Lots of people have asked me if studying biostats has actually been relevant in my career as a software engineer, and I’ve found the answer to be a resounding yes! It's super relevant in lots of engineering problems and in understanding the world generally. 🧵 follows!
When I worked on payment fraud prevention, I was always talking about diagnostic testing for rare diseases!
Diagnostic testing was something we studied at length in our early biostat & epi classes in grad school and it turns out “fraud” behaves similarly to a “rare disease” in a lot of ways.
Evaluating a predictive fraud model is a lot like evaluating a diagnostic test! In particular, ✨ PPV is a function of population prevalence ✨ or, as I said in many a meeting, ✨ model precision is a function of our overall fraud rate ✨
Model evaluation and comparison was much, much harder than I’d expected it to be when I first started in industry. There’s no one right answer! But having a good foundation for understanding the tradeoffs ended up being super useful.
Topics like positive/negative predictive value, uncertainty/risk assessment, and treatment effectiveness have also come very obviously front and center with covid, and I’ve seen them sneaking around in the background in politics too.
for example: to combat an outcome of voter fraud (the very definition of a rare event), politicians have proposed requiring ID to vote (a test). But the treatment adversely affects double-digit percentages of non-fraudulent votes, and has a positive predictive value of ~0.
This framing makes it pretty easy to see that voter ID requirements aren't really being made in good faith.
The thread has admittedly moved outside of tech now 😅 so circling back — a biostat background has served me super well! I use biostat skills basically daily at work in applied fields totally unrelated to biology, and beyond.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
