Data analysis

Network analysis, Bayesian Data Analysis, and pretty graphics with a story. In R (mostly).

I am passionate about thorough and reproducible ways to link data analysis to questions about environmental governance, policy and sustainability. I take an unreasonable amount of joy in finding innovative or innovatively simple ways to measure complex concepts, such as gaps in governance systems, how governance is structured or what shapes how actors interact in governance.

My main methodological asset is a good knowledge about the tools, possibilities, and sometimes silliness of network approaches to data analysis. I believe in the integration of relational aspects of data structures and the merits of studying networked phenomena in their own right, but I also think that treating network analysis as its own field in applied research misses out on a lot of advanced knowledge about data analysis from other established data analysis procedures (eg. clustering).

I am fascinated by representing and analyzing complex social-environmental settings as multi-level networks. I have spent quite some time working with Exponential Random Graph Models and developed a kind of love-hate relationship with them over time. Recently, this relationship has improved a bit with new developments in Bayesian ERGMs. Beyond this, I like to combine generalized linear regression models with causal graph approaches to causality.

Related: Software development on multi-level network motif analysis in R

I have become ever more Bayesian, for the very simple reason that Bayesian Data Analysis fits the way I have always thought about how we gain knowledge in research. But it took me some time to realize that there was a whole area of statistics in line with it. I would also say that this way of thinking is also not particularly different from how most researchers I know treat the interpretation their models, even though the way these models are implemented often disagrees with them.

I do not care much about simplistic rejections of straw-man null hypotheses based on arbitrary threshold values nor the cult of statistical significance. I also do not care much about the fetish for super-large sample sizes - evidence is evidence and if you have less data points, you can still learn something, but you will probably just be more uncertain. With the caveat that this creates a responsibility to quantify that uncertainty. And I hate hiding behind reflexive statements of “correlation is not causation”” statements, even though we can often offer the people to which our research might matter more than that.

I am obviously influenced by a host of researchers and statisticians who have thought much more deeply than me about these things. This especially because as an applied researcher, I am dependent on these people to think things through properly and make them accessible. I would unequivocally recommend anyone to dive into Richard McElreath and his brillant “Statistical Rethinking”(v2) as well as Judea Pearl’s work most accessible in the “Book of Why”.

However, I feel that there is also a responsibility for an applied researcher to not only apply but also try to understand approaches they use, which has always been very important to me. This is not an easy line to walk obviously, but an important one to be aware of.