Date: November 17, 2011
Time: 12:30PM to 1:30PM
Location: Room 241 Vanderbilt Kennedy Center, One Magnolia Circle
Sonya K. Sterba, Ph.D., Assistant Professor of Psychology (Peabody College)
Abstract: Power analyses are now requested by most granting agencies. In their introductory statistics courses, researchers are often taught how to conduct power analyses for hypothesis tests involving means, correlation coefficients, regression coefficients and/or population R-squared. However, when researchers' analysis goals require a latent variable model—such as a factor analysis, structural equation, or latent growth curve model—they may be unsure how to proceed with a power analysis. In latent variable models, hypotheses could be tested about different things: (1) individual parameters of a given model, (2) the overall fit of a single model in isolation, and (3) differences in fit between competing models. Different power analysis procedures are suited for each of these three hypothesis testing situations. This pedagogical talk will review approaches for conducting power analyses involving these three situations. These approaches will be illustrated for a latent growth curve model.
1. Brief refresher on power analysis concepts
2. Power analysis procedures for three different analysis goals in latent variable modeling
3. Illustration of each procedure and brief discussion of software implementation
1. To understand the logic of power analyses in a latent variable modeling context
2. To obtain introductory information on how to implement such power analyses in practice using available software
3. To understand the strengths and limitations of alternative procedures and what resources to consult for further information
Intended Audience: Researchers who use latent variable models but who are unacquainted with power analysis procedures for such models.
Speaker Description: Sonya Sterba is an Assistant Professor in the Quantitative Methods and Evaluation Program within the Department of Psychology and Human Development. She teaches graduate courses on Latent Growth Curve Modeling (Spring, 2012) and Applied Latent Class and Mixture Modeling (Fall, 2011). Her research involves evaluating the performance of latent variable models for longitudinal and cross-sectional data and involves extending such models to address research questions in developmental psychopathology.
Lunch will be provided. Please register by November 15.
Co-sponsored with the Department of Biostatistics.