Data Analysis Plan
We intend to use general linear mixed models (GLMM) using the package
lme4 in R to examine the effects of language forms and kinds of
properties (i.e., content) on different aspects of essentialist beliefs
(i.e., causal explanation and heritability). To analyze children’s data,
we plan is to run stepwise mixed-effect logistic regression models for
each of the measure (Explanation and Heritability). We will start with a
full model on each measure coded as a binomial variable
(essentialist/non-essentialist) with language forms (generic, specific),
kinds of properties (biological, cultural), age (continuous), and all
two way and the three-way interactions as fixed effects, and ID and
scenario as random effects. The non-significant interaction terms will
then be dropped in subsequent models and the model will be reduced
step-wise until we have the best fitting model. When applicable, we will
transform coefficients to odd ratios as measures of effect size.
Significant interactions will be examined using simple slope tests.
For the Explanation task, we also will compare children’s responses to
test properties (mentioned in the storybook) and their responses to the
control property (cross-condition properties) for each of the
experimental conditions. This will be a logistic regression model with
property (test, control), age, and the interaction terms as fixed
effects and ID and scenario as random effects. In addition, responses to
switched-at-birth questions will be compared with the chance rate using
t tests.
A similar procedure will be used for the adult sample, except that these
models would not include age. We also plan to analyze a full sample with
a similar plan and include age group as a categorical factor. This
enables us to explore any developmental shifts from childhood to
adulthood in how generics lead to essentialism.