Current Research Questions
Measurement and Modeling of Psychological Constructs
In this line of research, we examine the implications and consequences of one’s choice of measurement model. Latent variable models (e.g., structural equation models) have become firmly established as the gold standard in psychological measurement, because they are able to account for measurement error and disentangle that error from reliable sources of variation. As a result, it is increasingly common to see these models used in all areas of psychological research, to “deal with” measurement error. But these models can create new problems (e.g., biased estimates) when the model assumptions are not met, potentially leading to incorrect scientific inferences (e.g., Rhemtulla, van Bork, & Borsboom, 2020). Our ongoing research examines alternative measurement models and explores ways to diagnose these problems in empirical research.
Improving Robustness and Reliability of Psychological Research
Many fields of psychology are grappling with the realization that entrenched research methods have produced a body of work that often fails to replicate, prompting renewed interest in developing and adopting improved research methods. We collaborate with researchers in developmental, social, and neuropsychology to develop and disseminate methods that will allow researchers to address their questions in a more robust way to produce better science.
Advancing Methods for Network Psychometrics
Another line of research in our lab examines methods and practices within the young field of network psychometrics, in which network models –typically undirected graphical models (a.k.a., Markov random fields) –are fit to data from a set of observed variables as a way to understand the structure of a psychological construct. These models have become popular as an alternative to latent variable models, because they have the potential to foster a finer-grained understanding of how particular observed indicators of a construct relate to each other. Being a relatively new set of models for psychological data, though, there are many open methodological questions. We work on some of those questions, such as: which estimation methods produce network estimates with the best statistical properties? What factors lead networks to be more or less likely to replicate in new data?