Applying Linear Mixed Effects Models with Crossed Random Effects to Psycholinguistic Data: Multilevel Specification and Model Selection.

Abstract

Abstract Abstract Applying linear mixed effects regression (LMER) models to psycholinguistic data was made popular by Baayen, Davidson, and Bates (2008). However, applied researchers sometimes encounter model specification difficulties when using such models. This article presents a multilevel specification of LMERs customized for typical psycholinguistic studies. The proposed LMER specifications with crossed random effects allow different combinations of random intercept effects or random slope effects to be specified directly for subject and item covariates. As a result, this approach allows researchers to describe, specify, and interpret a wide range of effects in an LMER more easily. Next, the syntax and steps involved in using the PROC MIXED procedure in SAS to fit the discussed models are illustrated. Thirdly, various issues relating to model selection, specifically for the random component of LMER models with crossed random effects, are discussed. Finally, this article concludes with remarks about model specification and selection of the random structure in the context of analyzing psycholinguistic data using LMERs specifically. This paper provides readers conducting psycholinguistic research with a complete tutorial on how to select, apply, and interpret the multilevel specification of LMERs.

Correlates of health in adolescent elite athletes and adolescents: A cross‐sectional study of 1016 adolescents

“…The models were tested for random effect with varying intercepts and varying slopes of the competence-based self-esteem variable, with a variance component covariance structure. The best-fitted model, as determined by the Akaike Information Criterion (AIC) was chosen (Yu, 2015). This resulted in that a final model without varying intercepts and slopes was chosen for elite athletes and a final model with varying intercepts and slopes was chosen for non-elite adolescents (Supplementary file).…”

From Play to Language: Infants’ Actions on Objects Cascade to Word Learning

“…We selected the best fitting, most parsimonious model for RQ1 and RQ2. Specifically, we attempted to fit models with maximal random structure (Barr, Levy, Scheepers, & Tily, 2013), and if the model did not converge, we performed likelihood ratio tests to reduce the model (H. T. Yu, 2015). Each term was retained in the final model if it improved model fit relative to a model without the term (as described in the Results).…”

Specifying the random effect structure in linear mixed effect models for analyzing psycholinguistic data

“…Multiple software packages can implement LMEMs, including SAS (e.g., Yu, 2015), and R (e.g., Bates, Maechler, Bolker, & Walker, 2015). Several books and papers (e.g., Baayen et al, 2008) are often credited with popularizing the use of the R package lme4 (Bates, 2005;Bates, Maechler, Bolker, & Walker, 2015)…”

“…In the LMEM context, model selection aims to determine the fixed and random effects to include in the model. However, the best method for selecting the random component structure is particularly unclear (Barr et al, 2013;Yu, 2015).…”