Robert H. & Nancy Dedman Trustee Prof of Econ at @SMU, husband, son, proud papa, baseball junkie, Carroll Dragon. Mastodon: @dlmillimet@econtwitter.net
May 25, 2023 • 6 tweets • 2 min read
This is all excellent advice. I will reemphasize FINISH PAPERS. Being smart is necessary but not sufficient for tenure. Unfortunately you need to prove it with pubs and you cannot pub w/o submitting. Many smart people have not succeeded in academia (if that’s your goal) bc
they do not submit … out of a need for “perfection” (that is not attainable) or fear of rejection (which will happen A LOT). My addl piece of advice is that while you should DEFINITELY do things that interest you and not worry about the “profession”, you also need to be a little
Mar 8, 2023 • 4 tweets • 1 min read
For those who don't want to spend the time teaching themselves, there are 6 critical things to know about measurement error.
1. Classical ME in Y leads only to imprecision. Classical ME in X leads to attenuation bias. 2. The vast vast majority of ME is not classical.
3. ME in a binary (or any bounded var; e.g.,non-negative count var) cannot be classical. 4. In multiple regression, "small" ME can lead to huge bias since it depends on the share of the var in the mismeasured var that is due to ME after partialling out all the other covariates.
Mar 23, 2022 • 6 tweets • 2 min read
Covering a common theme I like to harp on in econometrics in class today. Related to IV. Say the model is
y = a + b1*x1 + b2*x2 + e,
where x1 and x2 are endogenous, b1 is the "parameter of interest", and z is a valid IV for x1 in the usual sense. You do not have an IV for x2.
Do we care that we don't have an IV for x2? Some might argue that since we only care about b1, we don't care about x2. However, for IV to give a consistent estimate of b1 we require more of z: z must (generally) be uncorrelated with not only e, but also x2.
Jul 13, 2021 • 7 tweets • 2 min read
The delta method is awesome. It is also used in most @stata nonlinear test commands. However, most people - I suspect - don’t understand how it works. Let me tell you…
When we define a new random variable that is a linear function of other RVs with known variances and covariances, it is trivial to find the variance of our new RV. However, when the new RV is a nonlinear function of other RVs, even with known vars and covs, we are out of luck.
Apr 30, 2019 • 12 tweets • 4 min read
So, I've decided to call my psuedo-blog on #EconTwitter about random econometric factoids "How the Sausage is Made". It will be me sharing (hopefully correct) tidbits that I think most applied types are unaware and may be of use. A bit of insight into "how the sausage"
is made without getting bogging down in all the metricy (h/t to @agoodmanbacon for using that term the other day) details. Today I'm getting into the weeds a bit about how "finicky" (as I describe it) the IV estimator is. Before describing some of the downside to IV, I think it
