Society for Research in Educational Effectiveness (SREE) Conference
Theme: Tensions and Tradeoffs: Responding to Diverse Demands for Evidence
March 6-9, 2019
Title: Reciprocal Effects of Reading and Mathematics? Beyond the Cross-Lagged Panel Model (Paper)
Authors: Drew Bailey, Yoonkyung Oh, George Farkas, Paul Morgan, Marianne Hillemeier
The cross-lagged panel model (CLPM) has long been a popular methodological approach for investigators with observational panel data seeking to estimate the mutual effects of two variables on one another. We re-estimate cross-lagged relations between reading and mathematics achievement using alternative models that better allow for the possibility that the achievement measures are differentially affected by unmeasured confounds throughout the elementary grades. We analyzed data from the public version of the Early Childhood Longitudinal Study –Kindergarten Class of 2010-11 (ECLS-K: 2011), which is a large-scale, longitudinal study of a nationally representative sample of U.S. children who were in kindergarten during the 2010-11 academic year and were followed through the fifth grade. We present estimates from a total of 8 models: (1) a CLPM with autoregressive and cross-lagged paths freely estimated, (2) another CLPM with autoregressive and cross-lagged paths held constant across waves, (3-4) a ST-CLPM and a RI-CLPM with autoregressive and cross-lagged paths freely estimated but factor loadings held constant across waves, (5-6) a ST-CLPM and a RI-CLPM with autoregressive and cross-lagged paths constrained across waves but factor loadings freely estimated across waves, and (7-8) a ST-CLPM and a RI-CLPM with autoregressive and cross-lagged paths and factor loadings held constant across waves. The modifications of the CLPM estimated here fit the longitudinal math and reading test score data better than the original CLPM and provide very different substantive interpretations than were provided by those models. The cross-lagged effects of math at time t to reading at t+1, or of reading at time t to math at t+1 are generally small but positive; to the extent they are asymmetrical, the effects from earlier reading to later math appear stronger than the effects from earlier math to later reading, a reversal of what is found in the CLPM in the same data.