aliquote

< a quantity that can be divided into another a whole number of time />

On consulting

June 30, 2022

I have been working as a full-time statistical consultant (mostly in biomedical research) for something like 13 years. Sometimes I miss the job, but in the end I’m better served where I am now. The academic world is worth what it is worth, but when you are looking for a quiet place it is one of the best options. Consulting is both challenging and entertaining: you learn a lot, often in a short period of time, which means you get continuing education (almost) for free, and your network will rapidly grow provided your work is appreciated, which means you get a lot more opportunities, hence you may become more demanding when selecting new projects. I already mentioned this in a previous post.

After 13 years I had a serious accident that forced me to review my perspectives both professionally and from a more personal point of view. Concerning the first point, it turned out that I was getting tired of repeating the same thing over and over again to the doctors I was interacting with, not to mention the fact that from then on I was going to find myself in the position of a patient for a good part of my days.1 But the real problem, it seems to me, is that at a certain point it becomes difficult to take the place of the principal investigator and to suggest alternative interpretations of the results. My job was to organize and analyze the data with sound statistical methods. It’s not that I didn’t care that much about what they might do with the results afterwards, but honestly that was not part of my job. I’m willing to learn a lot about domains I was not trained in, but it would be hard to become a specialist in each new domain in a few months, right? And finally, I was most often involved at later stages of clinical studies. You know the famous adage:

To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of. — Sir R. A. Fisher

They rarely consulted me before carrying out their studies, so I was left with the eventually flawed design, and already collected data to analyze. No prior power analysis, no interim analysis, and so on. I always did the best I could do given the context, and I believe no one was ever unhappy with my job. However, I felt something was broken in this workflow. You get paid for analyzing data, without being involved in the design of the study, results come afterwards and people ask to carry out subgroup analysis or devise new hypotheses to explain or at least exploit unattended results. First it sounds at an angle with scientific reasoning, second I’m surely not the right person to ask. I ended up believing that some people would benefit from taking a course in experimental design and causal inference.

A few days ago, I came across an interview with John D. Cook where he described his move to mathematical consulting after years in Academia and work in private sector. He has better words than me regarding part of the specific issue I mentioned above:

Applied statistics may be closer to epistemology than mathematics. You have to question what you know, why you think you know it, and how confident you are (or should be) in that knowledge. You must make simplifying assumptions and determine whether they are justified. Statistical modeling is hard to do well. It’s easy to write down a model if you’re not overly concerned with the accuracy of your results. As the old saying goes, fools rush in where angels fear to tread.
Sometimes I get to work on statistical problems with crisp mathematical statements, such as finding an efficient way to compute something. These projects are fun because the epistemological concerns are someone else’s responsibility.
My work in applied mathematics has been more objective than my work in statistics. When you’re dealing with voltages or velocities, what you’re measuring (and what it means) is pretty clear. There’s much less anxiety around modeling, and your level of success at the end is usually obvious. I prefer mathematical work, but there are more statistical opportunities in consulting. These days I often hand over statistical projects to someone else; I work on statistical problems about data privacy but am moving away from medical statistics.

♪ Gerald Clayton • Like Water


  1. I spent a total of 10 weeks in intensive care type hospitalization over a period of 5 months, 7 of which were consecutive. ↩︎

See Also

» Computing the inverse CDF of a Gaussian » The unquantified self #23 » Random sampling by the inverse CDF transformation » The unquantified self #22 » The unquantified self #21