2-131 - Evaluating and Improving the Validity of Causal Inferences in Child Development Research

Session Type: Paper Symposium

Integrative Statement

Valid causal inferences are required to move beyond descriptions of developmental data to obtain sufficiently accurate explanations of those data to inform optimization of human development via recommendations implied by empirical results. This symposium includes four presentations to improve the validity of causal inferences in child development research. The first paper summarizes how econometrics is improving the most valid quasi-experimental method, the regression discontinuity design. The second presentation shows how the magnitudes of longitudinal associations of academic achievement at age 16 with earnings in adulthood vary by the number and type of covariates in the model. The remaining associations also vary by age during the lifespan. The third presentation illustrates the recently developed Random-Intercepts Cross Lagged Panel Model (RI-CLPM) to isolate within-person effects by removing confounds due to between-person differences. Applying it to the question whether privacy invasion by parents predicts increased adolescent secrecy, isolating processes at the level of the family unit, indicates reverse theoretical inferences compared to a traditional cross-lagged panel model. The final presentation illustrates various longitudinal analyses of the effects of medication and therapeutic interventions on mothers' physical and mental health. Correlations and the most common longitudinal analyses make all those corrective interventions look harmful, presumably due to selection bias that is only partially controlled for. It explored whether RI-CLPM analyses could produce results more compatible with current medical evidence, but those analyses still show adverse effects for half of the cross-lagged coefficients, despite ruling out all time-invariant covariates.

Sub Unit

• Methods, History, Theory

Chair

• Robert E Larzelere, Oklahoma State University

Individual Presentations

• Exploiting Naturally Occurring Discontinuities to Infer Causal Effects - Presenting Author: Felix Thoemmes, Cornell University

• Revisiting the Correlation Between Adolescent Academic Achievement and Adult Economic Success - Presenting Author: Tyler Watts, University of California, Irvine

• Want to Know a Secret? Using Random-Intercept Cross-lagged Panel Models to Clarify the Link Between Secrecy and Privacy Invasion - Presenting Author: Loes Keijsers, Tilburg University; Evelien Dietvorst, Tilburg University; Ellen Hamaker, Utrecht University; Marieke Hiemstra, Utrecht University

• Trying to Overcome Selection Bias in Longitudinal Analyses of Corrective Actions by Professionals - Presenting Author: Robert E Larzelere, Oklahoma State University; Ronald Cox, Oklahoma State University; Isaac Washburn, Oklahoma State University; Hua Lin, Oklahoma State University