Brown, Smith, Samara, & Wonnacott

Set Up

Load Packages

rm(list = ls(all = TRUE))
library(languageR)
suppressPackageStartupMessages(library(lattice))
suppressPackageStartupMessages(library(lme4))
library(plotrix)
suppressPackageStartupMessages(library(irr))
library(plyr)
library(knitr)
library(ggplot2)
suppressPackageStartupMessages(library(reshape))
suppressPackageStartupMessages(library(reshape2))
library(WRS2)
library(boot)

## 
## Attaching package: 'boot'

## The following object is masked from 'package:lattice':
## 
##     melanoma

library(Hmisc)

## Loading required package: survival

## 
## Attaching package: 'survival'

## The following object is masked from 'package:boot':
## 
##     aml

## Loading required package: Formula

## 
## Attaching package: 'Hmisc'

## The following objects are masked from 'package:plyr':
## 
##     is.discrete, summarize

## The following objects are masked from 'package:base':
## 
##     format.pval, round.POSIXt, trunc.POSIXt, units

Load Helper Functions

SummarySE

This function can be found on the website “Cookbook for R”

http://www.cookbook-r.com/Manipulating_data/Summarizing_data/

It summarizes data, giving count, mean, standard deviation, standard error of the mean, and confidence interval (default 95%).

• data: a data frame
• measurevar: the name of a column that contains the variable to be summarized
• groupvars: a vector containing the names of the columns that contain grouping variables
• na.rm: a boolean that indicates whether to ignore NA’s
• conf.interval: the percent range of the confidence interval (default is 95%)

summarySE <- function(data=NULL, measurevar, groupvars=NULL, na.rm=FALSE,
           conf.interval=.95, .drop=TRUE) {
  require(plyr)

  # New version of length which can handle NA's: if na.rm==T, don't count them
  length2 <- function (x, na.rm=FALSE) {
    if (na.rm) sum(!is.na(x))
    else    length(x)
  }

  # This does the summary. For each group's data frame, return a vector with
  # N, mean, and sd
  datac <- ddply(data, groupvars, .drop=.drop,
   .fun = function(xx, col) {
    c(N  = length2(xx[[col]], na.rm=na.rm),
     mean = mean  (xx[[col]], na.rm=na.rm),
     sd  = sd   (xx[[col]], na.rm=na.rm)
    )
   },
   measurevar
  )

  # Rename the "mean" column  
  datac <- rename(datac, c("mean" = measurevar))

  datac$se <- datac$sd / sqrt(datac$N) # Calculate standard error of the mean

  # Confidence interval multiplier for standard error
  # Calculate t-statistic for confidence interval: 
  # e.g., if conf.interval is .95, use .975 (above/below), and use df=N-1
  ciMult <- qt(conf.interval/2 + .5, datac$N-1)
  datac$ci <- datac$se * ciMult

  return(datac)
}

SummarySEwithin

This function can be found on the website “Cookbook for R”

http://www.cookbook-r.com/Graphs/Plotting_means_and_error_bars_(ggplot2)/#Helper functions

From that website:

Summarizes data, handling within-subjects variables by removing inter-subject variability. It will still work if there are no within-subjects variables. Gives count, un-normed mean, normed mean (with same between-group mean), standard deviation, standard error of the mean, and confidence interval. If there are within-subject variables, calculate adjusted values using method from Morey (2008).

• data: a data frame
• measurevar: the name of a column that contains the variable to be summarized
• betweenvars: a vector containing names of columns that are between-subjects variables
• withinvars: a vector containing names of columns that are within-subjects variables
• idvar: the name of a column that identifies each subject (or matched subjects)
• na.rm: a boolean that indicates whether to ignore NA’s
• conf.interval: the percent range of the confidence interval (default is 95%)

summarySEwithin <- function(data=NULL, measurevar, betweenvars=NULL, withinvars=NULL,
              idvar=NULL, na.rm=FALSE, conf.interval=.95, .drop=TRUE) {

 # Ensure that the betweenvars and withinvars are factors
 factorvars <- vapply(data[, c(betweenvars, withinvars), drop=FALSE],
  FUN=is.factor, FUN.VALUE=logical(1))

 if (!all(factorvars)) {
  nonfactorvars <- names(factorvars)[!factorvars]
  message("Automatically converting the following non-factors to factors: ",
      paste(nonfactorvars, collapse = ", "))
  data[nonfactorvars] <- lapply(data[nonfactorvars], factor)
 }

 # Get the means from the un-normed data
 datac <- summarySE(data, measurevar, groupvars=c(betweenvars, withinvars),
           na.rm=na.rm, conf.interval=conf.interval, .drop=.drop)

 # Drop all the unused columns (these will be calculated with normed data)
 datac$sd <- NULL
 datac$se <- NULL
 datac$ci <- NULL

 # Norm each subject's data
 ndata <- normDataWithin(data, idvar, measurevar, betweenvars, na.rm, .drop=.drop)

 # This is the name of the new column
 measurevar_n <- paste(measurevar, "_norm", sep="")

 # Collapse the normed data - now we can treat between and within vars the same
 ndatac <- summarySE(ndata, measurevar_n, groupvars=c(betweenvars, withinvars),
           na.rm=na.rm, conf.interval=conf.interval, .drop=.drop)

 # Apply correction from Morey (2008) to the standard error and confidence interval
 # Get the product of the number of conditions of within-S variables
 nWithinGroups  <- prod(vapply(ndatac[,withinvars, drop=FALSE], FUN=nlevels,
              FUN.VALUE=numeric(1)))
 correctionFactor <- sqrt( nWithinGroups / (nWithinGroups-1) )

 # Apply the correction factor
 ndatac$sd <- ndatac$sd * correctionFactor
 ndatac$se <- ndatac$se * correctionFactor
 ndatac$ci <- ndatac$ci * correctionFactor

 # Combine the un-normed means with the normed results
 merge(datac, ndatac)
}

normDataWithin

This function is used by the SummarySEWithin function above. It can be found on the website “Cookbook for R”

http://www.cookbook-r.com/Graphs/Plotting_means_and_error_bars_(ggplot2)/#Helper functions

From that website:

Norms the data within specified groups in a data frame; it normalizes each subject (identified by idvar) so that they have the same mean, within each group specified by betweenvars.

• data: a data frame
• idvar: the name of a column that identifies each subject (or matched subjects)
• measurevar: the name of a column that contains the variable to be summarized
• betweenvars: a vector containing names of columns that are between-subjects variables
• na.rm: a boolean that indicates whether to ignore NA’s

normDataWithin <- function(data=NULL, idvar, measurevar, betweenvars=NULL,
              na.rm=FALSE, .drop=TRUE) {
  #library(plyr)

  # Measure var on left, idvar + between vars on right of formula.
  data.subjMean <- ddply(data, c(idvar, betweenvars), .drop=.drop,
   .fun = function(xx, col, na.rm) {
    c(subjMean = mean(xx[,col], na.rm=na.rm))
   },
   measurevar,
   na.rm
  )

  # Put the subject means with original data
  data <- merge(data, data.subjMean)

  # Get the normalized data in a new column
  measureNormedVar <- paste(measurevar, "_norm", sep="")
  data[,measureNormedVar] <- data[,measurevar] - data[,"subjMean"] +
                mean(data[,measurevar], na.rm=na.rm)

  # Remove this subject mean column
  data$subjMean <- NULL

  return(data)
}

myCenter

This function outputs the centered values of a variable, which can be a numeric variable, a factor, or a data frame. It was taken from Florian Jaegers blog.

https://hlplab.wordpress.com/2009/04/27/centering-several-variables/.

From his blog:

• If the input is a numeric variable, the output is the centered variable.
• If the input is a factor, the output is a numeric variable with centered factor level values. That is, the factor’s levels are converted into numerical values in their inherent order (if not specified otherwise, R defaults to alphanumerical order). More specifically, this centers any binary factor so that the value below 0 will be the 1st level of the original factor, and the value above 0 will be the 2nd level.
• If the input is a data frame or matrix, the output is a new matrix of the same dimension and with the centered values and column names that correspond to the colnames() of the input preceded by “c” (e.g. “Variable1” will be “cVariable1”).

myCenter= function(x) {
 if (is.numeric(x)) { return(x - mean(x, na.rm=T)) }
    if (is.factor(x)) {
        x= as.numeric(x)
        return(x - mean(x, na.rm=T))
    }
    if (is.data.frame(x) || is.matrix(x)) {
        m = matrix(nrow=nrow(x), ncol = ncol(x))
        colnames(m) = paste("c", colnames(x), sep="")
    
        for (i in 1:ncol(x)) {
        
            m[,i]= myCenter(x[,i])
        }
        return(as.data.frame(m))
    }
}

lizCenter

This function provides a wrapper around myCenter allowing you to center a specific list of variables from a data frame.

• x: data frame
• listfname: a list of the variables to be centered (e.g. list(variable1,variable2))

The output is a copy of the data frame with a column (always a numeric variable) added for each of the centered variables. These columns are labelled with the column’s previous name, but with “.ct” appended (e.g., “variable1” will become “variable1.ct”).

lizCenter= function(x, listfname) 
{
    for (i in 1:length(listfname)) 
    {
        fname = as.character(listfname[i])
        x[paste(fname,".ct", sep = "")] = myCenter(x[fname])
    }
        
    return(x)
}

lizContrasts

This function can be used to create two centered dummy variables which stand in place of a three-way factor (e.g., semantic consistency). This allows us to inspect each contrast separately, as well as their interactions with other factors. Other fixed effects in the model can be evaluated as the average effects across all levels of the factor.

The function takes a data frame (d), a factor from that data frame (condition), which must have three levels, and the name of the level of the factor which is to be used as the baseline for the contrasts (baselevel).

• d: data frame
• condition: three-way factor
• baselevel: name of the level of the factor that is to be used as the baseline for contrasts

lizContrasts= function(d, condition, baselevel) 
{
    condition = factor(condition)
 condition = relevel(condition, baselevel)

    a= (contrasts(condition)-apply(contrasts(condition),2,mean))
    d$dummy1[condition== rownames(a)[1]] <- a[1] 
    d$dummy1[condition== rownames(a)[2]] <- a[2] 
    d$dummy1[condition== rownames(a)[3]] <- a[3] 
    d$dummy2[condition== rownames(a)[1]] <- a[4] 
    d$dummy2[condition== rownames(a)[2]] <- a[5] 
    d$dummy2[condition== rownames(a)[3]] <- a[6] 

    name1 = paste(baselevel, rownames(a)[2],sep="_VS_")
    name2 = paste(baselevel, rownames(a)[3],sep="_VS_")

    d[name1] = d$dummy1 
    d[name2] = d$dummy2 

    d$dummy1 <-NULL 
    d$dummy2 <-NULL 
    
    return(d)
}

get_coeffs

This function allows us to inspect particular coefficients from the output of an lme model by putting them in table.

• x: the output returned when running lmer or glmer (i.e. an object of type lmerMod or glmerMod)
• list: a list of the names of the coefficients to be extracted (e.g. c(“variable1”, “variable1:variable2”))

get_coeffs <- function(x,list){(kable(as.data.frame(summary(x)$coefficients)[list,],digits=3))}

Set up Theme for Figures

mytheme <- theme(axis.line = element_line(colour = "grey"),
        text = element_text(size=12),
        panel.background = element_blank(),
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank(),
        strip.text.x = element_text(size=12, face="bold"),
        strip.background = element_rect(fill="white", colour = "white"))

Load Datasets

prod = read.csv("production.csv")
afc = read.csv("afc.csv")

Production Test: Accuracy

prod$session = as.factor(prod$session)
prod$awareness = as.factor(prod$awareness)

Data Preparation

Inter-Rater Reliability for Coding

Use Cohen’s Kappa (used for nominal/categorical variables)

• 1 = perfect agreement
• 0 = random agreement
• -1 = perfect disagreement

Landis & Koch (1977)

• 0.0-0.2 = slight agreement
• 0.21-0.40 = fair agreement
• 0.41-0.6 = moderate agreement
• 0.61-0.8 = substantial agreement
• 0.81-1.0 = perfect/near perfect agreement

Krippendorff (1980)

• 0.67-0.8 = allows tentative conclusions
• 0.81 or higher = allows definite conclusions

To compute Cohen’s Kappa we need to compare detHB and detAS columns for trials where there was a single phoneme error

prod$detHB = as.character(prod$detHB)
prod$detAS = as.character(prod$detAS)

# First select only items that were originally coded as "other" responses using a strict (fully correct vs. incorrect) criterion
IRR <- droplevels(subset(prod, det_used_logical=="other"))

# Then look at agreement between re-coding of these items between HB and AS
IRR <- cbind(IRR$detHB, IRR$detAS)
kappa2(IRR, weight = c("unweighted", "equal", "squared"), sort.levels = FALSE)

##  Cohen's Kappa for 2 Raters (Weights: unweighted)
## 
##  Subjects = 2478 
##    Raters = 2 
##     Kappa = 0.993 
## 
##         z = 65.6 
##   p-value = 0

Remove Trials without a Det1/Det2 Response

First remove trials where the noun produced was incorrect

prod1 <- droplevels(subset(prod, noun_correct=="1"))

Then remove trials in which the participant either did not produce a sentence-final particle or produced something other than one of the two trained particles

prod1 <- droplevels(subset(prod1, detHB!="none"))
prod1 <- droplevels(subset(prod1, detHB!="other"))

Create Separate Dataframes for Children and Adults

prodCHILD <- droplevels(subset(prod1, age=="Child"))
prodADULT <- droplevels(subset(prod1, age=="Adult"))

Trained Nouns

Figure 1

All Conditions

PRODOLD <- droplevels(subset(prod1, oldnew=="old"))

# Aggregate to obtain means per participant/day etc.
aggregated.PRODOLD = aggregate(det_correct ~ participant + session_fig + consistency_fig + typefreq + age, PRODOLD, FUN=mean)

# Re-order levels within the semantic consistency variable (the default is alphabetical: 1 = fully consistent, 2 = inconsistent, 3 = partially consistent; we want to swap the order of inconsistent/partially consistent)
aggregated.PRODOLD$consistency_fig = factor(aggregated.PRODOLD$consistency_fig, levels(aggregated.PRODOLD$consistency_fig)[c(1,3,2)])
levels(aggregated.PRODOLD$consistency_fig)

## [1] "Fully Consistent"     "Partially Consistent" "Inconsistent"

# Re-order levels within the age variable (the default is alphabetical: 1 = adult, 2 = child, we want to reverse this order)
aggregated.PRODOLD$age = factor(aggregated.PRODOLD$age,
levels(aggregated.PRODOLD$age)[c(2,1)])
levels(aggregated.PRODOLD$age)

## [1] "Child" "Adult"

p = ggplot(aggregated.PRODOLD, aes(x = consistency_fig, y = det_correct, fill = typefreq))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white"), name = "Type Frequency", labels = c("High", "Low"))
p = p + geom_hline(aes(yintercept=0.50),linetype="dashed") 
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(~ age + session_fig)
p = p + theme_bw()
p = p + mytheme
p = p + theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
print(p)

ggsave("Figure1_top_panel.jpg", width = 8, height = 5, dpi = 120)

Partially Consistent Condition Only

Compare exception/majority-particle nouns within the partially consistent language.

prodOLDP <- droplevels(subset(PRODOLD, consistency_fig == "Partially Consistent"))

# Aggregate to obtain means per participant/day etc.
aggregated.prodOLDP = aggregate(det_correct ~ participant + session_fig + exception + typefreq + age, prodOLDP, FUN = mean)

# Check the order of levels within the exception variable
levels(aggregated.prodOLDP$exception)

## [1] "0" "1"

# 0 = majority-particle, 1 = exception

# Re-order levels within the age variable (the default is alphabetical: 1 = adult, 2 = child, we want to reverse this order)
aggregated.prodOLDP$age = factor(aggregated.prodOLDP$age, levels(aggregated.prodOLDP$age)[c(2,1)])
levels(aggregated.prodOLDP$age)

## [1] "Child" "Adult"

# Now plot with separate panels for Session 1 and Session 4 for each age group, split by type frequency, then by noun-type (majority-particle vs. exception)
p = ggplot(aggregated.prodOLDP, aes(x = typefreq, y = det_correct, fill = exception))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white"), name = "Noun-Type", labels = c("Majority-Particle", "Exception"))
p = p + geom_hline(aes(yintercept=0.50),linetype="dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(~ age + session_fig)
# p = p + facet_grid(~ day)
p = p + theme_bw()
p = p + mytheme
print(p)

ggsave("Figure1_bottom_panel.jpg", width = 8, height = 4, dpi = 120)

Statistical Analyses

Children

All Conditions

First run a model in which the fully consistent condition (“consistent”) is set as the baseline for contrasts (i.e., the model compares the fully consistent condition to both the partially consistent condition (“Consistent_VS_Partial”) and the inconsistent condition (“Consistent_VS_Inconsistent”))

prodOLDchild <- droplevels(subset(prodCHILD, oldnew=="old"))
prodOLDchild <- lizCenter(prodOLDchild, list("typefreq", "session"))
prodOLDchild <- lizContrasts(prodOLDchild, prodOLDchild$consistency, "Consistent")

prodOLDchild.lmer1 = glmer(det_correct ~
                          + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * typefreq.ct * session.ct
                          + (1 + session.ct|participant),
                          data = prodOLDchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodOLDchild.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.909	0.090	10.132	0.000
Consistent_VS_Partial	-0.666	0.219	-3.047	0.002
Consistent_VS_Inconsistent	-0.756	0.219	-3.451	0.001
typefreq.ct	0.291	0.176	1.649	0.099
session.ct	1.115	0.106	10.497	0.000
Consistent_VS_Partial:typefreq.ct	-0.324	0.435	-0.746	0.456
Consistent_VS_Inconsistent:typefreq.ct	0.037	0.436	0.085	0.933
Consistent_VS_Partial:session.ct	-0.898	0.256	-3.515	0.000
Consistent_VS_Inconsistent:session.ct	-0.862	0.258	-3.345	0.001
typefreq.ct:session.ct	0.371	0.204	1.819	0.069
Consistent_VS_Partial:typefreq.ct:session.ct	-0.616	0.503	-1.225	0.221
Consistent_VS_Inconsistent:typefreq.ct:session.ct	0.359	0.507	0.708	0.479

We also need to compare the partially consistent and inconsistent conditions (“Inconsistent_VS_Partial”). To do this, rerun the model, with the inconsistent condition as the baseline. (The anova function is used to establish that this is the same model; only relevant parts of the output are printed)

prodOLDchild <- lizContrasts(prodOLDchild, prodOLDchild$consistency, "Inconsistent")

prodOLDchild.lmer2 = glmer(det_correct ~
                           + (Inconsistent_VS_Partial + Inconsistent_VS_Consistent) * typefreq.ct * session.ct
                           + (1 + session.ct|participant),
                           data = prodOLDchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodOLDchild.lmer1, prodOLDchild.lmer2)

## Data: prodOLDchild
## Models:
## prodOLDchild.lmer1: det_correct ~ +(Consistent_VS_Partial + Consistent_VS_Inconsistent) * 
## prodOLDchild.lmer1:     typefreq.ct * session.ct + (1 + session.ct | participant)
## prodOLDchild.lmer2: det_correct ~ +(Inconsistent_VS_Partial + Inconsistent_VS_Consistent) * 
## prodOLDchild.lmer2:     typefreq.ct * session.ct + (1 + session.ct | participant)
##                    Df    AIC    BIC  logLik deviance Chisq Chi Df
## prodOLDchild.lmer1 15 5779.2 5876.7 -2874.6   5749.2             
## prodOLDchild.lmer2 15 5779.2 5876.7 -2874.6   5749.2     0      0
##                    Pr(>Chisq)
## prodOLDchild.lmer1           
## prodOLDchild.lmer2          1

get_coeffs(prodOLDchild.lmer2, c("Inconsistent_VS_Partial", "Inconsistent_VS_Partial:typefreq.ct", "Inconsistent_VS_Partial:session.ct", "Inconsistent_VS_Partial:typefreq.ct:session.ct"))

	Estimate	Std. Error	z value	Pr(>|z|)
Inconsistent_VS_Partial	0.090	0.212	0.427	0.669
Inconsistent_VS_Partial:typefreq.ct	-0.361	0.424	-0.853	0.394
Inconsistent_VS_Partial:session.ct	-0.036	0.239	-0.153	0.879
Inconsistent_VS_Partial:typefreq.ct:session.ct	-0.974	0.478	-2.040	0.041

There is an effect of session in the first model. Check whether performance above chance in both sessions?

prodOLDchild <- lizContrasts(prodOLDchild, prodOLDchild$consistency, "Consistent")

prodOLDchild.lmer1a = glmer(det_correct ~
                            - 1 
                            + session
                            + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * typefreq.ct * session.ct
                            - session.ct
                            + (1 + session.ct|participant),
                            data = prodOLDchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodOLDchild.lmer1, prodOLDchild.lmer1a)

## Data: prodOLDchild
## Models:
## prodOLDchild.lmer1: det_correct ~ +(Consistent_VS_Partial + Consistent_VS_Inconsistent) * 
## prodOLDchild.lmer1:     typefreq.ct * session.ct + (1 + session.ct | participant)
## prodOLDchild.lmer1a: det_correct ~ -1 + session + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * 
## prodOLDchild.lmer1a:     typefreq.ct * session.ct - session.ct + (1 + session.ct | 
## prodOLDchild.lmer1a:     participant)
##                     Df    AIC    BIC  logLik deviance Chisq Chi Df
## prodOLDchild.lmer1  15 5779.2 5876.7 -2874.6   5749.2             
## prodOLDchild.lmer1a 15 5779.2 5876.7 -2874.6   5749.2     0      0
##                     Pr(>Chisq)    
## prodOLDchild.lmer1                
## prodOLDchild.lmer1a  < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

get_coeffs(prodOLDchild.lmer1a, c("session1", "session4"))

	Estimate	Std. Error	z value	Pr(>|z|)
session1	0.287	0.065	4.391	0
session4	1.403	0.127	11.077	0

The first model also showed that session interacted with both Consistent_VS_Inconsistent and Consistent_VS_Partial. We break down each of these interacts by running the model again (with “Consistent” as the baseline level for contrasts) but with (i) the main effect of each contrast and (ii) the interaction of each contrast with session removed, replaced with the effect of each contrast for each level of session (the anova function is used to show that this is equivalent to the original model).

prodOLDchild <- lizContrasts(prodOLDchild, prodOLDchild$consistency, "Consistent")

prodOLDchild.lmer1b = glmer(det_correct ~
                            + Consistent_VS_Partial:session
                            + Consistent_VS_Inconsistent:session
                            + typefreq.ct * session.ct
                            + (Consistent_VS_Partial + Consistent_VS_Inconsistent):typefreq.ct
                            + (Consistent_VS_Partial + Consistent_VS_Inconsistent):typefreq.ct:session.ct
                            + (1 + session.ct|participant),
                            data = prodOLDchild, family=binomial, control=glmerControl(optimizer = "bobyqa"))
anova(prodOLDchild.lmer1,prodOLDchild.lmer1b)

## Data: prodOLDchild
## Models:
## prodOLDchild.lmer1: det_correct ~ +(Consistent_VS_Partial + Consistent_VS_Inconsistent) * 
## prodOLDchild.lmer1:     typefreq.ct * session.ct + (1 + session.ct | participant)
## prodOLDchild.lmer1b: det_correct ~ +Consistent_VS_Partial:session + Consistent_VS_Inconsistent:session + 
## prodOLDchild.lmer1b:     typefreq.ct * session.ct + (Consistent_VS_Partial + Consistent_VS_Inconsistent):typefreq.ct + 
## prodOLDchild.lmer1b:     (Consistent_VS_Partial + Consistent_VS_Inconsistent):typefreq.ct:session.ct + 
## prodOLDchild.lmer1b:     (1 + session.ct | participant)
##                     Df    AIC    BIC  logLik deviance Chisq Chi Df
## prodOLDchild.lmer1  15 5779.2 5876.7 -2874.6   5749.2             
## prodOLDchild.lmer1b 15 5779.2 5876.7 -2874.6   5749.2     0      0
##                     Pr(>Chisq)    
## prodOLDchild.lmer1                
## prodOLDchild.lmer1b  < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

get_coeffs(prodOLDchild.lmer1b, c("Consistent_VS_Partial:session1", "Consistent_VS_Partial:session4", "session1:Consistent_VS_Inconsistent", "session4:Consistent_VS_Inconsistent"))

	Estimate	Std. Error	z value	Pr(>|z|)
Consistent_VS_Partial:session1	-0.165	0.159	-1.039	0.299
Consistent_VS_Partial:session4	-1.063	0.307	-3.459	0.001
session1:Consistent_VS_Inconsistent	-0.276	0.162	-1.706	0.088
session4:Consistent_VS_Inconsistent	-1.138	0.308	-3.696	0.000

Partially Consistent Condition Only

Do children differ in their performance with majority-particle and exception items in the partially consistent condition? To look at this run a new model just looking at the partially consistent condition, with exception (1 = exception noun; 0 = majority-particle noun) as a fixed effect, as well as interactions with the other factors.

prodOLDchildP <- droplevels(subset(prodOLDchild, consistency == "Partial"))
prodOLDchildP <- lizCenter(prodOLDchildP, list("typefreq", "session", "exception"))

prodOLDchildP.lmer = glmer(det_correct ~ 
                          + typefreq.ct * session.ct * exception.ct
                          + (1 + session.ct * exception.ct|participant), 
                          data = prodOLDchildP, family = binomial, control = glmerControl(optimizer = "bobyqa"))

get_coeffs(prodOLDchildP.lmer, c("exception.ct", "typefreq.ct:exception.ct", "session.ct:exception.ct", "typefreq.ct:session.ct:exception.ct"))

	Estimate	Std. Error	z value	Pr(>|z|)
exception.ct	0.003	0.183	0.014	0.989
typefreq.ct:exception.ct	0.405	0.354	1.145	0.252
session.ct:exception.ct	0.022	0.370	0.058	0.953
typefreq.ct:session.ct:exception.ct	-0.102	0.722	-0.141	0.888

There is no evidence that accuracy differs for exception and majority-particle nouns in the child data.

Adults

All Conditions

First run a model in which the fully consistent condition (“consistent”) is set as the baseline for contrasts (i.e., the model compares the fully consistent condition to both the partially consistent condition (“Consistent_VS_Partial”) and the inconsistent condition (“Consistent_VS_Inconsistent”)).

prodOLDadult <- droplevels(subset(prodADULT, oldnew=="old"))
prodOLDadult <- lizCenter(prodOLDadult, list("typefreq", "session"))
prodOLDadult <- lizContrasts(prodOLDadult, prodOLDadult$consistency, "Consistent")

prodOLDadult.lmer1 = glmer(det_correct ~
                          + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * typefreq.ct * session.ct
                          + (1 + session.ct|participant),
                          data = prodOLDadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodOLDadult.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	3.843	0.379	10.133	0.000
Consistent_VS_Partial	-2.460	0.761	-3.234	0.001
Consistent_VS_Inconsistent	-2.170	0.774	-2.803	0.005
typefreq.ct	1.193	0.585	2.038	0.042
session.ct	3.123	0.625	4.997	0.000
Consistent_VS_Partial:typefreq.ct	-0.413	1.513	-0.273	0.785
Consistent_VS_Inconsistent:typefreq.ct	0.943	1.554	0.606	0.544
Consistent_VS_Partial:session.ct	-0.997	1.101	-0.905	0.365
Consistent_VS_Inconsistent:session.ct	0.412	1.136	0.363	0.717
typefreq.ct:session.ct	-0.709	0.799	-0.887	0.375
Consistent_VS_Partial:typefreq.ct:session.ct	2.140	2.120	1.009	0.313
Consistent_VS_Inconsistent:typefreq.ct:session.ct	3.354	2.260	1.484	0.138

We also need to compare the partially consistent and inconsistent conditions (“Inconsistent_VS_Partial”). To do this, rerun the model with the inconsistent condition as the baseline.

prodOLDadult <- lizContrasts(prodOLDadult, prodOLDadult$consistency, "Inconsistent")

prodOLDadult.lmer2 = glmer(det_correct ~
                          + (Inconsistent_VS_Partial + Inconsistent_VS_Consistent) * typefreq.ct * session.ct
                          + (1 + session.ct|participant),
                          data = prodOLDadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodOLDadult.lmer1, prodOLDadult.lmer2)

## Data: prodOLDadult
## Models:
## prodOLDadult.lmer1: det_correct ~ +(Consistent_VS_Partial + Consistent_VS_Inconsistent) * 
## prodOLDadult.lmer1:     typefreq.ct * session.ct + (1 + session.ct | participant)
## prodOLDadult.lmer2: det_correct ~ +(Inconsistent_VS_Partial + Inconsistent_VS_Consistent) * 
## prodOLDadult.lmer2:     typefreq.ct * session.ct + (1 + session.ct | participant)
##                    Df    AIC    BIC  logLik deviance Chisq Chi Df
## prodOLDadult.lmer1 15 2082.1 2175.5 -1026.1   2052.1             
## prodOLDadult.lmer2 15 2082.1 2175.5 -1026.1   2052.1     0      0
##                    Pr(>Chisq)
## prodOLDadult.lmer1           
## prodOLDadult.lmer2          1

get_coeffs(prodOLDadult.lmer2, c("Inconsistent_VS_Partial", "Inconsistent_VS_Partial:typefreq.ct", "Inconsistent_VS_Partial:session","Inconsistent_VS_Partial:typefreq.ct:session.ct" ))

	Estimate	Std. Error	z value	Pr(>|z|)
Inconsistent_VS_Partial	-0.290	0.667	-0.435	0.663
Inconsistent_VS_Partial:typefreq.ct	-1.355	1.323	-1.024	0.306
Inconsistent_VS_Partial:session.ct	-1.409	0.859	-1.641	0.101
Inconsistent_VS_Partial:typefreq.ct:session.ct	-1.214	1.708	-0.711	0.477

There were main effects of both session and type frequency. Is performance above chance in each session for each level of type frequency?

prodOLDadult <- lizContrasts(prodOLDadult, prodOLDadult$consistency, "Consistent")

prodOLDadult.lmer2b = glmer(det_correct ~
                          - 1
                          + typefreq:session
                          + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * typefreq.ct * session.ct
                          - typefreq.ct * session.ct
                          + (1 + session.ct|participant),
                          data = prodOLDadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodOLDadult.lmer1, prodOLDadult.lmer2b)

## Data: prodOLDadult
## Models:
## prodOLDadult.lmer1: det_correct ~ +(Consistent_VS_Partial + Consistent_VS_Inconsistent) * 
## prodOLDadult.lmer1:     typefreq.ct * session.ct + (1 + session.ct | participant)
## prodOLDadult.lmer2b: det_correct ~ -1 + typefreq:session + (Consistent_VS_Partial + 
## prodOLDadult.lmer2b:     Consistent_VS_Inconsistent) * typefreq.ct * session.ct - 
## prodOLDadult.lmer2b:     typefreq.ct * session.ct + (1 + session.ct | participant)
##                     Df    AIC    BIC  logLik deviance Chisq Chi Df
## prodOLDadult.lmer1  15 2082.1 2175.5 -1026.1   2052.1             
## prodOLDadult.lmer2b 15 2082.1 2175.5 -1026.1   2052.1     0      0
##                     Pr(>Chisq)    
## prodOLDadult.lmer1                
## prodOLDadult.lmer2b  < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

get_coeffs(prodOLDadult.lmer2b, c("typefreqHigh:session1", "typefreqHigh:session4", "typefreqLow:session1", "typefreqLow:session4" ))

	Estimate	Std. Error	z value	Pr(>|z|)
typefreqHigh:session1	1.470	0.262	5.609	0
typefreqHigh:session4	4.943	0.777	6.361	0
typefreqLow:session1	3.027	0.322	9.385	0
typefreqLow:session4	5.791	0.817	7.092	0

Looking at session only - is performance above chance in each session?

prodOLDadult.lmer2c = glmer(det_correct ~
                            - 1 
                            + session
                            + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * typefreq.ct * session.ct
                            - session.ct
                            + (1 + session.ct|participant),
                            data = prodOLDadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodOLDadult.lmer1, prodOLDadult.lmer2c)

## Data: prodOLDadult
## Models:
## prodOLDadult.lmer1: det_correct ~ +(Consistent_VS_Partial + Consistent_VS_Inconsistent) * 
## prodOLDadult.lmer1:     typefreq.ct * session.ct + (1 + session.ct | participant)
## prodOLDadult.lmer2c: det_correct ~ -1 + session + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * 
## prodOLDadult.lmer2c:     typefreq.ct * session.ct - session.ct + (1 + session.ct | 
## prodOLDadult.lmer2c:     participant)
##                     Df    AIC    BIC  logLik deviance Chisq Chi Df
## prodOLDadult.lmer1  15 2082.1 2175.5 -1026.1   2052.1             
## prodOLDadult.lmer2c 15 2082.1 2175.5 -1026.1   2052.1     0      0
##                     Pr(>Chisq)
## prodOLDadult.lmer1            
## prodOLDadult.lmer2c          1

get_coeffs(prodOLDadult.lmer2c, c("session1", "session4"))

	Estimate	Std. Error	z value	Pr(>|z|)
session1	2.240	0.214	10.485	0
session4	5.363	0.654	8.196	0

Partially Consistent Condition Only

Do adults differ in their performance with majority-particle and exception items in the partially consistent condition? To look at this run a new model just looking at the partially consistent condition, with exception (1 = exception noun; 0 = majority-particle noun) as a fixed effect, as well as interactions with the other factors.

prodOLDadultP <- droplevels(subset(prodOLDadult, consistency == "Partial"))
prodOLDadultP <- lizCenter(prodOLDadultP, list("typefreq", "session", "exception"))


prodOLDadultP.lmer = glmer(det_correct ~ 
                          + typefreq.ct * session.ct * exception.ct
                          + (1 + session.ct * exception.ct|participant), 
                          data = prodOLDadultP, family = binomial, control = glmerControl(optimizer = "bobyqa"))

get_coeffs(prodOLDadultP.lmer, c("exception.ct", "typefreq.ct:exception.ct", "session.ct:exception.ct", "typefreq.ct:session.ct:exception.ct"))

	Estimate	Std. Error	z value	Pr(>|z|)
exception.ct	-0.930	0.660	-1.409	0.159
typefreq.ct:exception.ct	2.340	0.611	3.829	0.000
session.ct:exception.ct	1.714	1.392	1.231	0.218
typefreq.ct:session.ct:exception.ct	-2.353	1.742	-1.350	0.177

The main effect of noun-type is non-significant. There is however a significant interaction between noun-type and type-frequency: Figure 1 (bottom panel) suggest more errors with exception noun when there are more majority-particle nouns (i.e., in the high type frequency condition).

prodOLDadultP.lmer.b = glmer(det_correct ~ 
                             + typefreq:exception.ct
                             + session.ct * typefreq.ct
                             + session.ct:exception.ct
                             + typefreq.ct:session.ct:exception.ct
                             + (1 + session.ct * exception.ct|participant), 
                             data = prodOLDadultP, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodOLDadultP.lmer, prodOLDadultP.lmer.b)

## Data: prodOLDadultP
## Models:
## prodOLDadultP.lmer: det_correct ~ +typefreq.ct * session.ct * exception.ct + (1 + 
## prodOLDadultP.lmer:     session.ct * exception.ct | participant)
## prodOLDadultP.lmer.b: det_correct ~ +typefreq:exception.ct + session.ct * typefreq.ct + 
## prodOLDadultP.lmer.b:     session.ct:exception.ct + typefreq.ct:session.ct:exception.ct + 
## prodOLDadultP.lmer.b:     (1 + session.ct * exception.ct | participant)
##                      Df    AIC    BIC  logLik deviance Chisq Chi Df
## prodOLDadultP.lmer   18 771.78 863.41 -367.89   735.78             
## prodOLDadultP.lmer.b 18 771.78 863.41 -367.89   735.78     0      0
##                      Pr(>Chisq)    
## prodOLDadultP.lmer                 
## prodOLDadultP.lmer.b  < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

get_coeffs(prodOLDadultP.lmer.b, c("typefreqHigh:exception.ct","typefreqLow:exception.ct"))

	Estimate	Std. Error	z value	Pr(>|z|)
typefreqHigh:exception.ct	-2.066	0.699	-2.956	0.003
typefreqLow:exception.ct	0.274	0.757	0.362	0.717

Novel Nouns

Figure 2

PRODNEW <- droplevels(subset(prod1, oldnew =="new" & consistency!="Inconsistent"))

# Aggregate to obtain means per participant/day etc.
aggregated.PRODNEW = aggregate(det_correct ~ participant + session_fig + consistency_fig + typefreq + age, PRODNEW, FUN=mean)

#re-order levels of the semantic consistency variable
aggregated.PRODNEW$consistency_fig = factor(aggregated.PRODNEW$consistency_fig, levels(aggregated.PRODNEW$consistency_fig)[c(1,3,2)])
levels(aggregated.PRODNEW$consistency_fig)

## [1] "Fully Consistent"     "Partially Consistent"

# Re-order levels within the age variable (the default is alphabetical: 1 = adult, 2 = child, we want to reverse this order)
aggregated.PRODNEW$age = factor(aggregated.PRODNEW$age, levels(aggregated.PRODNEW$age)[c(2,1)])
levels(aggregated.PRODNEW$age)

## [1] "Child" "Adult"

# Now plot with separate panels for Session 1 and Session 4 for each age group, split by semantic consistency, then by type frequency
p = ggplot(aggregated.PRODNEW, aes(x = consistency_fig, y = det_correct, fill = typefreq))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white"), name = "Type Frequency", labels = c("High", "Low"))
p = p + geom_hline(aes(yintercept = 0.50), linetype = "dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(~ age + session_fig)
p = p + theme_bw()
p = p + mytheme
p = p + theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
print(p)

ggsave("Figure2.jpg", width = 8, height = 6, dpi = 120)

Statistical Analyses

For novel nouns we only include data from the fully consistent and partially consistent conditions. The inconsistent condition cannot be included since there is no “correct” (or majority) particle for semantic category, so no way to score accuracy of particle usage for novel nouns.

Children

prodNEWchild <- droplevels(subset(prod1, age == "Child" & oldnew == "new" & consistency != "Inconsistent"))
prodNEWchild <- lizCenter(prodNEWchild, list("consistency", "typefreq", "session"))

prodNEWchild.lmer1 = glmer(det_correct ~
                          + consistency.ct * typefreq.ct * session.ct
                          + (1 + session.ct|participant),
                          data = prodNEWchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodNEWchild.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.360	0.105	3.425	0.001
consistency.ct	-0.663	0.210	-3.155	0.002
typefreq.ct	-0.135	0.209	-0.644	0.519
session.ct	0.388	0.093	4.146	0.000
consistency.ct:typefreq.ct	-0.377	0.419	-0.899	0.369
consistency.ct:session.ct	-0.504	0.186	-2.706	0.007
typefreq.ct:session.ct	-0.127	0.182	-0.700	0.484
consistency.ct:typefreq.ct:session.ct	0.234	0.365	0.640	0.522

Break down the interaction between semantic consistency and session. Figure 2 suggests that the improvement in performance between sessions is greater in the fully consistent condition.

prodNEWchild.lmer2 = glmer(det_correct ~
                          + session.ct:consistency
                          + (consistency.ct * typefreq.ct * session.ct)
                          - session.ct
                          - session.ct:consistency.ct
                          + (1 + session.ct|participant),
                          data = prodNEWchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodNEWchild.lmer1, prodNEWchild.lmer2)

## Data: prodNEWchild
## Models:
## prodNEWchild.lmer1: det_correct ~ +consistency.ct * typefreq.ct * session.ct + (1 + 
## prodNEWchild.lmer1:     session.ct | participant)
## prodNEWchild.lmer2: det_correct ~ +session.ct:consistency + (consistency.ct * typefreq.ct * 
## prodNEWchild.lmer2:     session.ct) - session.ct - session.ct:consistency.ct + (1 + 
## prodNEWchild.lmer2:     session.ct | participant)
##                    Df  AIC    BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## prodNEWchild.lmer1 11 4162 4228.8  -2070     4140                        
## prodNEWchild.lmer2 11 4162 4228.8  -2070     4140     0      0          1

get_coeffs(prodNEWchild.lmer2, c("session.ct:consistencyConsistent", "session.ct:consistencyPartial"))

	Estimate	Std. Error	z value	Pr(>|z|)
session.ct:consistencyConsistent	0.646	0.141	4.597	0.000
session.ct:consistencyPartial	0.142	0.123	1.149	0.251

Are they above chance in each condition in each session? Re-fit the model removing the intercept, main effects of session and consistency, and session by consistency interaction, adding in an intercept for each session in each condition.

prodNEWchild.lmer3 = glmer(det_correct ~
                          - 1               
                          + session: consistency
                          + (consistency.ct * typefreq.ct * session.ct)
                          - session.ct * consistency.ct
                          + (1 + session.ct|participant),
                          data = prodNEWchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodNEWchild.lmer1, prodNEWchild.lmer3)

## Data: prodNEWchild
## Models:
## prodNEWchild.lmer1: det_correct ~ +consistency.ct * typefreq.ct * session.ct + (1 + 
## prodNEWchild.lmer1:     session.ct | participant)
## prodNEWchild.lmer3: det_correct ~ -1 + session:consistency + (consistency.ct * typefreq.ct * 
## prodNEWchild.lmer3:     session.ct) - session.ct * consistency.ct + (1 + session.ct | 
## prodNEWchild.lmer3:     participant)
##                    Df  AIC    BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## prodNEWchild.lmer1 11 4162 4228.8  -2070     4140                        
## prodNEWchild.lmer3 11 4162 4228.8  -2070     4140     0      0  < 2.2e-16
##                       
## prodNEWchild.lmer1    
## prodNEWchild.lmer3 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

get_coeffs(prodNEWchild.lmer3, c("session1:consistencyConsistent", "session4:consistencyConsistent", "session1:consistencyPartial", "session4:consistencyPartial"))

	Estimate	Std. Error	z value	Pr(>|z|)
session1:consistencyConsistent	0.345	0.130	2.663	0.008
session4:consistencyConsistent	0.991	0.192	5.154	0.000
session1:consistencyPartial	-0.041	0.125	-0.326	0.745
session4:consistencyPartial	0.101	0.181	0.559	0.576

Adults

prodNEWadult <- droplevels(subset(prod1, age == "Adult" & oldnew == "new" & consistency != "Inconsistent"))
prodNEWadult <- lizCenter(prodNEWadult, list("consistency", "typefreq", "session"))

prodNEWadult.lmer1 = glmer(det_correct ~
                           + consistency.ct * typefreq.ct * session.ct
                           + (1 + session.ct|participant),
                           data = prodNEWadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodNEWadult.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	4.311	0.633	6.805	0.000
consistency.ct	-3.443	1.029	-3.345	0.001
typefreq.ct	0.277	1.009	0.275	0.784
session.ct	2.216	1.022	2.167	0.030
consistency.ct:typefreq.ct	-1.622	1.980	-0.819	0.413
consistency.ct:session.ct	-1.892	1.427	-1.325	0.185
typefreq.ct:session.ct	-1.619	1.350	-1.199	0.230
consistency.ct:typefreq.ct:session.ct	3.926	2.591	1.515	0.130

Are they above chance in each condition in each session? Re-fit the model removing the intercept, main effects of session and consistency, and session by consistency interaction, adding in an intercept for each session in each condition.

prodNEWadult.lmer2 = glmer(det_correct ~
                          - 1               
                          + session:consistency
                          + (consistency.ct * typefreq.ct * session.ct)
                          - session.ct * consistency.ct
                          + (1 + session.ct|participant),
                          data = prodNEWadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodNEWadult.lmer1, prodNEWadult.lmer2)

## Data: prodNEWadult
## Models:
## prodNEWadult.lmer1: det_correct ~ +consistency.ct * typefreq.ct * session.ct + (1 + 
## prodNEWadult.lmer1:     session.ct | participant)
## prodNEWadult.lmer2: det_correct ~ -1 + session:consistency + (consistency.ct * typefreq.ct * 
## prodNEWadult.lmer2:     session.ct) - session.ct * consistency.ct + (1 + session.ct | 
## prodNEWadult.lmer2:     participant)
##                    Df    AIC    BIC  logLik deviance Chisq Chi Df
## prodNEWadult.lmer1 11 1327.7 1391.5 -652.85   1305.7             
## prodNEWadult.lmer2 11 1327.7 1391.5 -652.85   1305.7     0      0
##                    Pr(>Chisq)
## prodNEWadult.lmer1           
## prodNEWadult.lmer2          1

get_coeffs(prodNEWadult.lmer2, c("session1:consistencyConsistent", "session4:consistencyConsistent", "session1:consistencyPartial", "session4:consistencyPartial"))

	Estimate	Std. Error	z value	Pr(>|z|)
session1:consistencyConsistent	4.354	0.833	5.227	0.000
session4:consistencyConsistent	7.491	1.454	5.152	0.000
session1:consistencyPartial	1.896	0.656	2.890	0.004
session4:consistencyPartial	3.141	0.929	3.383	0.001

Two-Alternative Forced Choice (2AFC) Test

afc$exception = as.factor(afc$exception)

Trained Nouns

Figure 3

All Conditions

afcOLD <- droplevels(subset(afc, oldnew == "old"))

# Aggregate to obtain means per participant/day etc.
aggregated.afcOLD = aggregate(correct ~ participant + consistency_fig + typefreq + age, afcOLD, FUN=mean)

# Re-order levels of the semantic consistency variable (the defauly is alphabetical)
aggregated.afcOLD$consistency_fig = factor(aggregated.afcOLD$consistency_fig,levels(aggregated.afcOLD$consistency_fig)[c(1,3,2)])
levels(aggregated.afcOLD$consistency_fig)

## [1] "Fully Consistent"     "Partially Consistent" "Inconsistent"

# Re-order levels of the age variable (the defauly is alphabetical)
aggregated.afcOLD$age = factor(aggregated.afcOLD$age, levels(aggregated.afcOLD$age)[c(2,1)])
levels(aggregated.afcOLD$age)

## [1] "Child" "Adult"

# Now plot separate panels for each age group, split by semantic consistency, then by type frequency (recall that participants completed the 2afc task only in Session 4, so there is no "session" variable)
p = ggplot(aggregated.afcOLD, aes(x = consistency_fig, y = correct, fill = typefreq))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white"), name = "Type Frequency", labels = c("High", "Low"))
p = p + geom_hline(aes(yintercept = 0.50),linetype = "dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(~ age)
p = p + theme_bw()
p = p + mytheme
p = p + theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
print(p)

ggsave("Figure3_top_panel.jpg", width = 8, height = 5, dpi = 120)

Partially Consistent Condition Only

afcOLDP <- droplevels(subset(afc, oldnew == "old" & consistency == "Partial"))

# Aggregate to obtain means per participant/day etc.
aggregated.afcOLDP = aggregate(correct ~ participant + exception + typefreq + age, afcOLDP, FUN=mean)

# Re-order levels of the age variable (the defauly is alphabetical)
aggregated.afcOLDP$age = factor(aggregated.afcOLDP$age, levels(aggregated.afcOLDP$age)[c(2,1)])
levels(aggregated.afcOLDP$age)

## [1] "Child" "Adult"

p = ggplot(aggregated.afcOLDP, aes(x = typefreq, y = correct, fill = exception))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white"), name = "Noun Type", labels = c("Majority-Particle", "Exception"))
p = p + geom_hline(aes(yintercept = 0.50),linetype = "dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(~ age)
p = p + theme_bw()
p = p + mytheme
print(p)

ggsave("Figure3_bottom_panel.jpg", width = 8, height = 4, dpi = 120)

Statistical Analyses

Children

All Conditions

First run a model in which the fully consistent condition (“consistent”) is set as the baseline for contrasts (i.e., the model compares the fully consistent condition to both the partially consistent condition (“Consistent_VS_Partial”) and the inconsistent condition (“Consistent_VS_Inconsistent”))

afcOLDchild <- droplevels(subset(afc, age == "Child" & oldnew == "old"))
afcOLDchild <- lizCenter(afcOLDchild, list("typefreq"))
afcOLDchild <- lizContrasts(afcOLDchild, afcOLDchild$consistency, "Consistent")

afcOLDchild.lmer1 = glmer(correct ~
                         + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * typefreq.ct
                         + (1|participant),
                         data = afcOLDchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(afcOLDchild.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	1.103	0.132	8.326	0.000
Consistent_VS_Partial	-0.731	0.316	-2.310	0.021
Consistent_VS_Inconsistent	-0.771	0.316	-2.440	0.015
typefreq.ct	0.092	0.253	0.364	0.716
Consistent_VS_Partial:typefreq.ct	0.039	0.629	0.062	0.951
Consistent_VS_Inconsistent:typefreq.ct	-0.148	0.630	-0.234	0.815

Also compare the partially consistent and inconsistent conditions

afcOLDchild <- lizContrasts(afcOLDchild, afcOLDchild$consistency, "Inconsistent")

afcOLDchild.lmer2 = glmer(correct ~
                          + (Inconsistent_VS_Partial + Inconsistent_VS_Consistent) * typefreq.ct
                          + (1|participant),
                          data = afcOLDchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(afcOLDchild.lmer1, afcOLDchild.lmer2)

## Data: afcOLDchild
## Models:
## afcOLDchild.lmer1: correct ~ +(Consistent_VS_Partial + Consistent_VS_Inconsistent) * 
## afcOLDchild.lmer1:     typefreq.ct + (1 | participant)
## afcOLDchild.lmer2: correct ~ +(Inconsistent_VS_Partial + Inconsistent_VS_Consistent) * 
## afcOLDchild.lmer2:     typefreq.ct + (1 | participant)
##                   Df    AIC   BIC  logLik deviance Chisq Chi Df Pr(>Chisq)
## afcOLDchild.lmer1  7 835.45 867.5 -410.73   821.45                        
## afcOLDchild.lmer2  7 835.45 867.5 -410.73   821.45     0      0          1

get_coeffs(afcOLDchild.lmer2, c("Inconsistent_VS_Partial", "Inconsistent_VS_Partial:typefreq.ct"))

	Estimate	Std. Error	z value	Pr(>|z|)
Inconsistent_VS_Partial	0.041	0.301	0.135	0.893
Inconsistent_VS_Partial:typefreq.ct	0.186	0.602	0.310	0.757

Performance is significantly higher in the fully consistent condition compared to both the partially consistent and inconsistent conditions. There is no difference between the partially consistent and inconsistent conditions.

Partially Consistent Condition Only

Do children differ in their performance with majority-particle and exception items in the partially consistent condition? To look at this run a new model looking only at the partially consistent data, with exception (1 = exception noun; 0 = majority-particle noun) as a fixed effect, as well as interactions with the other factors.

afcOLDchildP <- droplevels(subset(afcOLDchild, consistency == "Partial"))
afcOLDchildP <- lizCenter(afcOLDchildP, list("typefreq", "exception"))

afcOLDchildP.lmer = glmer(correct ~ 
                          + typefreq.ct * exception.ct
                          + (1 + exception.ct|participant), 
                          data = afcOLDchildP, family = binomial, control = glmerControl(optimizer = "bobyqa"))

get_coeffs(afcOLDchildP.lmer, c("exception.ct", "typefreq.ct:exception.ct"))

	Estimate	Std. Error	z value	Pr(>|z|)
exception.ct	-0.425	0.324	-1.310	0.190
typefreq.ct:exception.ct	-0.426	0.637	-0.668	0.504

No evidence of a difference between majority-particle and exception nouns.

Adults

All Conditions

First run a model in which the fully consistent condition (“consistent”) is set as the baseline for contrasts (i.e., the model compares the fully consistent condition to both the partially consistent condition (“Consistent_VS_Partial”) and the inconsistent condition (“Consistent_VS_Inconsistent”)).

It was not possible to fit the full model (correlation of fixed factors = 1), therefore, we removed the interaction between semantic consistency and type frequency for both models.

afcOLDadult <- droplevels(subset(afc, age == "Adult" & oldnew == "old"))
afcOLDadult <- lizCenter(afcOLDadult, list("typefreq"))
afcOLDadult <- lizContrasts(afcOLDadult, afcOLDadult$consistency, "Consistent")

afcOLDadult.lmer1 = glmer(correct ~
                          + (Consistent_VS_Partial + Consistent_VS_Inconsistent) + typefreq.ct
                          + (1|participant),
                          data = afcOLDadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(afcOLDadult.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	6.087	1.533	3.970	0.000
Consistent_VS_Partial	-2.816	1.710	-1.647	0.100
Consistent_VS_Inconsistent	-1.707	1.692	-1.009	0.313
typefreq.ct	1.087	1.258	0.864	0.388

Also compare the partially consistent and inconsistent conditions

afcOLDadult <- lizContrasts(afcOLDadult, afcOLDadult$consistency, "Inconsistent")

afcOLDadult.lmer2 = glmer(correct ~
                          + (Inconsistent_VS_Partial + Inconsistent_VS_Consistent) + typefreq.ct
                          + (1|participant),
                          data = afcOLDadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))

anova(afcOLDadult.lmer1,afcOLDadult.lmer2)

## Data: afcOLDadult
## Models:
## afcOLDadult.lmer1: correct ~ +(Consistent_VS_Partial + Consistent_VS_Inconsistent) + 
## afcOLDadult.lmer1:     typefreq.ct + (1 | participant)
## afcOLDadult.lmer2: correct ~ +(Inconsistent_VS_Partial + Inconsistent_VS_Consistent) + 
## afcOLDadult.lmer2:     typefreq.ct + (1 | participant)
##                   Df    AIC    BIC  logLik deviance Chisq Chi Df
## afcOLDadult.lmer1  5 199.85 220.72 -94.924   189.85             
## afcOLDadult.lmer2  5 199.85 220.72 -94.924   189.85     0      0
##                   Pr(>Chisq)
## afcOLDadult.lmer1           
## afcOLDadult.lmer2          1

get_coeffs(afcOLDadult.lmer2, c("Inconsistent_VS_Partial"))

	Estimate	Std. Error	z value	Pr(>|z|)
Inconsistent_VS_Partial	-1.109	1.389	-0.798	0.425

Partially Consistent Condition Only

Do adults differ in their performance with majority-particle and exception items in the partially consistent condition? To look at this run a new model just looking at the partially consistent condition, with exception (1 = exception noun; 0 = majority-particle noun) as a fixed effect, as well as interactions with the other factors.

afcOLDadultP <- droplevels(subset(afcOLDadult, consistency == "Partial"))
afcOLDadultP <- lizCenter(afcOLDadultP, list("typefreq", "exception"))

afcOLDadultP.lmer = glmer(correct ~ 
                          + typefreq.ct * exception.ct
                          + (exception.ct|participant), 
                          data = afcOLDadultP, family = binomial, control = glmerControl(optimizer = "bobyqa"))

get_coeffs(afcOLDadultP.lmer, c("exception.ct", "typefreq.ct:exception.ct"))

	Estimate	Std. Error	z value	Pr(>|z|)
exception.ct	-0.703	1.617	-0.435	0.664
typefreq.ct:exception.ct	1.104	1.295	0.852	0.394

Novel Nouns

As in the production task, data from the inconsistent condition were not included in the novel nouns analyses

Figure 4

afcNEW <- droplevels(subset(afc, oldnew == "new" & consistency != "Inconsistent"))

# Aggregate to obtain means per participant/day etc.
aggregated.afcNEW = aggregate(correct ~ participant + consistency_fig + typefreq + age, afcNEW, FUN = mean)

# Re-order levels of the age variable (the default is alphabetical).
aggregated.afcNEW$age = factor(aggregated.afcNEW$age, levels(aggregated.afcNEW$age)[c(2,1)])
levels(aggregated.afcNEW$age)

## [1] "Child" "Adult"

# Now plot separate panels for each age group, split by consistency, then by type frequency
p = ggplot(aggregated.afcNEW, aes(x = consistency_fig, y = correct, fill = typefreq))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white"), name = "Type Frequency", labels = c("High", "Low"))
p = p + geom_hline(aes(yintercept = 0.50),linetype = "dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(~ age)
p = p + theme_bw()
p = p + mytheme
print(p)

ggsave("Figure4.jpg", width = 8, height = 6, dpi = 120)

Statistical Analyses

Children

Fully versus Partially Consistent Conditions

afcNEWchild <- droplevels(subset(afc, age == "Child" & oldnew == "new" & consistency != "Inconsistent"))
afcNEWchild <- lizCenter(afcNEWchild, list("consistency", "typefreq"))

afcNEWchild.lmer1 = glmer(correct ~
                          + consistency.ct * typefreq.ct
                          + (1|participant),
                          data = afcNEWchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(afcNEWchild.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.676	0.194	3.484	0.000
consistency.ct	-1.252	0.386	-3.241	0.001
typefreq.ct	-0.381	0.380	-1.004	0.315
consistency.ct:typefreq.ct	0.366	0.758	0.482	0.630

Is Performance Above Chance?

afcNEWchild.lmer2 = glmer(correct ~
                          + consistency
                          + consistency.ct * typefreq.ct
                          - 1
                          - consistency.ct
                          + (1|participant),
                          data = afcNEWchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(afcNEWchild.lmer1, afcNEWchild.lmer2)

## Data: afcNEWchild
## Models:
## afcNEWchild.lmer1: correct ~ +consistency.ct * typefreq.ct + (1 | participant)
## afcNEWchild.lmer2: correct ~ +consistency + consistency.ct * typefreq.ct - 1 - consistency.ct + 
## afcNEWchild.lmer2:     (1 | participant)
##                   Df    AIC    BIC  logLik deviance Chisq Chi Df
## afcNEWchild.lmer1  5 585.05 605.92 -287.52   575.05             
## afcNEWchild.lmer2  5 585.05 605.92 -287.52   575.05     0      0
##                   Pr(>Chisq)    
## afcNEWchild.lmer1               
## afcNEWchild.lmer2  < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

get_coeffs(afcNEWchild.lmer2, c("consistencyConsistent", "consistencyPartial"))

	Estimate	Std. Error	z value	Pr(>|z|)
consistencyConsistent	1.302	0.288	4.526	0.000
consistencyPartial	0.050	0.259	0.195	0.846

Adults

Fully versus Partially Consistent Condition

It was not possible to fit the full model (correlation of fixed factors = 1), therefore, as with trained nouns, we removed the interaction between semantic consistency and type frequency.

afcNEWadult <- droplevels(subset(afc, age == "Adult" & oldnew == "new" & consistency != "Inconsistent"))
afcNEWadult <- lizCenter(afcNEWadult, list("consistency", "typefreq"))

afcNEWadult.lmer1 = glmer(correct ~
                          + consistency.ct + typefreq.ct
                          + (1|participant),
                          data = afcNEWadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(afcNEWadult.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	4.566	0.979	4.664	0.000
consistency.ct	-3.918	1.394	-2.810	0.005
typefreq.ct	0.819	1.198	0.684	0.494

Is Performance Above Chance?

afcNEWadult.lmer2 = glmer(correct ~
                          + consistency
                          + consistency.ct + typefreq.ct
                          - 1
                          - consistency.ct
                          + (1|participant),
                          data = afcNEWadult, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(afcNEWadult.lmer1, afcNEWadult.lmer2)

## Data: afcNEWadult
## Models:
## afcNEWadult.lmer1: correct ~ +consistency.ct + typefreq.ct + (1 | participant)
## afcNEWadult.lmer2: correct ~ +consistency + consistency.ct + typefreq.ct - 1 - consistency.ct + 
## afcNEWadult.lmer2:     (1 | participant)
##                   Df    AIC    BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## afcNEWadult.lmer1  4 174.12 189.19 -83.06   166.12                        
## afcNEWadult.lmer2  4 174.12 189.19 -83.06   166.12     0      0  < 2.2e-16
##                      
## afcNEWadult.lmer1    
## afcNEWadult.lmer2 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

get_coeffs(afcNEWadult.lmer2, c("consistencyConsistent", "consistencyPartial"))

	Estimate	Std. Error	z value	Pr(>|z|)
consistencyConsistent	6.525	1.469	4.441	0.000
consistencyPartial	2.607	0.855	3.051	0.002

Awareness

Data Preparation

We need a column coding whether the participant is aware/unaware of the semantic conditioning. This is coded in the column “awareness”. However the coding for inconsistent is inappropriate as there are no semantic patterns in the data. We therefore re-code this as “na”.

prod1$awareness2 = as.character(prod1$awareness)
prod1$awareness2[prod1$consistency == "Inconsistent"] = "na"
prod1$awareness2 = factor(prod1$awareness2)

afc$awareness2 = as.character(afc$awareness)
afc$awareness2[afc$consistency == "Inconsistent"] = "na"
afc$awareness2 = factor(afc$awareness2)

Table 4 Data

Means: Production Test

aggregated.prod = aggregate(det_correct ~ participant + consistency + age + oldnew + awareness2 + typefreq + session + exception, data = prod1, FUN = mean)

aggregated.prod.M <- summarySEwithin(aggregated.prod, measurevar = "det_correct", betweenvars = c("oldnew", "age", "awareness2", "consistency", "typefreq"), withinvars = c("session", "exception"),  na.rm = FALSE, conf.interval = .95)

## Warning in qt(conf.interval/2 + 0.5, datac$N - 1): NaNs produced

## Warning in qt(conf.interval/2 + 0.5, datac$N - 1): NaNs produced

kable(aggregated.prod.M, digits = 3)

oldnew	age	awareness2	consistency	typefreq	session	exception	N	det_correct	det_correct_norm	sd	se	ci
new	Adult	0	Consistent	High	1	na	1	0.483	0.686	NA	NA	NA
new	Adult	0	Consistent	High	4	na	1	0.469	0.672	NA	NA	NaN
new	Adult	0	Partial	High	1	0	5	0.549	0.688	0.376	0.168	0.467
new	Adult	0	Partial	High	4	0	5	0.531	0.670	0.189	0.085	0.235
new	Adult	0	Partial	Low	1	0	5	0.547	0.638	0.227	0.102	0.282
new	Adult	0	Partial	Low	4	0	5	0.629	0.720	0.104	0.047	0.129
new	Adult	1	Consistent	High	1	na	9	0.760	0.559	0.418	0.139	0.321
new	Adult	1	Consistent	High	4	na	9	1.000	0.799	0.000	0.000	0.000
new	Adult	1	Consistent	Low	1	na	10	0.981	0.674	0.043	0.014	0.031
new	Adult	1	Consistent	Low	4	na	10	0.991	0.684	0.023	0.007	0.017
new	Adult	1	Partial	High	1	0	5	0.908	0.633	0.160	0.071	0.198
new	Adult	1	Partial	High	4	0	5	1.000	0.725	0.000	0.000	0.000
new	Adult	1	Partial	Low	1	0	5	0.866	0.621	0.106	0.047	0.132
new	Adult	1	Partial	Low	4	0	5	0.981	0.736	0.046	0.021	0.057
new	Adult	na	Inconsistent	High	1	na	10	0.304	0.626	0.301	0.095	0.215
new	Adult	na	Inconsistent	High	4	na	10	0.410	0.732	0.196	0.062	0.140
new	Adult	na	Inconsistent	Low	1	na	10	0.433	0.629	0.254	0.080	0.181
new	Adult	na	Inconsistent	Low	4	na	10	0.534	0.729	0.192	0.061	0.137
new	Child	0	Consistent	High	1	na	7	0.477	0.660	0.057	0.022	0.053
new	Child	0	Consistent	High	4	na	7	0.514	0.698	0.042	0.016	0.039
new	Child	0	Consistent	Low	1	na	10	0.497	0.659	0.168	0.053	0.120
new	Child	0	Consistent	Low	4	na	9	0.539	0.701	0.191	0.064	0.147
new	Child	0	Partial	High	1	0	11	0.476	0.662	0.113	0.034	0.076
new	Child	0	Partial	High	4	0	13	0.508	0.694	0.087	0.024	0.053
new	Child	0	Partial	Low	1	0	15	0.445	0.658	0.104	0.027	0.057
new	Child	0	Partial	Low	4	0	15	0.486	0.700	0.123	0.032	0.068
new	Child	1	Consistent	High	1	na	8	0.635	0.570	0.200	0.071	0.167
new	Child	1	Consistent	High	4	na	8	0.854	0.788	0.128	0.045	0.107
new	Child	1	Consistent	Low	1	na	4	0.766	0.594	0.209	0.105	0.333
new	Child	1	Consistent	Low	4	na	5	0.920	0.747	0.127	0.057	0.157
new	Child	1	Partial	High	1	0	2	0.716	0.630	0.334	0.236	3.000
new	Child	1	Partial	High	4	0	2	0.814	0.728	0.192	0.136	1.725
new	Child	na	Inconsistent	High	1	na	13	0.521	0.670	0.110	0.030	0.066
new	Child	na	Inconsistent	High	4	na	15	0.537	0.686	0.127	0.033	0.070
new	Child	na	Inconsistent	Low	1	na	12	0.570	0.730	0.171	0.049	0.108
new	Child	na	Inconsistent	Low	4	na	15	0.477	0.638	0.062	0.016	0.034
old	Adult	0	Consistent	High	1	na	1	0.500	0.585	NA	NA	NaN
old	Adult	0	Consistent	High	4	na	1	0.688	0.773	NA	NA	NaN
old	Adult	0	Partial	High	1	0	5	0.856	0.821	0.164	0.073	0.203
old	Adult	0	Partial	High	1	1	5	0.450	0.415	0.439	0.197	0.546
old	Adult	0	Partial	High	4	0	5	0.850	0.815	0.153	0.069	0.191
old	Adult	0	Partial	High	4	1	5	0.700	0.665	0.382	0.171	0.475
old	Adult	0	Partial	Low	1	0	5	0.799	0.677	0.180	0.081	0.224
old	Adult	0	Partial	Low	1	1	5	0.739	0.618	0.265	0.119	0.329
old	Adult	0	Partial	Low	4	0	5	0.789	0.668	0.136	0.061	0.168
old	Adult	0	Partial	Low	4	1	5	0.875	0.753	0.097	0.043	0.120
old	Adult	1	Consistent	High	1	na	9	0.866	0.614	0.227	0.076	0.175
old	Adult	1	Consistent	High	4	na	9	0.997	0.744	0.011	0.004	0.009
old	Adult	1	Consistent	Low	1	na	10	0.985	0.675	0.027	0.009	0.019
old	Adult	1	Consistent	Low	4	na	10	0.994	0.683	0.014	0.005	0.010
old	Adult	1	Partial	High	1	0	5	0.882	0.761	0.201	0.090	0.249
old	Adult	1	Partial	High	1	1	5	0.325	0.204	0.229	0.102	0.285
old	Adult	1	Partial	High	4	0	5	0.992	0.871	0.020	0.009	0.025
old	Adult	1	Partial	High	4	1	5	1.000	0.879	0.000	0.000	0.000
old	Adult	1	Partial	Low	1	0	5	0.717	0.488	0.481	0.215	0.598
old	Adult	1	Partial	Low	1	1	4	0.917	0.688	0.183	0.091	0.291
old	Adult	1	Partial	Low	4	0	5	1.000	0.771	0.000	0.000	0.000
old	Adult	1	Partial	Low	4	1	5	1.000	0.771	0.000	0.000	0.000
old	Adult	na	Inconsistent	High	1	na	10	0.619	0.531	0.151	0.048	0.108
old	Adult	na	Inconsistent	High	4	na	10	0.916	0.827	0.169	0.053	0.121
old	Adult	na	Inconsistent	Low	1	na	10	0.840	0.617	0.161	0.051	0.115
old	Adult	na	Inconsistent	Low	4	na	10	0.962	0.740	0.118	0.037	0.085
old	Child	0	Consistent	High	1	na	7	0.514	0.574	0.064	0.024	0.059
old	Child	0	Consistent	High	4	na	7	0.725	0.784	0.153	0.058	0.142
old	Child	0	Consistent	Low	1	na	10	0.536	0.544	0.203	0.064	0.145
old	Child	0	Consistent	Low	4	na	10	0.806	0.814	0.139	0.044	0.100
old	Child	0	Partial	High	1	0	13	0.519	0.598	0.124	0.034	0.075
old	Child	0	Partial	High	1	1	13	0.504	0.583	0.265	0.073	0.160
old	Child	0	Partial	High	4	0	13	0.688	0.767	0.178	0.049	0.108
old	Child	0	Partial	High	4	1	13	0.690	0.768	0.245	0.068	0.148
old	Child	0	Partial	Low	1	0	15	0.574	0.600	0.174	0.045	0.096
old	Child	0	Partial	Low	1	1	15	0.609	0.635	0.249	0.064	0.138
old	Child	0	Partial	Low	4	0	15	0.702	0.729	0.194	0.050	0.107
old	Child	0	Partial	Low	4	1	15	0.726	0.752	0.268	0.069	0.148
old	Child	1	Consistent	High	1	na	8	0.665	0.559	0.165	0.058	0.138
old	Child	1	Consistent	High	4	na	8	0.905	0.799	0.099	0.035	0.082
old	Child	1	Consistent	Low	1	na	5	0.792	0.584	0.272	0.122	0.338
old	Child	1	Consistent	Low	4	na	5	0.981	0.774	0.046	0.021	0.057
old	Child	1	Partial	High	1	0	2	0.739	0.788	0.337	0.238	3.026
old	Child	1	Partial	High	1	1	2	0.250	0.299	0.387	0.274	3.480
old	Child	1	Partial	High	4	0	2	0.958	1.008	0.065	0.046	0.580
old	Child	1	Partial	High	4	1	2	0.571	0.621	0.664	0.469	5.965
old	Child	na	Inconsistent	High	1	na	15	0.519	0.625	0.114	0.029	0.063
old	Child	na	Inconsistent	High	4	na	15	0.628	0.733	0.177	0.046	0.098
old	Child	na	Inconsistent	Low	1	na	14	0.583	0.585	0.151	0.040	0.087
old	Child	na	Inconsistent	Low	4	na	15	0.765	0.767	0.176	0.046	0.098

Means: 2AFC Test

aggregated.afc = aggregate(correct ~ participant + consistency + age + oldnew + awareness2 + typefreq + exception, data = afc, FUN = mean)

aggregated.afc.M <- summarySEwithin(aggregated.afc, measurevar = "correct", betweenvars = c("oldnew", "age", "awareness2", "consistency", "typefreq"), withinvars = "exception",  na.rm = FALSE, conf.interval = .95)

## Warning in qt(conf.interval/2 + 0.5, datac$N - 1): NaNs produced

## Warning in qt(conf.interval/2 + 0.5, datac$N - 1): NaNs produced

kable(aggregated.afc.M, digits = 3)

oldnew	age	awareness2	consistency	typefreq	exception	N	correct	correct_norm	sd	se	ci
new	Adult	0	Consistent	High	0	1	0.500	0.722	NA	NA	NaN
new	Adult	0	Partial	High	0	5	0.550	0.722	0.202	0.090	0.250
new	Adult	0	Partial	Low	0	5	0.600	0.722	0.316	0.141	0.393
new	Adult	1	Consistent	High	0	9	1.000	0.722	0.000	0.000	0.000
new	Adult	1	Consistent	Low	0	10	1.000	0.722	0.000	0.000	0.000
new	Adult	1	Partial	High	0	5	1.000	0.722	0.000	0.000	0.000
new	Adult	1	Partial	Low	0	5	1.000	0.722	0.000	0.000	0.000
new	Adult	na	Inconsistent	High	0	10	0.450	0.722	0.325	0.103	0.232
new	Adult	na	Inconsistent	Low	0	10	0.462	0.722	0.312	0.099	0.223
new	Child	0	Consistent	High	0	7	0.536	0.722	0.318	0.120	0.294
new	Child	0	Consistent	Low	0	10	0.512	0.722	0.328	0.104	0.234
new	Child	0	Partial	High	0	13	0.452	0.722	0.266	0.074	0.160
new	Child	0	Partial	Low	0	15	0.492	0.722	0.324	0.084	0.179
new	Child	1	Consistent	High	0	8	0.969	0.722	0.082	0.029	0.068
new	Child	1	Consistent	Low	0	5	1.000	0.722	0.000	0.000	0.000
new	Child	1	Partial	High	0	2	1.000	0.722	0.000	0.000	0.000
new	Child	na	Inconsistent	High	0	15	0.450	0.722	0.274	0.071	0.152
new	Child	na	Inconsistent	Low	0	15	0.567	0.722	0.175	0.045	0.097
old	Adult	0	Consistent	High	0	1	0.250	0.722	NA	NA	NaN
old	Adult	0	Partial	High	0	5	0.800	0.822	0.387	0.173	0.481
old	Adult	0	Partial	High	1	5	0.600	0.622	0.592	0.265	0.735
old	Adult	0	Partial	Low	0	5	0.733	0.738	0.316	0.141	0.393
old	Adult	0	Partial	Low	1	5	0.700	0.705	0.387	0.173	0.481
old	Adult	1	Consistent	High	0	9	1.000	0.722	0.000	0.000	0.000
old	Adult	1	Consistent	Low	0	10	1.000	0.722	0.000	0.000	0.000
old	Adult	1	Partial	High	0	5	1.000	0.722	0.000	0.000	0.000
old	Adult	1	Partial	High	1	5	1.000	0.722	0.000	0.000	0.000
old	Adult	1	Partial	Low	0	5	1.000	0.722	0.000	0.000	0.000
old	Adult	1	Partial	Low	1	5	1.000	0.722	0.000	0.000	0.000
old	Adult	na	Inconsistent	High	0	10	0.912	0.722	0.168	0.053	0.120
old	Adult	na	Inconsistent	Low	0	10	0.975	0.722	0.112	0.035	0.080
old	Child	0	Consistent	High	0	7	0.643	0.722	0.238	0.090	0.220
old	Child	0	Consistent	Low	0	10	0.725	0.722	0.310	0.098	0.221
old	Child	0	Partial	High	0	13	0.628	0.709	0.239	0.066	0.144
old	Child	0	Partial	High	1	13	0.654	0.734	0.531	0.147	0.321
old	Child	0	Partial	Low	0	15	0.733	0.788	0.293	0.076	0.162
old	Child	0	Partial	Low	1	15	0.600	0.655	0.293	0.076	0.162
old	Child	1	Consistent	High	0	8	0.922	0.722	0.188	0.066	0.157
old	Child	1	Consistent	Low	0	5	0.975	0.722	0.079	0.035	0.098
old	Child	1	Partial	High	0	2	1.000	0.972	0.000	0.000	0.000
old	Child	1	Partial	High	1	2	0.500	0.472	1.000	0.707	8.985
old	Child	na	Inconsistent	High	0	15	0.675	0.722	0.312	0.081	0.173
old	Child	na	Inconsistent	Low	0	15	0.667	0.722	0.388	0.100	0.215

Compare Number of Aware/Unaware Participants (Chi Square)

adults.aware = with(droplevels(subset(aggregated.afc, age == "Adult" & consistency != "Inconsistent" & oldnew == "new")), table(consistency, awareness2))
adults.aware

##             awareness2
## consistency   0  1
##   Consistent  1 19
##   Partial    10 10

chisq.test(adults.aware)

## 
##  Pearson's Chi-squared test with Yates' continuity correction
## 
## data:  adults.aware
## X-squared = 8.0251, df = 1, p-value = 0.004613

child.aware = with(droplevels(subset(aggregated.afc, age == "Child" & consistency != "Inconsistent" & oldnew == "new")), table(consistency, awareness2))
child.aware

##             awareness2
## consistency   0  1
##   Consistent 17 13
##   Partial    28  2

chisq.test(child.aware)

## 
##  Pearson's Chi-squared test with Yates' continuity correction
## 
## data:  child.aware
## X-squared = 8.8889, df = 1, p-value = 0.002869

For Table 4 we also need to know the number of aware/unaware participant split by type frequency

adults.awareT4 = with(droplevels(subset(aggregated.afc, age == "Adult" & consistency != "Inconsistent" & oldnew == "new")), table(consistency, typefreq, awareness2))
adults.awareT4

## , , awareness2 = 0
## 
##             typefreq
## consistency  High Low
##   Consistent    1   0
##   Partial       5   5
## 
## , , awareness2 = 1
## 
##             typefreq
## consistency  High Low
##   Consistent    9  10
##   Partial       5   5

child.awareT4 = with(droplevels(subset(aggregated.afc, age == "Child" & consistency != "Inconsistent" & oldnew == "new")), table(consistency, typefreq, awareness2))
child.awareT4

## , , awareness2 = 0
## 
##             typefreq
## consistency  High Low
##   Consistent    7  10
##   Partial      13  15
## 
## , , awareness2 = 1
## 
##             typefreq
## consistency  High Low
##   Consistent    8   5
##   Partial       2   0

Trained Nouns: Production Test

Children, Fully Consistent Condition

Figure 5

PRODOLD.unaware <- droplevels(subset(prod1, oldnew == "old" & age == "Child"))

# Aggregate to obtain means per participant/day etc.
aggregated.PRODOLD.unaware = aggregate(det_correct ~ participant + session_fig + consistency_fig + awareness2, PRODOLD.unaware, FUN = mean)

# Re-order levels of the semantic consistency variable (the default is alphabetical)
aggregated.PRODOLD.unaware$consistency_fig = factor(aggregated.PRODOLD.unaware$consistency_fig, levels(aggregated.PRODOLD.unaware$consistency_fig)[c(1,3,2)])
levels(aggregated.PRODOLD.unaware$consistency_fig)

## [1] "Fully Consistent"     "Partially Consistent" "Inconsistent"

# This plot creates separate panels for Session 1 and Session 4, split by semantic consistency, then by awareness
p = ggplot(aggregated.PRODOLD.unaware, aes(x = consistency_fig, y = det_correct, fill = awareness2))
p = p + stat_summary(fun.y = mean, geom = "bar", position = "dodge", colour = "black", size = .3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width = .4, position = position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white", "black"), name = "Awareness", labels = c("Unaware", "Aware", "NA"))
p = p + geom_hline(aes(yintercept = 0.50), linetype = "dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(~ session_fig)
p = p + theme_bw()
p = p + mytheme
p = p + theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
print(p)

ggsave("Figure5.jpg", width = 6, height = 6, dpi = 120)

Statistical Analyses: Unaware Participants

Comparisons across noun types, children, production

prodOLDchild.unaware = droplevels(subset(prod1, age == "Child" & oldnew == "old" & awareness2 != 1))
prodOLDchild.unaware <- lizCenter(prodOLDchild.unaware, list("typefreq", "session"))
prodOLDchild.unaware <- lizContrasts(prodOLDchild.unaware, prodOLDchild.unaware$consistency, "Consistent")

prodOLDchild.unaware.lmer2 = glmer(det_correct ~
                                  + (Consistent_VS_Inconsistent + Consistent_VS_Partial) * typefreq.ct * session.ct
                                  + (1 + session.ct|participant),
                                  data = prodOLDchild.unaware, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodOLDchild.unaware.lmer2)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.693	0.075	9.174	0.000
Consistent_VS_Inconsistent	-0.186	0.191	-0.971	0.331
Consistent_VS_Partial	-0.159	0.193	-0.823	0.411
typefreq.ct	0.312	0.150	2.076	0.038
session.ct	0.942	0.106	8.917	0.000
Consistent_VS_Inconsistent:typefreq.ct	0.045	0.384	0.118	0.906
Consistent_VS_Partial:typefreq.ct	-0.201	0.387	-0.519	0.604
Consistent_VS_Inconsistent:session.ct	-0.406	0.267	-1.522	0.128
Consistent_VS_Partial:session.ct	-0.542	0.267	-2.026	0.043
typefreq.ct:session.ct	0.389	0.209	1.860	0.063
Consistent_VS_Inconsistent:typefreq.ct:session.ct	0.359	0.533	0.673	0.501
Consistent_VS_Partial:typefreq.ct:session.ct	-0.474	0.536	-0.884	0.376

There is an interaction between Consistent_VS_Partial and session.

prodOLDchild.unaware.lmer2b = glmer(det_correct ~
                                  + Consistent_VS_Partial:session
                                  + Consistent_VS_Inconsistent:session
                                  + (Consistent_VS_Partial + Consistent_VS_Inconsistent):typefreq.ct
                                  + (Consistent_VS_Partial + Consistent_VS_Inconsistent):typefreq.ct:session.ct
                                  + typefreq.ct * session.ct
                                  + (1 + session.ct|participant),
                                  data = prodOLDchild.unaware, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodOLDchild.unaware.lmer2, prodOLDchild.unaware.lmer2b)

## Data: prodOLDchild.unaware
## Models:
## prodOLDchild.unaware.lmer2: det_correct ~ +(Consistent_VS_Inconsistent + Consistent_VS_Partial) * 
## prodOLDchild.unaware.lmer2:     typefreq.ct * session.ct + (1 + session.ct | participant)
## prodOLDchild.unaware.lmer2b: det_correct ~ +Consistent_VS_Partial:session + Consistent_VS_Inconsistent:session + 
## prodOLDchild.unaware.lmer2b:     (Consistent_VS_Partial + Consistent_VS_Inconsistent):typefreq.ct + 
## prodOLDchild.unaware.lmer2b:     (Consistent_VS_Partial + Consistent_VS_Inconsistent):typefreq.ct:session.ct + 
## prodOLDchild.unaware.lmer2b:     typefreq.ct * session.ct + (1 + session.ct | participant)
##                             Df    AIC    BIC  logLik deviance Chisq Chi Df
## prodOLDchild.unaware.lmer2  15 5096.4 5191.1 -2533.2   5066.4             
## prodOLDchild.unaware.lmer2b 15 5096.4 5191.1 -2533.2   5066.4     0      0
##                             Pr(>Chisq)
## prodOLDchild.unaware.lmer2            
## prodOLDchild.unaware.lmer2b          1

get_coeffs(prodOLDchild.unaware.lmer2b, c("Consistent_VS_Partial:session1", "Consistent_VS_Partial:session4"))

	Estimate	Std. Error	z value	Pr(>|z|)
Consistent_VS_Partial:session1	0.143	0.151	0.948	0.343
Consistent_VS_Partial:session4	-0.399	0.284	-1.406	0.160

Adults, Partially Consistent Condition

Figure 6

prodOLDP <- droplevels(subset(prod1, oldnew == "old" & age == "Adult" & consistency == "Partial"))

# Aggregate across participants etc.
aggregated.prodOLDP = aggregate(det_correct ~ participant + session_fig + exception + typefreq + age + awareness2, prodOLDP, FUN = mean)

# Now plot
p = ggplot(aggregated.prodOLDP, aes(x = typefreq, y = det_correct, fill = exception))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white"), name = "Noun-Type", labels = c("Majority-Particle", "Exception"))
p = p + geom_hline(aes(yintercept = 0.50), linetype = "dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(awareness2 ~ session_fig)
p = p + theme_bw()
p = p + mytheme
print(p)

ggsave("Figure6.jpg", width = 8, height = 6, dpi = 120)

Statistical Analyses: Aware versus Unaware Participants

The full model would not converge. Correlations between slopes were removed using “||” syntax

prodOLDadultP.unaware <- droplevels(subset(prod1, oldnew == "old" & age == "Adult" & consistency == "Partial" & awareness2 == "0"))
prodOLDadultP.unaware <- lizCenter(prodOLDadultP.unaware, list("typefreq", "session", "exception"))

prodOLDadultP.unaware.lmer = glmer(det_correct ~ 
                                + typefreq.ct * session.ct * exception.ct
                                + (1 + session.ct * exception.ct||participant), 
                                data = prodOLDadultP.unaware, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodOLDadultP.unaware.lmer)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	1.755	0.389	4.515	0.000
typefreq.ct	-0.102	0.768	-0.132	0.895
session.ct	0.265	0.222	1.191	0.234
exception.ct	-0.717	0.257	-2.791	0.005
typefreq.ct:session.ct	-0.115	0.444	-0.260	0.795
typefreq.ct:exception.ct	1.978	0.516	3.833	0.000
session.ct:exception.ct	1.208	0.948	1.274	0.203
typefreq.ct:session.ct:exception.ct	-0.471	1.897	-0.248	0.804

Break down the interaction between type frequency and noun-type (“exception”)

prodOLDadultP.unaware.lmer.b = glmer(det_correct ~ 
                                + exception.ct :typefreq
                                + typefreq.ct * session.ct 
                                + session.ct : exception.ct
                                + typefreq.ct : session.ct : exception.ct
                                + (1 + session.ct * exception.ct||participant), 
                                data = prodOLDadultP.unaware, family = binomial, control = glmerControl(optimizer = "bobyqa"))
anova(prodOLDadultP.unaware.lmer, prodOLDadultP.unaware.lmer.b)

## Data: prodOLDadultP.unaware
## Models:
## prodOLDadultP.unaware.lmer: det_correct ~ +typefreq.ct * session.ct * exception.ct + (1 + 
## prodOLDadultP.unaware.lmer:     session.ct * exception.ct || participant)
## prodOLDadultP.unaware.lmer.b: det_correct ~ +exception.ct:typefreq + typefreq.ct * session.ct + 
## prodOLDadultP.unaware.lmer.b:     session.ct:exception.ct + typefreq.ct:session.ct:exception.ct + 
## prodOLDadultP.unaware.lmer.b:     (1 + session.ct * exception.ct || participant)
##                              Df    AIC    BIC  logLik deviance Chisq
## prodOLDadultP.unaware.lmer   12 584.21 637.62 -280.11   560.21      
## prodOLDadultP.unaware.lmer.b 12 584.21 637.62 -280.11   560.21     0
##                              Chi Df Pr(>Chisq)
## prodOLDadultP.unaware.lmer                    
## prodOLDadultP.unaware.lmer.b      0          1

get_coeffs(prodOLDadultP.unaware.lmer.b, c("exception.ct:typefreqHigh","exception.ct:typefreqLow"))

	Estimate	Std. Error	z value	Pr(>|z|)
exception.ct:typefreqHigh	-1.708	0.364	-4.697	0.000
exception.ct:typefreqLow	0.270	0.365	0.741	0.459

Novel Nouns: Production Test

Figure 7

prodNEW <- droplevels(subset(prod1,  oldnew == "new" & consistency != "Inconsistent"))

# Aggregate across participants, session etc.
aggregated.prodNEW = aggregate(det_correct ~ participant + session + consistency_fig + awareness2 + age, prodNEW, FUN = mean)

# Reorder levels of the factor age (default is alphabetical)
aggregated.prodNEW$age = factor(aggregated.prodNEW$age, levels(aggregated.prodNEW$age)[c(2,1)])
levels(aggregated.prodNEW$age)

## [1] "Child" "Adult"

# Now plot, with separate panels for each age, split by semantic consistency and awareness
p = ggplot(aggregated.prodNEW, aes(x = consistency_fig, y = det_correct, fill = awareness2))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white"), name = "Awareness", labels = c("Unaware", "Aware"))
p = p + geom_hline(aes(yintercept = 0.50), linetype = "dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(awareness2 ~ age + session)
p = p + theme_bw()
p = p + mytheme
p = p + theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
print(p)

## Warning: Removed 2 rows containing missing values (geom_errorbar).

ggsave("Figure7.jpg", width = 8, height = 6, dpi = 120)

## Warning: Removed 2 rows containing missing values (geom_errorbar).

Statitsical Analyses: Unaware Participants

Is performance in unaware participants above chance?

Children, Fully Consistent Condition

prodNEWchild.con.unaware = droplevels(subset(prod1, age == "Child" & oldnew == "new" & consistency == "Consistent" & awareness2 == "0"))
prodNEWchild.con.unaware = lizCenter(prodNEWchild.con.unaware, list("typefreq", "session"))

prodNEWchild.con.unaware.lmer = glmer(det_correct ~
                              + typefreq.ct * session.ct
                              + (session.ct|participant),
                              data = prodNEWchild.con.unaware, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodNEWchild.con.unaware.lmer)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.062	0.125	0.501	0.616
typefreq.ct	0.139	0.251	0.553	0.580
session.ct	0.118	0.140	0.838	0.402
typefreq.ct:session.ct	-0.031	0.282	-0.110	0.912

Adults, Partially Consistent Condition

prodNEWadult.par.unaware <- droplevels(subset(prod1, age == "Adult" & oldnew == "new" & consistency == "Partial" & awareness2 == "0"))
prodNEWadult.par.unaware = lizCenter(prodNEWadult.par.unaware, list("typefreq", "session"))

prodNEWadult.par.unaware.lmer = glmer(det_correct ~
                              + typefreq.ct*session.ct
                              + (session.ct|participant),
                              data = prodNEWadult.par.unaware, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodNEWadult.par.unaware.lmer)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.346	0.199	1.741	0.082
typefreq.ct	0.124	0.396	0.313	0.754
session.ct	-0.021	0.523	-0.040	0.968
typefreq.ct:session.ct	0.566	1.045	0.542	0.588

Trained Nouns, 2AFC Test

Children, Fully Consistent Condition

Figure 8

afcOLD.child <- droplevels(subset(afc, oldnew == "old" & age == "Child"))

# Aggregate to obtain means per participant etc.
aggregated.afcOLD.child = aggregate(correct ~ participant + consistency_fig + awareness2, afcOLD.child, FUN = mean)

#re-order levels of consistency variable, the default is alphabetical.
aggregated.afcOLD.child$consistency_fig = factor(aggregated.afcOLD.child$consistency_fig, levels(aggregated.afcOLD.child$consistency_fig)[c(1,3,2)])
levels(aggregated.afcOLD.child$consistency_fig)

## [1] "Fully Consistent"     "Partially Consistent" "Inconsistent"

# Now plot, splitting the data by semantic consistency, then by awareness
p = ggplot(aggregated.afcOLD.child, aes(x = consistency_fig, y = correct, fill = awareness2))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white", "black"), name = "Awareness", labels = c("Unaware", "Aware", "NA"))
p = p + geom_hline(aes(yintercept = 0.50),linetype = "dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + theme_bw()
p = p + mytheme
print(p)

ggsave("Figure8.jpg", width = 6, height = 4, dpi = 120)

Statistical Analyses: Unaware Participants

afcOLDchild.unaware = droplevels(subset(afc, oldnew == "old" & age == "Child" & awareness2 != 1))
afcOLDchild.unaware <- lizContrasts(afcOLDchild.unaware, afcOLDchild.unaware$consistency, "Consistent")
afcOLDchild.unaware <- lizCenter(afcOLDchild.unaware, list("typefreq"))

afcOLDchild.unaware.lmer2 = glmer(correct ~
                                  + (Consistent_VS_Inconsistent + Consistent_VS_Partial) * typefreq.ct 
                                  + (1|participant),
                                  data = afcOLDchild.unaware, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(afcOLDchild.unaware.lmer2)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.810	0.126	6.437	0.000
Consistent_VS_Inconsistent	-0.075	0.314	-0.239	0.811
Consistent_VS_Partial	-0.093	0.317	-0.293	0.770
typefreq.ct	0.241	0.247	0.976	0.329
Consistent_VS_Inconsistent:typefreq.ct	-0.456	0.630	-0.724	0.469
Consistent_VS_Partial:typefreq.ct	-0.108	0.637	-0.170	0.865

Novel Nouns: 2AFC Test

Figure 9

afcNEW <- droplevels(subset(afc, oldnew == "new" & consistency != "Inconsistent"))

# Aggregate across participants etc.
aggregated.afcNEW = aggregate(correct ~ participant + consistency_fig + age + awareness2, afcNEW, FUN = mean)

# Reorder levels of the factor age (default is alphabetical)
aggregated.afcNEW$age = factor(aggregated.afcNEW$age, levels(aggregated.afcNEW$age)[c(2,1)])
levels(aggregated.afcNEW$age)

## [1] "Child" "Adult"

# This plot creates separate panels for S1 and S4, split by consistency, then by type-freq
p = ggplot(aggregated.afcNEW, aes(x = consistency_fig, y = correct, fill = awareness2))
p = p + stat_summary(fun.y = mean, geom = "bar", position="dodge", colour="black", size=.3) 
p = p + stat_summary(fun.data = mean_cl_boot, geom = "errorbar", width=.4, position=position_dodge(.9)) 
p = p + xlab("") + ylab("Proportion Correct")
p = p + scale_fill_manual(values = c("grey", "white"), name = "Awareness", labels = c("Unaware", "Aware"))
p = p + geom_hline(aes(yintercept = 0.50),linetype = "dashed")
p = p + coord_cartesian(ylim = c(0, 1))
p = p + facet_grid(awareness2 ~ age)
p = p + theme_bw()
p = p + mytheme
p = p + theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
print(p)

## Warning: Removed 1 rows containing missing values (geom_errorbar).

ggsave("Figure9.jpg", width = 6, height = 6, dpi = 120)

## Warning: Removed 1 rows containing missing values (geom_errorbar).

Statistical Analyses: Unaware Participants

Children, Fully Consistent Condition

afcNEWchild.con.unaware <- droplevels(subset(afc, age == "Child" & oldnew == "new" & consistency == "Consistent" & awareness2 == "0"))
afcNEWchild.con.unaware = lizCenter(afcNEWchild.con.unaware, list("typefreq"))

afcNEWchild.con.unaware.lmer = glmer(correct ~
                              + typefreq.ct
                              + (1|participant),
                              data = afcNEWchild.con.unaware, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(afcNEWchild.con.unaware.lmer)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.096	0.221	0.435	0.663
typefreq.ct	-0.098	0.449	-0.218	0.827

Adults, Partially Consistent Condition

afcNEWadult.par.unaware <- droplevels(subset(afc, age == "Adult" & oldnew == "new" & consistency == "Partial" & awareness2 == "0"))
afcNEWadult.par.unaware = lizCenter(afcNEWadult.par.unaware, list("typefreq"))

afcNEWadult.par.unaware.lmer = glmer(correct ~
                              + typefreq.ct
                              + (1|participant),
                              data = afcNEWadult.par.unaware, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(afcNEWadult.par.unaware.lmer)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.303	0.226	1.338	0.181
typefreq.ct	0.205	0.453	0.452	0.651

Supplementary Analyses

Memory Measures

Means (Table 1)

Children

aggregated.child.mem = aggregate(cbind(age_months, stm_ver_raw, stm_ver_stand, stm_nonver_raw, stm_nonver_stand, wm_ver_raw, wm_ver_stand, mem_composite_raw) ~ participant + consistency + typefreq, subset(prod1, age == "Child"), FUN = mean)

kable(summarySE(aggregated.child.mem, measurevar= "stm_ver_stand", groupvars = c( "consistency", "typefreq")), digits = 2)

consistency	typefreq	N	stm_ver_stand	sd	se	ci
Consistent	High	15	113.27	12.60	3.25	6.98
Consistent	Low	15	111.40	13.69	3.54	7.58
Inconsistent	High	15	101.73	13.12	3.39	7.27
Inconsistent	Low	15	105.40	16.11	4.16	8.92
Partial	High	15	102.13	17.39	4.49	9.63
Partial	Low	15	102.13	13.27	3.43	7.35

kable(summarySE(aggregated.child.mem, measurevar= "stm_nonver_stand", groupvars = c( "consistency", "typefreq")), digits = 2)

consistency	typefreq	N	stm_nonver_stand	sd	se	ci
Consistent	High	15	112.07	10.91	2.82	6.04
Consistent	Low	15	118.40	15.79	4.08	8.74
Inconsistent	High	15	106.93	16.45	4.25	9.11
Inconsistent	Low	15	109.27	13.10	3.38	7.26
Partial	High	15	112.60	18.53	4.78	10.26
Partial	Low	15	116.67	12.62	3.26	6.99

kable(summarySE(aggregated.child.mem, measurevar= "wm_ver_stand", groupvars = c( "consistency", "typefreq")), digits = 2)

consistency	typefreq	N	wm_ver_stand	sd	se	ci
Consistent	High	15	108.07	8.59	2.22	4.76
Consistent	Low	15	112.53	12.67	3.27	7.02
Inconsistent	High	15	108.13	16.66	4.30	9.23
Inconsistent	Low	15	110.40	20.03	5.17	11.09
Partial	High	15	106.27	20.21	5.22	11.19
Partial	Low	15	112.93	12.66	3.27	7.01

Adults

aggregated.adult.mem = aggregate(cbind(age_months, stm_ver_raw, stm_ver_stand, stm_nonver_raw, stm_nonver_stand, wm_ver_raw, wm_ver_stand, mem_composite_raw) ~ participant + consistency + typefreq, subset(prod1, age == "Adult"), FUN = mean)

kable(summarySE(aggregated.adult.mem, measurevar= "stm_ver_stand", groupvars = c( "consistency", "typefreq")), digits = 2)

consistency	typefreq	N	stm_ver_stand	sd	se	ci
Consistent	High	10	106.4	15.23	4.82	10.90
Consistent	Low	10	99.0	19.07	6.03	13.64
Inconsistent	High	10	100.7	16.68	5.27	11.93
Inconsistent	Low	10	101.4	16.52	5.22	11.82
Partial	High	10	105.0	19.77	6.25	14.14
Partial	Low	10	104.5	12.07	3.82	8.63

kable(summarySE(aggregated.adult.mem, measurevar= "stm_nonver_stand", groupvars = c( "consistency", "typefreq")), digits = 2)

consistency	typefreq	N	stm_nonver_stand	sd	se	ci
Consistent	High	10	93.8	18.30	5.79	13.09
Consistent	Low	10	99.7	13.06	4.13	9.35
Inconsistent	High	10	95.0	15.42	4.88	11.03
Inconsistent	Low	10	98.7	15.03	4.75	10.75
Partial	High	10	97.4	10.75	3.40	7.69
Partial	Low	10	94.2	10.24	3.24	7.32

kable(summarySE(aggregated.adult.mem, measurevar= "wm_ver_stand", groupvars = c( "consistency", "typefreq")), digits = 2)

consistency	typefreq	N	wm_ver_stand	sd	se	ci
Consistent	High	10	101.7	21.21	6.71	15.17
Consistent	Low	10	101.9	19.63	6.21	14.04
Inconsistent	High	10	111.5	16.70	5.28	11.95
Inconsistent	Low	10	105.3	13.46	4.26	9.63
Partial	High	10	101.5	10.58	3.34	7.56
Partial	Low	10	103.6	8.24	2.60	5.89

Compare Groups

Note that for analyses we use raw scores rather than standard scores

Children

Verbal Short Term Memory

anova.child.verSTM = aov(stm_ver_raw ~ consistency + typefreq, data = aggregated.child.mem)
summary(anova.child.verSTM)

##             Df Sum Sq Mean Sq F value  Pr(>F)   
## consistency  2  196.4   98.18   5.014 0.00872 **
## typefreq     1   11.4   11.38   0.581 0.44800   
## Residuals   86 1684.1   19.58                   
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

There is a significant difference between the semantic consistency groups. Run t-tests to find out which groups differ

# Get means
kable(summarySE(aggregated.child.mem, measurevar= "stm_ver_raw", groupvars = "consistency"), digits = 2)

consistency	N	stm_ver_raw	sd	se	ci
Consistent	30	20.13	3.89	0.71	1.45
Inconsistent	30	17.00	4.70	0.86	1.75
Partial	30	17.00	4.61	0.84	1.72

# T-tests

# Fully consistent versus Partially Consistent
t.test(subset(aggregated.child.mem, consistency == "Consistent")$stm_ver_raw, subset(aggregated.child.mem, consistency == "Partial")$stm_ver_raw)

## 
##  Welch Two Sample t-test
## 
## data:  subset(aggregated.child.mem, consistency == "Consistent")$stm_ver_raw and subset(aggregated.child.mem, consistency == "Partial")$stm_ver_raw
## t = 2.8447, df = 56.422, p-value = 0.006182
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
##  0.9272386 5.3394281
## sample estimates:
## mean of x mean of y 
##  20.13333  17.00000

# Fully Consistent versus Inconsistent
t.test(subset(aggregated.child.mem, consistency == "Consistent")$stm_ver_raw, subset(aggregated.child.mem, consistency == "Inconsistent")$stm_ver_raw)

## 
##  Welch Two Sample t-test
## 
## data:  subset(aggregated.child.mem, consistency == "Consistent")$stm_ver_raw and subset(aggregated.child.mem, consistency == "Inconsistent")$stm_ver_raw
## t = 2.8129, df = 56.065, p-value = 0.006756
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
##  0.9019869 5.3646797
## sample estimates:
## mean of x mean of y 
##  20.13333  17.00000

# Partially Consistent versus Inconsistent
t.test(subset(aggregated.child.mem, consistency == "Inconsistent")$stm_ver_raw, subset(aggregated.child.mem, consistency == "Partial")$stm_ver_raw)

## 
##  Welch Two Sample t-test
## 
## data:  subset(aggregated.child.mem, consistency == "Inconsistent")$stm_ver_raw and subset(aggregated.child.mem, consistency == "Partial")$stm_ver_raw
## t = 0, df = 57.979, p-value = 1
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
##  -2.405145  2.405145
## sample estimates:
## mean of x mean of y 
##        17        17

Short term verbal memory is significantly higher in the fully consistent condition compared to the other two conditions

Non-Verbal Short Term Memory

anova.child.nonverSTM = aov(stm_nonver_raw ~ consistency + typefreq, data = aggregated.child.mem)
summary(anova.child.nonverSTM)

##             Df Sum Sq Mean Sq F value Pr(>F)  
## consistency  2  149.0   74.48   2.881 0.0615 .
## typefreq     1  108.9  108.90   4.212 0.0432 *
## Residuals   86 2223.3   25.85                 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

There is a significant difference between high and low type frequency groups. Get means

kable(summarySE(aggregated.child.mem, measurevar= "stm_nonver_raw", groupvars = "typefreq"), digits = 2)

typefreq	N	stm_nonver_raw	sd	se	ci
High	45	13.64	5.47	0.81	1.64
Low	45	15.84	4.90	0.73	1.47

Verbal Working Memory

anova.child.WM = aov(wm_ver_raw ~ consistency + typefreq, data = aggregated.child.mem)
summary(anova.child.WM)

##             Df Sum Sq Mean Sq F value Pr(>F)  
## consistency  2   10.7    5.34   0.548 0.5801  
## typefreq     1   37.4   37.38   3.832 0.0535 .
## Residuals   86  838.8    9.75                 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

There is a significant difference between high and low type frequency groups. Get means

kable(summarySE(aggregated.child.mem, measurevar= "wm_ver_raw", groupvars = "typefreq"), digits = 2)

typefreq	N	wm_ver_raw	sd	se	ci
High	45	8.24	3.34	0.50	1.00
Low	45	9.53	2.85	0.42	0.86

Adults

Verbal Short Term Memory

anova.adult.verSTM = aov(stm_ver_raw ~ consistency + typefreq, data = aggregated.adult.mem)
summary(anova.adult.verSTM)

##             Df Sum Sq Mean Sq F value Pr(>F)
## consistency  2    7.2   3.617   0.267  0.766
## typefreq     1    6.7   6.667   0.493  0.486
## Residuals   56  757.8  13.533

Non-Verbal Short Term Memory

anova.adult.nonverSTM = aov(stm_nonver_raw ~ consistency + typefreq, data = aggregated.adult.mem)
summary(anova.adult.nonverSTM)

##             Df Sum Sq Mean Sq F value Pr(>F)
## consistency  2    1.0   0.517   0.039  0.962
## typefreq     1    4.3   4.267   0.318  0.575
## Residuals   56  750.4  13.401

Verbal Working Memory

anova.adult.verWM = aov(wm_ver_raw ~ consistency + typefreq, data = aggregated.adult.mem)
summary(anova.adult.verWM)

##             Df Sum Sq Mean Sq F value Pr(>F)
## consistency  2  111.4   55.72   1.765  0.181
## typefreq     1   11.3   11.27   0.357  0.553
## Residuals   56 1768.0   31.57

Add Memory Measures to LMEs

Summary of group difference

Children

• Non-verbal STM and Verbal WM: low TF > high TF (however, given the absence of TF effects in the main analyses we did not explore these differences further)
• Verbal STM: Fully consistent > Partially Consistent and Inconsistent
• To explore the latter effect we include “stm_ver_raw” in all models that compare performance between the fully consistent and partially or inconsistent conditions

Adult

• There were no group differences

Children: Verbal STM

Production: Trained Nouns

Fully Consistent consistent set as baseline

prodOLDchild <- droplevels(subset(prodCHILD, oldnew == "old"))
prodOLDchild <- lizCenter(prodOLDchild, list("typefreq", "session", "stm_ver_raw"))
prodOLDchild <- lizContrasts(prodOLDchild, prodOLDchild$consistency, "Consistent")

prodOLDchild.lmer1mem = glmer(det_correct ~
                          + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * typefreq.ct * session.ct + stm_ver_raw.ct
                          + (1 + session.ct|participant),
                          data = prodOLDchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodOLDchild.lmer1mem)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.908	0.089	10.231	0.000
Consistent_VS_Partial	-0.706	0.221	-3.200	0.001
Consistent_VS_Inconsistent	-0.797	0.221	-3.602	0.000
typefreq.ct	0.296	0.175	1.695	0.090
session.ct	1.112	0.106	10.495	0.000
stm_ver_raw.ct	-0.014	0.015	-0.927	0.354
Consistent_VS_Partial:typefreq.ct	-0.307	0.430	-0.715	0.474
Consistent_VS_Inconsistent:typefreq.ct	0.080	0.433	0.184	0.854
Consistent_VS_Partial:session.ct	-0.894	0.255	-3.510	0.000
Consistent_VS_Inconsistent:session.ct	-0.856	0.257	-3.334	0.001
typefreq.ct:session.ct	0.372	0.204	1.826	0.068
Consistent_VS_Partial:typefreq.ct:session.ct	-0.611	0.501	-1.220	0.222
Consistent_VS_Inconsistent:typefreq.ct:session.ct	0.360	0.505	0.713	0.476

Production: Novel Nouns

prodNEWchild <- droplevels(subset(prod1, age == "Child" & oldnew == "new" & consistency != "Inconsistent"))
prodNEWchild <- lizCenter(prodNEWchild, list("consistency", "typefreq", "session", "stm_ver_raw"))

prodNEWchild.lmer1 = glmer(det_correct ~
                          + consistency.ct * typefreq.ct * session.ct + stm_ver_raw.ct
                          + (1 + session.ct|participant),
                          data = prodNEWchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(prodNEWchild.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.360	0.105	3.435	0.001
consistency.ct	-0.735	0.220	-3.337	0.001
typefreq.ct	-0.141	0.209	-0.677	0.499
session.ct	0.390	0.094	4.138	0.000
stm_ver_raw.ct	-0.023	0.021	-1.087	0.277
consistency.ct:typefreq.ct	-0.354	0.419	-0.845	0.398
consistency.ct:session.ct	-0.510	0.188	-2.715	0.007
typefreq.ct:session.ct	-0.118	0.184	-0.643	0.520
consistency.ct:typefreq.ct:session.ct	0.226	0.368	0.613	0.540

2AFC: Trained Nouns

Fully consistent versus Partially consistent

afcOLDchild <- droplevels(subset(afc, age == "Child" & oldnew == "old"))
afcOLDchild <- lizCenter(afcOLDchild, list("typefreq", "stm_ver_raw"))
afcOLDchild <- lizContrasts(afcOLDchild, afcOLDchild$consistency, "Consistent")

afcOLDchild.lmer1 = glmer(correct ~
                         + (Consistent_VS_Partial + Consistent_VS_Inconsistent) * typefreq.ct + stm_ver_raw.ct
                         + (1|participant),
                         data = afcOLDchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(afcOLDchild.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	1.103	0.132	8.352	0.000
Consistent_VS_Partial	-0.657	0.328	-2.002	0.045
Consistent_VS_Inconsistent	-0.703	0.326	-2.155	0.031
typefreq.ct	0.076	0.253	0.302	0.763
stm_ver_raw.ct	0.023	0.029	0.793	0.428
Consistent_VS_Partial:typefreq.ct	0.018	0.628	0.028	0.978
Consistent_VS_Inconsistent:typefreq.ct	-0.198	0.631	-0.314	0.754

2AFC: Novel Nouns

afcNEWchild <- droplevels(subset(afc, age == "Child" & oldnew == "new" & consistency != "Inconsistent"))
afcNEWchild <- lizCenter(afcNEWchild, list("consistency", "typefreq", "stm_ver_raw"))

afcNEWchild.lmer1 = glmer(correct ~
                          + consistency.ct * typefreq.ct + stm_ver_raw.ct
                          + (1|participant),
                          data = afcNEWchild, family = binomial, control = glmerControl(optimizer = "bobyqa"))
kable(summary(afcNEWchild.lmer1)$coefficients, digits = 3)

	Estimate	Std. Error	z value	Pr(>|z|)
(Intercept)	0.678	0.193	3.504	0.000
consistency.ct	-1.111	0.407	-2.734	0.006
typefreq.ct	-0.378	0.378	-1.001	0.317
stm_ver_raw.ct	0.045	0.045	0.989	0.323
consistency.ct:typefreq.ct	0.337	0.756	0.445	0.656