John Dziak, Ph.D.
Research Associate, The Methodology Center
The Methodology Center
The Pennsylvania State University
427 Health and Human Development Building
University Park, PA 16802
Ph.D., The Pennsylvania State University, 2006 (Statistics)
M.S., The Pennsylvania State University, 2004 (Statistics)
M.A., The Catholic University of America, 2001 (Psychology)
B.S., The University of Scranton, 1999 (Psychology)
Research Interests and Collaborators
One of my main roles here at the Center is working on software development, in particular for our SAS procedures. I helped expand the functionality of our latent class analysis procedure PROC LCA, to allow for standard error estimation and for pseudolikelihood estimation with sampling weights. Additionally, I’ve been working with Bethany Bray and Stephanie Lanza on how best to include later outcome variables into latent class analysis (LCA) models, and the implications of different existing approaches to doing this. I help with simulating different scenarios and comparing outcomes of different methods under those scenarios.
I am very interested in the analysis of longitudinal data and am working with Runze Li, Xianming Tan, Anne Buu, and Stephanie Lanza on research related variations of the time-varying effect model (TVEM) for specific situations such as nonnumerical data, weighted survey data, or latent mixed populations. I have also been studying a related model, scalar-on-function functional regression, in which a time-varying predictor has a time-varying contribution to a non-time-varying outcome. In the usual TVEM, the outcome itself is also varying over time.
I’ve also been working with Linda Collins, Kari Kugler, and Inbal Nahum-Shani on the uses of factorial experimental designs for studying multiple-factor interventions. One area of interest is in the special challenges that arise in common situations in which study participants must be treated in clusters such as teams, schools or hospitals rather than being treated as individuals, and how this affects statistical power and required sample size. Another is how best to convert analyses of the results of factorial experiments into practical recommendations for improving interventions, in terms of efficacy or of efficiency.
I’m also very interested in model selection. My thesis had to do with the application of the SCAD (smoothly clipped absolute deviation) complexity penalty to functions other than likelihoods, such as generalized least squares or quadratic inference functions.