PPVT & Associative Strength
Correlation matrix of child-level variables (overall performance)
propIncorrect_choseAssoc is the proportion of incorrect trials where there were 1-2 semantically-related words and the child chose one of them.
child_level_vars <- select(child_item_vars, subjID, Age_years, totalAttempted, totalCorrect, totalIncorrect,
num_incorrect_choseAssociate, propPosAssn_correct, propNoAssn_correct, propCorrect,
propIncorrect, propIncorrect_choseAssoc) %>%
distinct()
child_level_corrs <- select(child_level_vars, propCorrect, propPosAssn_correct, propNoAssn_correct,
propIncorrect_choseAssoc, Age_years) %>%
cor(use="pairwise.complete.obs", method="pearson")
p.mat_child <- cor.mtest(child_level_corrs)
pMatrix_child <- p.mat_child$p
corrplot(child_level_corrs, method = 'color', type='lower', diag = TRUE, addCoef.col = "grey",
tl.col = "black", number.font=2, number.cex=.8, p.mat=pMatrix_child, sig.level = 0.05, insig = "blank")
How does associative strength affect the likelihood of an trial being correct, controlling for age & word difficulty (using AOA as proxy)?
summary(glmer(is.Correct ~ Age_years + KupermanAOA + sumStrength + (1|subjID), child_item_vars, family="binomial"))## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: binomial ( logit )
## Formula: is.Correct ~ Age_years + KupermanAOA + sumStrength + (1 | subjID)
## Data: child_item_vars
##
## AIC BIC logLik deviance df.resid
## 8087.8 8123.1 -4038.9 8077.8 8591
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -8.3874 0.1957 0.3646 0.5299 1.8111
##
## Random effects:
## Groups Name Variance Std.Dev.
## subjID (Intercept) 0.3768 0.6139
## Number of obs: 8596, groups: subjID, 193
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 1.92002 0.15625 12.288 < 2e-16 ***
## Age_years 0.55305 0.03441 16.073 < 2e-16 ***
## KupermanAOA -0.52089 0.02094 -24.871 < 2e-16 ***
## sumStrength -3.05610 0.58874 -5.191 2.09e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) Ag_yrs KprAOA
## Age_years -0.610
## KupermanAOA -0.377 -0.432
## sumStrength 0.043 -0.060 -0.087Correlation matrix of item-level variables
Formula used to get t-value for each item: totalCorrect ~ is.Correct + Age_years + totalAttempted
item_level_vars <- select(child_item_vars, item, BlockNumber, WordNumber, sumStrength, numNonZero,
hyper_z, KupermanAOA, itemMeanCorrect, n_responded, tval_item) %>%
distinct()
item_level_corrs <- select(item_level_vars, sumStrength, numNonZero, hyper_z,
KupermanAOA, itemMeanCorrect, tval_item) %>%
cor(use="pairwise.complete.obs", method="pearson")
p.mat_item <- cor.mtest(item_level_corrs)
pMatrix_item <- p.mat_item$p
corrplot(item_level_corrs, method = 'color', type='lower', diag = TRUE, addCoef.col = "grey",
tl.col = "black", number.font=2, number.cex=.8, p.mat=pMatrix_item, sig.level = 0.05, insig = "blank")
No correlation between t-value and sumStrength, but are there individual words that are high on both?
ggplot(filter(item_level_vars, sumStrength > 0), aes(sumStrength, tval_item))+
geom_point()+
geom_label_repel(aes(label=ifelse((tval_item>2)|(sumStrength>.3), as.character(item),'')),
box.padding=.35,
point.padding=.2)+
theme_classic()## Warning: Removed 1 rows containing missing values (geom_point).## Warning: Removed 1 rows containing missing values (geom_label_repel).
posAssn_highT_words <- c("penguin","clarinet","canoe","clamp","fly","castle","boulder",
"group","aquarium","vine","flamingo","carpenter","pastry",
"wrench","cactus","empty","vest","tusk","arrow","interior")
associations <- read.csv("associativeStrength_byTrial.csv") %>%
select(-X) %>%
rename(item=TargetWord) %>%
filter(item %in% posAssn_highT_words) %>%
arrange(desc(numNonZero))
DT::datatable(associations)How does PPVT performance predict W-J Science?
- Specifically looking at performance on positive-association (i.e., harder) trials or performance on incorrect trials where they chose a semantically related lure
- Controlling for age and PPVT percentile
child_vars_DAS <- select(child_item_vars_DAS, subjID, totalAttempted, totalCorrect, propCorrect, Age_years,
propPosAssn_correct, propNoAssn_correct, num_incorrect_choseAssociate,
propIncorrect_choseAssoc, WJNumCorrect, PPVTRawScore, PPVTStandardScore, PPVTPercentile) %>%
distinct()
summary(lm(WJNumCorrect ~ Age_years + PPVTPercentile + propPosAssn_correct, child_vars_DAS))##
## Call:
## lm(formula = WJNumCorrect ~ Age_years + PPVTPercentile + propPosAssn_correct,
## data = child_vars_DAS)
##
## Residuals:
## Min 1Q Median 3Q Max
## -5.2905 -1.8229 0.0659 1.4067 7.6815
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1.85078 2.94400 -0.629 0.53200
## Age_years 1.71710 0.19392 8.855 2e-12 ***
## PPVTPercentile 0.05563 0.01731 3.214 0.00212 **
## propPosAssn_correct 2.49732 3.47633 0.718 0.47536
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.655 on 59 degrees of freedom
## (4 observations deleted due to missingness)
## Multiple R-squared: 0.6243, Adjusted R-squared: 0.6052
## F-statistic: 32.69 on 3 and 59 DF, p-value: 1.416e-12summary(lm(WJNumCorrect ~ Age_years + PPVTPercentile + propIncorrect_choseAssoc, child_vars_DAS))##
## Call:
## lm(formula = WJNumCorrect ~ Age_years + PPVTPercentile + propIncorrect_choseAssoc,
## data = child_vars_DAS)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.0766 -1.9960 0.0346 1.4919 6.7947
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.86849 2.48092 0.350 0.7277
## Age_years 1.63242 0.21273 7.674 4.19e-10 ***
## PPVTPercentile 0.05088 0.02040 2.494 0.0159 *
## propIncorrect_choseAssoc 3.27901 6.88471 0.476 0.6359
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.683 on 52 degrees of freedom
## (11 observations deleted due to missingness)
## Multiple R-squared: 0.5494, Adjusted R-squared: 0.5234
## F-statistic: 21.14 on 3 and 52 DF, p-value: 4.367e-09