Matt Blackwell
data, causal inference, experiments, politics
https://mattblackwell.org
- Narrative violation: Harvard is viewed more favorably than Trump, Vance, or Musk.
- [Not loaded yet]
- [Not loaded yet]
- Lol got me!
- Unfortunately, partial identification is usually only achieved in terms of population parameters. Still have to deal with sampling uncertainty!
- Seeing AI people refer to 100% confidence intervals, we’re ngmi
- Does anybody have a good visualization to explain how interactions can be confounded, and why interactions require interaction controls? @urisohn.bsky.social maybe? (Asking because I have an idea, but want to check out what exists already before investing the effort)
- [Not loaded yet]
- Yes, we call this omitted interaction bias. Hadn’t seen the ReStat article but glad to see people raising the issue in Econ as well!
- Gen alpha lingo that is starting to grate my ears: “I ‘searched up’ Minecraft mods”
- The problem w this approach is that to get to the hard stuff, one has to spend a lot of time doing the stuff that a computer is very good at. AI in education allows you complete work without mastery, leaving you without the foundations for the hard stuff
- The other aspect of this is that LLMs turn us from creators to evaluators and editors. But to be successful editors and evaluators, we need to know enough to (a) judge the output accurately and (b) fix the output as needed.
- After listening to the Marxist, feminist professor rail against God, a Marine in the back of the class stood up and said “grok, is this true?”
