In our series Interview With A Team Member, we aim to introduce the people behind the JASP project. Today we are interviewing Erik-Jan van Kesteren, one of our software developers.
Erik-Jan is a software developer. At JASP, is responsible for adding plots, functions, UI elements, and interfacing R and C++.
What is your professional background?
I am currently a PhD candidate at Utrecht University, working on extending Structural Equation Modeling for novel types of data. For example: how can we run SEM-like models when we have more variables than cases? I love the job because it allows me to keep learning interesting concepts around probability and statistics, data science, and computer science, and, best of all, I can combine it well with my work for JASP!
What is your favorite statistical test?
Well, because of what I study I should say I really like SEM for how intuitive it is to graphically specify the ideas you have about the world in a graphical way. SEM allows you to statistically test systems as you would imagine them in theory. But secretly my favourite statistical method is linear regression. I used to consider it as one of the more ‘basic’ methods, but its elegant simplicity is deceptive and it continues to surprise me. It has a very strong connection to a relatively new love of mine (linear algebra), and it’s also the building block for many other methods in virtually every field of statistics. Its extensions rival the performance of the most fancy ‘machine learning’ and ‘artificial intelligence’ methods. If I were to be put on an island with only one statistical method, it would be linear regression.
What is your relation to JASP?
I started at JASP in 2016 to do an internship for my masters in Methodology & Statistics. There, I performed some research and implemented the VS-MPR, an upper-bound for the Bayes Factor based only on a transformation of the p-value. Since then, I kind of stuck around because I got more and more into programming. When I went into my PhD project, it was clear to me that I would want to continue working at JASP to learn more about software development, open source, and to make it easier for scientists to do statistical analysis. Also, the colleagues are really nice 🙂
What feature of JASP do you like best?
I really like being able to click on your output to change the options. But one of the more recent killer features of JASP is the extensive data library, introduced in version 0.9. It is really nice to be able to have datasets and example analyses of many statistical tests at your fingertips. I can just get lost in trying out different datasets! My favourite one is the Titanic dataset (in the regression folder).
What aspect of JASP would you like to see improved in a future version?
Maybe it does not have a lot to do with the usability, but I would love to see JASP become more approachable for developers. Although we are making great progress on this topic, specialized knowledge of the JASP architecture is still needed to implement an analysis into JASP. A more user-centric change I would like to see is plot editing, so users can change the look and content of their plots before saving them.
Are you a Bayesian, a frequentist, an agnostic, a pragmatist, or perhaps something else?
I like Bayesian estimation a lot: it’s intuitive, it formalises what we know beforehand, and I weirdly love probability distributions. But I am not a dogmatist: I like any method that allows us to answer questions about the world, as long as it incorporates the level of uncertainty about this answer. For example, the scatter plot is a great tool that shows the level and strength of a relationship between two variables, as well as uncertainty through shape and density.
What question would you like to answer?
What is your favourite probability distribution?
The Beta distribution. I like how many different shapes it can take and the fact that it is bounded between 0 and 1. It also works really neatly as a conjugate prior for a binomial test.