- Location: Hobbs Hall • 1818 Horton Ave • Nashville, TN 37212
- Room: 105
- Contact: Kris Preacher
- Email: firstname.lastname@example.org
- Audience: Free and Open to the Public
Society of Multivariate Experimental Psychology (SMEP) Conference Talks
Structural Equation Modeling Strategies for Assessing Within-Participant Mediation
Kristopher J. Preacher, Vanderbilt University, Department of Psychology and Human Development
In contrast to the dominant between-participant framework for assessing mediation, Judd, Kenny, and McClelland (2001) proposed a three-step regression-based method for situations where participants are measured on both a mediator and an outcome under at least two treatment conditions. Building on Judd et al.’s method, Montoya and Hayes (2017) proposed a path-analytic framework permitting direct estimation of within-participant indirect effects, the generation of bootstrap confidence intervals, and the inclusion of multiple mediators. Here I extend Montoya and Hayes’ (2017) method to a full structural equation modeling approach. The extension (a) circumvents the need to manually compute difference scores, (b) permits the use of latent variables with multiple observed indicators, (c) employs modern missing data handling techniques for repeated measures of the mediator and/or outcome, and (d) avoids some common criticisms associated with the use of observed difference scores. Relationships to other approaches for assessing within-participant mediation are elaborated.
A problem, a method, and a dataset: Addressing selection bias using sibling control methods with the NLSY data
Joseph Rodgers, Vanderbilt University, Department of Psychology and Human Development
Do small families create smart children, or do smart parents make small families? Do mothers smoking during pregnancy cause problem behaviors in their children, or would the children of mothers who smoke have problems behaviors anyway? Do smart teens delay age at first intercourse because of their intelligence, or do other background factors cause smart teens and delayed AFI to co-occur? Challenges to internal validity, the validity of causal inference, abound in social/behavioral science research – especially in quasi-experimental design settings. Tenuous causal directionality is often confidently asserted by sophisticated researchers, though often without apparent awareness of the threats to validity. Perhaps the most pernicious problem is the challenge of selection bias. Methods to handle selection bias in quasi-experimental research settings are in their ascendancy, including using covariates, instrumental variable approaches, and propensity score methods. Another powerful design methodology to help address selection bias is the discordant sibling design. A data source with flexible longitudinal, within-family, and cross-generational data patterns is the National Longitudinal Survey of Youth, which has three separate data sources – the original NLSY79 sample, the NLSY-Children/Young Adult Sample, and the NLSY97 replication sample. Using maternal sibling pairs from the NLSY79, and their biological offspring from the NLSY-C/YA, we can control for most background sources of unobserved heterogeneity. I describe the dataset, the design, and findings from published studies using this methodology to address selection bias. To anticipate several findings, suggested above: Small families do not create smart children; smart parents do make small families. Maternal smoking does not cause problem behaviors in children. Intelligence is not what causes smart teens to delay first intercourse.