On January 23, 2018 The RMC will host Dr. Laura Stapleton who will present, “Using Confirmatory Factor Analysis with Data from Multistage Sample Designs,” in the EHE Commons, Ramseyer 260, from 10:00 – 11:30 am.
Laura M. Stapleton is a Professor in Measurement, Statistics and Evaluation (EDMS) in the Department of Human Development and Quantitative Methodology at the University of Maryland. Additionally, she serves as the Associate Director of the Research Branch of the Maryland State Longitudinal Data System Center. She joined the faculty of EDMS in Fall 2011 after being on the faculty in Psychology at the University of Maryland, Baltimore County and in Educational Psychology at the University of Texas, Austin. Each year she also serves on the faculty of the National Center for Education Research (NCER) funded Summer Research Training Institute on Cluster Randomized Trials at Northwestern University. Prior to earning her Ph.D. in Measurement, Statistics and Evaluation from the University of Maryland in 2001, she was an economist at the Bureau of Labor Statistics and, subsequently, conducted educational research at the American Association of State Colleges and Universities and as Associate Director of institutional research at the University of Maryland.
Meetings with Dr. Stapleton on the afternoon of January 23 are available to faculty, students, and researchers who would like to consult with her about their work. To inquire, please contact Sandy Reed at email@example.com.
In social science research, latent constructs are often inferred from sets of items intended to measure those constructs. When data are collected via multistage sampling designs (e.g., students within schools or teachers within districts) the construct of interest might exist at multiple levels. In this talk, I will consider how researchers might approach construct meaning and construct validation when working with data that are nested. I will first present extensions of the single-level confirmatory factor analysis (CFA) approach to a simple multilevel CFA (MCFA) when data are nested. I then will wade through the murky conceptual landscape that exists when considering measurement models at both the individual and cluster levels and introduce conceptual distinctions between constructs across levels and among different types of constructs at the cluster level. Specifically, I will discuss how items might be used to measure “shared” and “configural” cluster-level constructs. While shared constructs would reflect a shared element of the cluster (wherein individuals would be viewed as exchangeable within a cluster), configural constructs represent aggregation of characteristics of the individuals within the cluster. Importantly, an often-overlooked characteristic of configural constructs would be an evaluation of dispersion within clusters. Additionally, although empirical data may show cluster dependency, theoretically the construct may be an individual level one only but the data reflect a spurious intraclass correlation (ICC) or a spurious contextual effect due to measurement non-invariance. Under certain circumstances, specifically, when the composition of individuals differs across clusters and there is measurement non-equivalence at the item level, the data will yield a spurious ICC in the construct validation process. Monte Carlo simulation results are used to demonstrate the concepts.
To RSVP for this event, click here