A lot of the research and consulting work I do involves intensive statistical computing. I also dabble a bit in R development.

Statistical Computing

If I had to give myself an academic label, it would probably be “computational statistician”. A great deal of my work revolves around statistical computing. In my methodological research, I use Monte Carlo simulation as the primary tool by which to explore the performance of the methods I’m developing or evaluating. As these methods tend to be some flavor of multiple imputation—which is already computationally intensive—these simulations often place prohibitive demands on computing resources. Hence, I have become a serial abuser of high-performance computing (HPC) clusters such as the Lisa Compute Cluster. My eternal gratitude goes out to the folks at SURF who administer Lisa and many similar services and provide excellent support for their products.

Likewise, most of my collaborative work is computational. In my experience, substantive researchers don’t need statistical experts to help them run t-tests or simple linear models, but these simple methods are often inadequate. As my coauthors and I argue in this paper, plausible representations of modern social-scientific theories—and reasonable tests of the associated hypotheses—often require advanced modeling approaches such as structural equation modeling, multilevel modeling, conditional process analysis, or Bayesian modeling. Similarly, when working with messy real-world data, the asymptotic assumptions underlying simple, traditional statistical tools are often grossly violated. In such cases, computationally intensive methods like cross-validation or resampling methods (e.g., bootstrapping, permutation tests) are often required to support reasonable statistical modeling, inference, and prediction.

Development

As a natural consequence of spending so much of my time immersed in statistical computation, I also dabble in software development, particularly R package development. I created my first R package (MIBRR) after realizing that my life would be much easier if I packaged the code I was using to implement the imputation algorithm I was developing for my PhD project. During my postdoc at Texas Tech University, one of my primary tasks was developing the PcAux R package. After a few more people joined the PcAux project, I took on the role of development team lead.

I have also done a few non-statistical programming projects such as this utility that I developed for the Tilburg University Student Administration to merge the results from the online and on-campus versions of hybrid exams.

Pretty much all of my potentially interesting programming/development work lives on my GitHub page. If you would like to contribute to any of my projects, feel free to submit a pull request.