As a quantitative psychologist and statistician, my goal is to advance the quantitative, social, and behavioral sciences by developing, evaluating, and applying statistical, psychometric, and computational methods that provide researchers with tools for answering their research questions.
The focus of my current research program is in the area of generalized latent variable modeling, specifically on a family of models broadly known as diagnostic classification models (DCMs). DCMs are generalized latent variable models – in fact, they are restricted latent class models – that model individuals’ cognitive or psychological states (e.g., symptoms associated with a particular psychological disorder) based on their response patterns to a set of manifest variables. Different from traditional latent variable modeling frameworks, DCMs typically operationalize latent variables as categorical rather than continuous. In the context of measurement of depression, for instance, the categorical latent variables – generically called attributes – may represent the set of depressive symptoms as defined by the DSM-V such that symptoms are either “active/present” or “inactive/not present.” Whereas traditional latent variable models such as factor analysis would provide an overall “depression score” on some continuum, DCMs instead perform classification of individuals by providing symptom profiles – generically called attribute profiles – that describe the specific pattern of depressive symptoms individuals displayed according to their response patterns. Hence, DCMs provide rich and nuanced information about psychological constructs that traditional latent variable models cannot without extensive post-hoc analyses. Due in part to their generality and flexibility, DCMs have been applied in several domains across the social, behavioral, and psychological sciences to study personality structures, evaluate cognitive skills, and to identify psychological disorders such as bipolar disorder, major depressive disorder, schizotypal personality disorder, and pathological gambling.
A major research interest of mine is in developing efficient and scalable estimation algorithms for generalized latent variable models (GLVMs). GLVMs are notorious for being difficult to estimate, especially in high-dimensional settings if the number of latent variables in the model is large. This is primarily due to a phenomenon known as the ``curse of dimensionality" which fills the data likelihood with intractible integrals that need to be approximated with numerical methods. In my work, I have explored and developed novel estimation techniques for addressing this issue from a variety of perspectives, including Bayesian estimation algorithms, maximum likelihood estimators, and variational inference methods. For example, my dissertation work is exploring the use of Hamiltonian dynamics with Gibbs sampling for estimation of diagnostic classification models within a fully Bayesian framework. Examples of my work with maximum likelihood estimators and variational inference algorithms can be found here and here, respectively.
I am currently interested on the development of efficient maximum likelihood estimators for generalized structural equation models and currently have a few projects underway. More information will be posted here as it becomes available.
I am intersted in the development of novel Bayesian methods. My previous work has focused on identifying the use of cross-loadings in higher-order factor analysis models. Recent work my colleagues Dr. Hyeri Hong and Dr. Walter Vispoel has focused on investigating the use of exploratory SEM and Bayesian SEM with and without the use of cross-loadings.
As a quantitative methodologists, collaborating with scholars from other disciplines within the social and behavioral sciences is one of the most satisfying aspects of my work. Previous research efforts have been to study the impact of self-efficacy on healthcare career interest, the development of psychometrically and emprically valid measures of ability in ESL assessments, and so on.