Subject) %>%
mutate(Session = "First",
Language = sample(langs, 1),
Bilingual = FALSE) %>%
arrange(Lab,Subject,Trial) %>%
select(-starts_with("."))Additional data labs contribute after testing the “promised” children is complete.
- Number of labs: 5 (instead of 20)
- Lab-specific difference decreased (from 0-.5 to 0-.25)
- Looking times slightly decreased
additional_data <- expand.grid(Lab = factor(paste0('Lab',1:5)),
Subject = 1:30,
Trial = 1:16)
additional_data %<>%
mutate(Subject = factor(toupper(paste0(Lab,'-',Subject))),
Block = ((Trial-1) %/% 4) + 1) %>%
group_by(Subject,Block) %>%
mutate(Condition = sample(blocks)) %>%
group_by(Lab) %>%
mutate(.LabDiff = runif(1, .0, .25)) %>%
ungroup() %>%
mutate(LT = ifelse(Condition == 'IDS',
rlnorm(n(), 1.1 + .LabDiff, .8),
rlnorm(n(), 1.1, .8))) %>%
group_by(Subject) %>%
mutate(LT = LT + runif(1,0,.5)) %>%
ungroup() %>%
mutate(LT = ifelse(LT > lt_max, lt_max, LT)) %>%
group_by(Lab) %>%
mutate(.LabMeanAge = round(runif(1, 3, 12))) %>%
ungroup() %>%
group_by(Subject) %>%
mutate(Age = round(runif(1, .LabMeanAge-.5, .LabMeanAge+.5),2)) %>%
ungroup() %>%
group_by(Lab) %>%
mutate(Method = sample(methods,1)) %>%
ungroup() %>%
group_by(Subject) %>%
mutate(Session = sample(sessions,1),
Language = sample(langs, 1),
Bilingual = sample(bilingual, 1)) %>%
arrange(Lab,Subject,Trial) %>%
select(-starts_with("."))Data Cleaning
- exclude outliers based on log-transformed LT
- filter based on minimum trials of each type
- remove hidden fields
lt_min <- 2 # minumum LT for inclusion
z_threshold <- 3 # outlier threshold (sds)
min_trials_per_type <- 4 # min trials per type for inclusion
data_clean <- data %>%
filter(LT >= lt_min) %>%
group_by(Subject) %>%
mutate(log_lt = log(LT),
.scaled_log_lt = as.numeric(langcog::scale(log_lt))) %>%
filter(abs(.scaled_log_lt) < z_threshold) %>%
group_by(Subject) %>%
mutate(.N_IDS = sum(Condition == "IDS"),
.N_ADS = sum(Condition == "ADS")) %>%
filter(.N_IDS >= min_trials_per_type &
.N_ADS >= min_trials_per_type) %>%
select(-starts_with("."))
additional_data_clean <- additional_data %>%
filter(LT >= lt_min) %>%
group_by(Subject) %>%
mutate(log_lt = log(LT),
.scaled_log_lt = as.numeric(langcog::scale(log_lt))) %>%
filter(abs(.scaled_log_lt) < z_threshold) %>%
group_by(Subject) %>%
mutate(.N_IDS = sum(Condition == "IDS"),
.N_ADS = sum(Condition == "ADS")) %>%
filter(.N_IDS >= min_trials_per_type &
.N_ADS >= min_trials_per_type) %>%
select(-starts_with("."))Join two datasets for further analyses
all_data_clean = rbind(data_clean, additional_data_clean)Distributions
“initial” data
ggplot(data_clean, aes(x = LT)) +
geom_histogram() +
facet_grid(~Condition)
all data
ggplot(all_data_clean, aes(x = LT)) +
geom_histogram() +
facet_grid(~Condition)
Transformation
ggplot(data_clean, aes(x=log_lt)) +
geom_histogram() +
facet_grid(~Condition)
Create Aggregated Datasets
agg_subjects <- data_clean %>%
group_by(Lab, Method, Subject, Language, Condition, Age) %>%
summarise(MeanLogLT = mean(log_lt)) %>%
mutate(ConditionC = ifelse(Condition == "IDS", .5, -.5)) %>%
mutate(Native = ifelse(Language == "American English", TRUE, FALSE))
agg_subjects_paired <- agg_subjects %>%
select(-ConditionC) %>%
spread(Condition, MeanLogLT) %>%
mutate(Diff = IDS - ADS,
Prop = IDS / (IDS + ADS))
all_agg_subjects <- all_data_clean %>%
group_by(Lab, Method, Session, Subject, Language, Bilingual, Condition, Age) %>%
summarise(MeanLogLT = mean(log_lt)) %>%
mutate(ConditionC = ifelse(Condition == "IDS", .5, -.5)) %>%
mutate(Native = ifelse(Language == "American English", TRUE, FALSE))
all_agg_subjects_paired <- all_agg_subjects %>%
select(-ConditionC) %>%
spread(Condition, MeanLogLT) %>%
mutate(Diff = IDS - ADS,
Prop = IDS / (IDS + ADS))Hypothesis Tests
Overall Preference for IDS v ADS
ggplot(agg_subjects, aes(x=Condition, y=MeanLogLT, group=Subject)) +
geom_boxplot() +
geom_jitter(width = .25, size =.5) +
theme_manylabs
ggplot(agg_subjects_paired, aes(x='Overall', y=Diff)) +
geom_boxplot() +
geom_jitter(width = .25, size =.5) +
geom_hline(yintercept=0, linetype="dashed", alpha=.5) +
scale_x_discrete('') +
theme_manylabs
t.test(agg_subjects_paired$Diff)##
## One Sample t-test
##
## data: agg_subjects_paired$Diff
## t = 10.071, df = 599, p-value < 2.2e-16
## alternative hypothesis: true mean is not equal to 0
## 95 percent confidence interval:
## 0.1021410 0.1516268
## sample estimates:
## mean of x
## 0.1268839mean(agg_subjects_paired$Diff) / sd(agg_subjects_paired$Diff)## [1] 0.411156model <- lmer(MeanLogLT ~ ConditionC +
(ConditionC | Lab) +
(1 | Subject),
data=agg_subjects, REML=FALSE)
summary(model)## Linear mixed model fit by maximum likelihood ['lmerMod']
## Formula: MeanLogLT ~ ConditionC + (ConditionC | Lab) + (1 | Subject)
## Data: agg_subjects
##
## AIC BIC logLik deviance df.resid
## -285.5 -249.9 149.8 -299.5 1193
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.91446 -0.67929 -0.03054 0.65325 3.12569
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## Subject (Intercept) 0.001024 0.03200
## Lab (Intercept) 0.001487 0.03856
## ConditionC 0.009001 0.09488 0.82
## Residual 0.043038 0.20746
## Number of obs: 1200, groups: Subject, 600; Lab, 20
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 1.74427 0.01058 164.89
## ConditionC 0.12688 0.02436 5.21
##
## Correlation of Fixed Effects:
## (Intr)
## ConditionC 0.582drop1(model,~.,test="Chi")## Single term deletions
##
## Model:
## MeanLogLT ~ ConditionC + (ConditionC | Lab) + (1 | Subject)
## Df AIC LRT Pr(Chi)
## <none> -285.51
## ConditionC 1 -270.37 17.141 3.47e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Lab Variability
ggplot(agg_subjects_paired, aes(x=Lab, y=Diff)) +
stat_summary(fun.y='mean', geom='point') +
stat_summary(fun.data='mean_cl_normal', geom='errorbar', width=.1, fun.args=list(mult=2)) +
geom_hline(yintercept=0, linetype="dashed", alpha=.5) +
scale_x_discrete('') +
coord_flip() +
theme_manylabs
model <- lmer(MeanLogLT ~ ConditionC + (1 + ConditionC | Lab) + (1 | Subject), data=agg_subjects, REML=F)
fixed_effect <- fixef(model)[['ConditionC']]
lab_ranefs <- data.frame(Lab = factor(rownames(ranef(model)$Lab),
levels=rownames(ranef(model)$Lab)),
ConditionRanef = ranef(model)$Lab$ConditionC +
fixed_effect)
ggplot(lab_ranefs, aes(x=Lab, y=ConditionRanef, group=Lab)) +
geom_point() +
geom_errorbar(aes(ymin=fixed_effect, ymax=ConditionRanef), width=.1) +
geom_hline(yintercept=0, linetype="dashed", alpha=.5) +
geom_hline(yintercept=fixed_effect, linetype="solid", alpha=.5) +
scale_x_discrete('') +
scale_y_continuous('Random Effect') +
coord_flip() +
theme_manylabs
Does IDS preference change by age?
agg_subjects %<>%
ungroup() %>%
mutate(AgeC = Age - mean(Age))
ggplot(agg_subjects_paired, aes(x=Age, y=Diff)) +
geom_point() +
stat_smooth() +
geom_hline(yintercept=0, linetype="dashed", alpha=.5) +
theme_manylabs
model <- lmer(MeanLogLT ~ ConditionC*AgeC +
(1 + ConditionC + AgeC | Lab) +
(1 | Subject), data=agg_subjects,
REML=FALSE)
summary(model)## Linear mixed model fit by maximum likelihood ['lmerMod']
## Formula: MeanLogLT ~ ConditionC * AgeC + (1 + ConditionC + AgeC | Lab) +
## (1 | Subject)
## Data: agg_subjects
##
## AIC BIC logLik deviance df.resid
## -276.7 -215.6 150.3 -300.7 1188
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.89873 -0.68124 -0.03555 0.65026 3.14672
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## Subject (Intercept) 1.046e-03 0.032342
## Lab (Intercept) 1.124e-03 0.033532
## ConditionC 8.672e-03 0.093124 0.79
## AgeC 1.494e-05 0.003865 0.86 0.99
## Residual 4.305e-02 0.207483
## Number of obs: 1200, groups: Subject, 600; Lab, 20
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 1.7435936 0.0099751 174.79
## ConditionC 0.1268839 0.0240230 5.28
## AgeC 0.0009819 0.0032051 0.31
## ConditionC:AgeC 0.0044125 0.0080797 0.55
##
## Correlation of Fixed Effects:
## (Intr) CndtnC AgeC
## ConditionC 0.512
## AgeC 0.305 0.232
## CondtnC:AgC 0.203 0.000 0.424drop1(model,~.,test="Chi")## Single term deletions
##
## Model:
## MeanLogLT ~ ConditionC * AgeC + (1 + ConditionC + AgeC | Lab) +
## (1 | Subject)
## Df AIC LRT Pr(Chi)
## <none> -276.69
## ConditionC 1 -261.29 17.4094 3.013e-05 ***
## AgeC 1 -278.61 0.0873 0.7676
## ConditionC:AgeC 1 -278.41 0.2825 0.5951
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Does quadratic age term improve the fit?
# model with linear+quadratic random effect of age, but only linear fixed effect
model <- lmer(MeanLogLT ~ ConditionC*poly(AgeC,1) +
(1 + ConditionC + poly(AgeC,2) | Lab) +
(1 | Subject),
data=agg_subjects, REML=FALSE)
# model with linear+quadratic random and fixed effects of age
model_2 <- lmer(MeanLogLT ~ ConditionC*poly(AgeC,2) +
(1 + ConditionC + poly(AgeC,2) | Lab) +
(1 | Subject), data=agg_subjects,
REML=FALSE)
anova(model,model_2)## Data: agg_subjects
## Models:
## model: MeanLogLT ~ ConditionC * poly(AgeC, 1) + (1 + ConditionC + poly(AgeC,
## model: 2) | Lab) + (1 | Subject)
## model_2: MeanLogLT ~ ConditionC * poly(AgeC, 2) + (1 + ConditionC + poly(AgeC,
## model_2: 2) | Lab) + (1 | Subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## model 16 -271.83 -190.39 151.92 -303.83
## model_2 18 -268.21 -176.58 152.10 -304.21 0.3703 2 0.831Are there trial order effects?
data_clean %<>%
ungroup() %>%
mutate(TrialC = Trial - mean(Trial),
ConditionC = ifelse(Condition == "IDS", .5, -.5),
AgeC = Age - mean(Age))
ggplot(data_clean, aes(x=Trial, y=log_lt, color=Condition)) +
stat_summary(fun.y='mean', geom='point') +
stat_summary(fun.y='mean', geom='line') +
stat_summary(fun.data='mean_cl_normal', geom='errorbar', width=.1, fun.args=list(mult=2)) +
theme_manylabs
# model <- lmer(log_lt ~ ConditionC*AgeC*TrialC +
# (1 + ConditionC + AgeC + TrialC | Lab) +
# (1 + TrialC + ConditionC | Subject),
# data=data_clean, REML=FALSE)
# summary(model)
# drop1(model,~.,test="Chi")Moderator Analyses
Method
Note age | lab doesn’t converge.
#NB: for some reason I cannot knit without adding this bit again.
agg_subjects %<>%
ungroup() %>%
mutate(AgeC = Age - mean(Age))
ggplot(agg_subjects_paired, aes(x=Method, y=Diff)) +
geom_boxplot() +
geom_jitter(width = .25, size =.5) +
geom_hline(yintercept=0, linetype="dashed", alpha=.5) +
scale_x_discrete('') +
theme_manylabs
contrasts(agg_subjects$Method) <- contr.sum(length(unique(agg_subjects$Method)))
model <- lmer(MeanLogLT ~ ConditionC * Method * AgeC +
(1 + ConditionC | Lab) +
(1 | Subject),
data=agg_subjects, REML=FALSE)
summary(model)## Linear mixed model fit by maximum likelihood ['lmerMod']
## Formula:
## MeanLogLT ~ ConditionC * Method * AgeC + (1 + ConditionC | Lab) +
## (1 | Subject)
## Data: agg_subjects
##
## AIC BIC logLik deviance df.resid
## -288.1 -201.5 161.0 -322.1 1183
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.0178 -0.6766 -0.0249 0.6659 3.1945
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## Subject (Intercept) 0.0007343 0.02710
## Lab (Intercept) 0.0006466 0.02543
## ConditionC 0.0072901 0.08538 1.00
## Residual 0.0429444 0.20723
## Number of obs: 1200, groups: Subject, 600; Lab, 20
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 1.6721338 0.0280075 59.70
## ConditionC 0.1034075 0.0694399 1.49
## Method1 0.1031838 0.0292002 3.53
## Method2 0.0618840 0.0288342 2.15
## AgeC 0.0287053 0.0106858 2.69
## ConditionC:Method1 0.0331589 0.0729471 0.45
## ConditionC:Method2 0.0064572 0.0718737 0.09
## ConditionC:AgeC 0.0164965 0.0262665 0.63
## Method1:AgeC -0.0273000 0.0109887 -2.48
## Method2:AgeC -0.0254217 0.0109723 -2.32
## ConditionC:Method1:AgeC -0.0008044 0.0271583 -0.03
## ConditionC:Method2:AgeC -0.0255790 0.0271076 -0.94
##
## Correlation of Fixed Effects:
## (Intr) CndtnC Methd1 Methd2 AgeC CnC:M1 CnC:M2 CnC:AC Mt1:AC
## ConditionC 0.417
## Method1 -0.876 -0.347
## Method2 -0.913 -0.368 0.795
## AgeC -0.917 -0.355 0.891 0.897
## CndtnC:Mth1 -0.345 -0.853 0.431 0.300 0.346
## CndtnC:Mth2 -0.366 -0.897 0.300 0.427 0.347 0.759
## CondtnC:AgC -0.358 -0.900 0.351 0.352 0.401 0.872 0.878
## Method1:AgC 0.904 0.353 -0.843 -0.885 -0.917 -0.321 -0.345 -0.355
## Method2:AgC 0.900 0.350 -0.875 -0.860 -0.921 -0.341 -0.330 -0.358 0.841
## CndtC:M1:AC 0.354 0.885 -0.325 -0.348 -0.353 -0.815 -0.864 -0.900 0.411
## CndtC:M2:AC 0.351 0.881 -0.344 -0.333 -0.356 -0.852 -0.835 -0.906 0.312
## Mt2:AC CC:M1:
## ConditionC
## Method1
## Method2
## AgeC
## CndtnC:Mth1
## CndtnC:Mth2
## CondtnC:AgC
## Method1:AgC
## Method2:AgC
## CndtC:M1:AC 0.312
## CndtC:M2:AC 0.410 0.811model_null <- lmer(MeanLogLT ~ ConditionC*AgeC +
(1 + ConditionC + AgeC | Lab) +
(1 | Subject), data=agg_subjects,
REML=FALSE)
anova(model, model_null)## Data: agg_subjects
## Models:
## model_null: MeanLogLT ~ ConditionC * AgeC + (1 + ConditionC + AgeC | Lab) +
## model_null: (1 | Subject)
## model: MeanLogLT ~ ConditionC * Method * AgeC + (1 + ConditionC | Lab) +
## model: (1 | Subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## model_null 12 -276.69 -215.61 150.35 -300.69
## model 17 -288.07 -201.54 161.03 -322.07 21.373 5 0.0006887 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# post-hoc least-squares contrasts
lstrends(model, ~ Method, var="ConditionC", adjust="none")## Method ConditionC.trend SE df lower.CL upper.CL
## AnotherMethod 0.13656641 0.04540421 26.85 0.04337977 0.2297531
## HPP 0.10986467 0.03768034 26.83 0.03252801 0.1872013
## SingleScreen 0.06379142 0.22557736 83.57 -0.38482725 0.5124101
##
## Confidence level used: 0.95Session
Tested on all data, but very imbalanced. But: How to select subset?
all_agg_subjects %<>%
ungroup() %>%
mutate(AgeC = Age - mean(Age))
ggplot(all_agg_subjects_paired, aes(x=Session, y=Diff)) +
geom_boxplot()+
geom_jitter(width = .25, size =.5)+
geom_hline(yintercept=0, linetype="dashed", alpha=.5) +
scale_x_discrete('') +
theme_manylabs
all_agg_subjects$Session = as.factor(all_agg_subjects$Session)
contrasts(all_agg_subjects$Session) <- contr.sum(length(unique(all_agg_subjects$Session)))
model <- lmer(MeanLogLT ~ ConditionC*Session*AgeC + (1 + ConditionC + AgeC | Lab) + (1 | Subject), data=all_agg_subjects, REML=F)
summary(model)## Linear mixed model fit by maximum likelihood ['lmerMod']
## Formula:
## MeanLogLT ~ ConditionC * Session * AgeC + (1 + ConditionC + AgeC |
## Lab) + (1 | Subject)
## Data: all_agg_subjects
##
## AIC BIC logLik deviance df.resid
## -233.8 -149.0 132.9 -265.8 1464
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.9514 -0.6655 -0.0237 0.6497 3.5202
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## Subject (Intercept) 0.0000000 0.00000
## Lab (Intercept) 0.0057544 0.07586
## ConditionC 0.0056142 0.07493 0.07
## AgeC 0.0009693 0.03113 -0.56 0.45
## Residual 0.0461829 0.21490
## Number of obs: 1480, groups: Subject, 600; Lab, 20
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 1.677279 0.023485 71.42
## ConditionC 0.049797 0.027050 1.84
## Session1 0.026245 0.012577 2.09
## AgeC 0.003532 0.010224 0.35
## ConditionC:Session1 0.069150 0.020691 3.34
## ConditionC:AgeC 0.004796 0.009662 0.50
## Session1:AgeC 0.012048 0.005664 2.13
## ConditionC:Session1:AgeC 0.004325 0.009921 0.44
##
## Correlation of Fixed Effects:
## (Intr) CndtnC Sessn1 AgeC CnC:S1 CnC:AC Ss1:AC
## ConditionC 0.042
## Session1 -0.402 -0.003
## AgeC -0.340 0.218 -0.144
## CndtnC:Sss1 -0.015 -0.660 0.004 -0.029
## CondtnC:AgC 0.036 0.121 -0.016 0.015 -0.160
## Sessin1:AgC -0.048 -0.012 0.139 -0.493 0.013 -0.009
## CndtC:S1:AC 0.022 -0.092 0.034 -0.007 0.122 -0.821 0.025model_null <- lmer(MeanLogLT ~ ConditionC*AgeC + (1 + ConditionC + AgeC | Lab) + (1 | Subject), data=all_agg_subjects, REML=F)
anova(model,model_null)## Data: all_agg_subjects
## Models:
## model_null: MeanLogLT ~ ConditionC * AgeC + (1 + ConditionC + AgeC | Lab) +
## model_null: (1 | Subject)
## model: MeanLogLT ~ ConditionC * Session * AgeC + (1 + ConditionC + AgeC |
## model: Lab) + (1 | Subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## model_null 12 -224.53 -160.93 124.26 -248.53
## model 16 -233.76 -148.97 132.88 -265.76 17.24 4 0.001736 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# post-hoc least-squares contrasts
lstrends(model, ~ Session, var="ConditionC", adjust="none")## Session ConditionC.trend SE df lower.CL upper.CL
## First 0.11894732 0.02155302 22.86 0.07434595 0.16354870
## Second -0.01935252 0.04505958 244.44 -0.10810712 0.06940208
##
## Confidence level used: 0.95Nativeness of the test (American English versus everyone else)
Tested only on the “initial” dataset, analysis conditional on meeting a minimum N for each language to be included
agg_subjects %<>%
ungroup() %>%
mutate(AgeC = Age - mean(Age))
ggplot(agg_subjects_paired, aes(x=Native, y=Diff)) +
geom_boxplot()+
geom_jitter(width = .25, size =.5)+
geom_hline(yintercept=0, linetype="dashed", alpha=.5) +
scale_x_discrete('') +
theme_manylabs
contrasts(agg_subjects$Native) <- contr.sum(length(unique(agg_subjects$Native)))
model <- lmer(MeanLogLT ~ ConditionC*Native*AgeC + (1 + ConditionC + AgeC | Lab) + (1 | Subject), data=agg_subjects, REML=F)
summary(model)## Linear mixed model fit by maximum likelihood ['lmerMod']
## Formula:
## MeanLogLT ~ ConditionC * Native * AgeC + (1 + ConditionC + AgeC |
## Lab) + (1 | Subject)
## Data: agg_subjects
##
## AIC BIC logLik deviance df.resid
## -270.1 -188.7 151.1 -302.1 1184
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.89672 -0.66647 -0.03374 0.64422 3.13995
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## Subject (Intercept) 1.079e-03 0.032854
## Lab (Intercept) 1.145e-03 0.033840
## ConditionC 8.861e-03 0.094131 0.78
## AgeC 1.327e-05 0.003642 0.86 0.99
## Residual 4.295e-02 0.207236
## Number of obs: 1200, groups: Subject, 600; Lab, 20
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 1.7400607 0.0111155 156.54
## ConditionC 0.1216380 0.0260154 4.68
## Native1 0.0057668 0.0078573 0.73
## AgeC 0.0013457 0.0036832 0.37
## ConditionC:Native1 0.0085479 0.0153529 0.56
## ConditionC:AgeC 0.0018628 0.0088963 0.21
## Native1:AgeC -0.0005989 0.0028030 -0.21
## ConditionC:Native1:AgeC 0.0039913 0.0054757 0.73
##
## Correlation of Fixed Effects:
## (Intr) CndtnC Nativ1 AgeC CnC:N1 CnC:AC Nt1:AC
## ConditionC 0.433
## Native1 -0.438 -0.002
## AgeC 0.226 0.178 0.007
## CndtnC:Ntv1 -0.004 -0.366 0.006 -0.003
## CondtnC:AgC 0.157 -0.001 0.001 0.345 -0.002
## Native1:AgC 0.005 -0.002 -0.001 -0.487 0.004 0.003
## CndtC:N1:AC -0.003 0.003 0.004 -0.002 -0.001 -0.400 0.005model_null <- lmer(MeanLogLT ~ ConditionC*AgeC + (1 + ConditionC + AgeC | Lab) + (1 | Subject), data=agg_subjects, REML=F)
anova(model,model_null)## Data: agg_subjects
## Models:
## model_null: MeanLogLT ~ ConditionC * AgeC + (1 + ConditionC + AgeC | Lab) +
## model_null: (1 | Subject)
## model: MeanLogLT ~ ConditionC * Native * AgeC + (1 + ConditionC + AgeC |
## model: Lab) + (1 | Subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## model_null 12 -276.69 -215.61 150.35 -300.69
## model 16 -270.11 -188.66 151.05 -302.11 1.4109 4 0.8423# post-hoc least-squares contrasts
lstrends(model, ~ Native, var="ConditionC", adjust="none")## Native ConditionC.trend SE df lower.CL upper.CL
## FALSE 0.1301858 0.02614705 24.54 0.07628317 0.1840885
## TRUE 0.1130901 0.03567997 87.80 0.04218145 0.1839988
##
## Confidence level used: 0.95Native Language
Tested only on the “initial” dataset, analysis conditional on meeting a minimum N for each language to be included
ggplot(agg_subjects_paired, aes(x=Language, y=Diff)) +
geom_boxplot()+
geom_jitter(width = .25, size =.5)+
geom_hline(yintercept=0, linetype="dashed", alpha=.5) +
scale_x_discrete('') +
theme_manylabs
contrasts(agg_subjects$Language) <- contr.sum(length(unique(agg_subjects$Language)))
model <- lmer(MeanLogLT ~ ConditionC*Language*AgeC + (1 + ConditionC + AgeC | Lab) + (1 | Subject), data=agg_subjects, REML=F)
summary(model)## Linear mixed model fit by maximum likelihood ['lmerMod']
## Formula: MeanLogLT ~ ConditionC * Language * AgeC + (1 + ConditionC +
## AgeC | Lab) + (1 | Subject)
## Data: agg_subjects
##
## AIC BIC logLik deviance df.resid
## -252.0 -89.1 158.0 -316.0 1168
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.8947 -0.6715 -0.0304 0.6536 3.0588
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## Subject (Intercept) 1.179e-03 0.034334
## Lab (Intercept) 8.627e-04 0.029372
## ConditionC 8.881e-03 0.094239 0.98
## AgeC 6.475e-05 0.008046 0.66 0.49
## Residual 4.233e-02 0.205749
## Number of obs: 1200, groups: Subject, 600; Lab, 20
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 1.7437674 0.0101251 172.22
## ConditionC 0.1272437 0.0242266 5.25
## Language1 -0.0102703 0.0130506 -0.79
## Language2 0.0098594 0.0136133 0.72
## Language3 0.0110850 0.0138097 0.80
## Language4 0.0075649 0.0145975 0.52
## Language5 -0.0015339 0.0139931 -0.11
## AgeC 0.0010895 0.0034523 0.32
## ConditionC:Language1 -0.0142435 0.0254293 -0.56
## ConditionC:Language2 -0.0223540 0.0265449 -0.84
## ConditionC:Language3 0.0691236 0.0269506 2.56
## ConditionC:Language4 0.0009791 0.0284860 0.03
## ConditionC:Language5 -0.0114283 0.0273439 -0.42
## ConditionC:AgeC 0.0050678 0.0081435 0.62
## Language1:AgeC 0.0014774 0.0046596 0.32
## Language2:AgeC -0.0007674 0.0046821 -0.16
## Language3:AgeC -0.0003546 0.0046187 -0.08
## Language4:AgeC -0.0023421 0.0050166 -0.47
## Language5:AgeC -0.0046067 0.0051137 -0.90
## ConditionC:Language1:AgeC -0.0070830 0.0090741 -0.78
## ConditionC:Language2:AgeC 0.0017861 0.0091328 0.20
## ConditionC:Language3:AgeC -0.0082213 0.0090034 -0.91
## ConditionC:Language4:AgeC 0.0106457 0.0097789 1.09
## ConditionC:Language5:AgeC 0.0007898 0.0099847 0.08model_null <- lmer(MeanLogLT ~ ConditionC*AgeC + (1 + ConditionC + AgeC | Lab) + (1 | Subject), data=agg_subjects, REML=F)
anova(model,model_null)## Data: agg_subjects
## Models:
## model_null: MeanLogLT ~ ConditionC * AgeC + (1 + ConditionC + AgeC | Lab) +
## model_null: (1 | Subject)
## model: MeanLogLT ~ ConditionC * Language * AgeC + (1 + ConditionC +
## model: AgeC | Lab) + (1 | Subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## model_null 12 -276.69 -215.614 150.35 -300.69
## model 32 -251.98 -89.096 157.99 -315.98 15.284 20 0.76# post-hoc least-squares contrasts
lstrends(model, ~ Language, var="ConditionC", adjust="none")## Language ConditionC.trend SE df lower.CL upper.CL
## American English 0.1130002 0.03568783 87.52 0.04207276 0.1839277
## British English 0.1048897 0.03690440 99.12 0.03166447 0.1781149
## French 0.1963672 0.03734649 103.48 0.12230334 0.2704311
## German 0.1282227 0.03906431 122.30 0.05089293 0.2055525
## Japanese 0.1158153 0.03777210 106.68 0.04093397 0.1906967
## Spanish 0.1051668 0.03728583 103.02 0.03121934 0.1791143
##
## Confidence level used: 0.95Biligualism
Tested on all data, but very imbalanced. But: How to select subset?
ggplot(all_agg_subjects_paired, aes(x=Bilingual, y=Diff)) +
geom_boxplot()+
geom_jitter(width = .25, size =.5)+
geom_hline(yintercept=0, linetype="dashed", alpha=.5) +
scale_x_discrete('') +
theme_manylabs
contrasts(all_agg_subjects$Bilingual) <- contr.sum(length(unique(all_agg_subjects$Bilingual)))
model <- lmer(MeanLogLT ~ ConditionC*Bilingual*AgeC + (1 + ConditionC + AgeC | Lab) + (1 | Subject), data=all_agg_subjects, REML=F)
summary(model)## Linear mixed model fit by maximum likelihood ['lmerMod']
## Formula: MeanLogLT ~ ConditionC * Bilingual * AgeC + (1 + ConditionC +
## AgeC | Lab) + (1 | Subject)
## Data: all_agg_subjects
##
## AIC BIC logLik deviance df.resid
## -228.2 -143.4 130.1 -260.2 1464
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.4108 -0.6649 -0.0200 0.6500 3.4557
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## Subject (Intercept) 0.000000 0.00000
## Lab (Intercept) 0.005266 0.07257
## ConditionC 0.005734 0.07572 0.24
## AgeC 0.000811 0.02848 -0.48 0.30
## Residual 0.046393 0.21539
## Number of obs: 1480, groups: Subject, 600; Lab, 20
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 1.6647332 0.0226689 73.44
## ConditionC 0.0810985 0.0270180 3.00
## Bilingual1 0.0390717 0.0123599 3.16
## AgeC 0.0148000 0.0096759 1.53
## ConditionC:Bilingual1 0.0335077 0.0205000 1.63
## ConditionC:AgeC 0.0018924 0.0094660 0.20
## Bilingual1:AgeC -0.0007156 0.0056052 -0.13
## ConditionC:Bilingual1:AgeC 0.0100267 0.0097042 1.03
##
## Correlation of Fixed Effects:
## (Intr) CndtnC Blngl1 AgeC CnC:B1 CnC:AC Bl1:AC
## ConditionC 0.125
## Bilingual1 -0.416 -0.006
## AgeC -0.266 0.149 -0.160
## CndtnC:Bln1 -0.019 -0.652 0.007 -0.026
## CondtnC:AgC 0.030 0.175 -0.018 0.024 -0.229
## Bilngl1:AgC -0.060 -0.013 0.141 -0.502 0.014 -0.007
## CndtC:B1:AC 0.014 -0.147 0.028 0.005 0.190 -0.806 0.022model_null <- lmer(MeanLogLT ~ ConditionC*AgeC + (1 + ConditionC + AgeC | Lab) + (1 | Subject), data=all_agg_subjects, REML=F)
anova(model,model_null)## Data: all_agg_subjects
## Models:
## model_null: MeanLogLT ~ ConditionC * AgeC + (1 + ConditionC + AgeC | Lab) +
## model_null: (1 | Subject)
## model: MeanLogLT ~ ConditionC * Bilingual * AgeC + (1 + ConditionC +
## model: AgeC | Lab) + (1 | Subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## model_null 12 -224.53 -160.93 124.26 -248.53
## model 16 -228.24 -143.44 130.12 -260.24 11.715 4 0.0196 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# post-hoc least-squares contrasts
lstrends(model, ~ Bilingual, var="ConditionC", adjust="none")## Bilingual ConditionC.trend SE df lower.CL upper.CL
## FALSE 0.11460620 0.02178942 23.13 0.06954487 0.1596675
## TRUE 0.04759074 0.04469517 239.75 -0.04045465 0.1356361
##
## Confidence level used: 0.95