My college is floating the idea of investing in GPT-type technology to help researchers code text data. This reminds me of my longtime belief that the distinction between “qual” and “quant” is often less important than the distinction between different research paradigms.

I got my job largely because I can work with Twitter data, and my tenure application is built on the premise that I do good Twitter research. I probably shouldn’t take as much pleasure as I do from watching the platform fall apart right now, but I was ready to move on anyway.

One of those afternoons where I’m auditing someone’s analysis code, but it’s an analysis of 4M rows of data, so I’m also doing spurts of grading while I wait for code to execute.

Today’s manuscript revision fun is detangling the results of a coding error that left out 3 hours and 56 minutes worth of tweets from my analysis. Just enough to make some very small differences in reported results.

I usually do too much qualitative work for them to be useful, but .Rmd-based slides are a delight to work in.