Can someone help me find this recent paper that talks about interpreting the coefficients on dummy posttreatment variables in a diff in diff when the dependent variable is logged? I remember the looked at AER papers and most of them incorrectly interpreted the coffee as a pct change in the dv . Can't find it again, need it.
Recent paper about logged DVs in a diff in diff

suppose that ln (y) = B*x where x is a dummy
remember the change in a the dummy dx = (1  0) = 1
so interpretation is d ln y = B*dx is B = dln (y)
now the instantaneous change (local change) in a log can be interpreted as
d ln y = d/dy (ln y) * dy = 1/y * dy = dy/y or a local percent change.

suppose that ln (y) = B*x where x is a dummy
remember the change in a the dummy dx = (1  0) = 1
so interpretation is d ln y = B*dx is B = dln (y)
now the instantaneous change (local change) in a log can be interpreted as
d ln y = d/dy (ln y) * dy = 1/y * dy = dy/y or a local percent change.This is wrong, as the paper shows

This is not it. That just shows the proper exponentiated transformation of a coeff in a semilog model
The paper is here: https://ashecon.confex.com/ashecon/2018/webprogram/Paper6783.html

https://www.jstor.org/stable/1805237
This is not it. That just shows the proper exponentiated transformation of a coeff in a semilog model
The paper is here: https://ashecon.confex.com/ashecon/2018/webprogram/Paper6783.htmlSorrythis is the OP. I found it after some more searches.

suppose that ln (y) = B*x where x is a dummy
remember the change in a the dummy dx = (1  0) = 1
so interpretation is d ln y = B*dx is B = dln (y)
now the instantaneous change (local change) in a log can be interpreted as
d ln y = d/dy (ln y) * dy = 1/y * dy = dy/y or a local percent change.This is wrong, as the paper shows
WTF??? You need a paper to show you that this is wrong? What on earth did you learn in calculus and undergrad econometrics?

suppose that ln (y) = B*x where x is a dummy
remember the change in a the dummy dx = (1  0) = 1
so interpretation is d ln y = B*dx is B = dln (y)
now the instantaneous change (local change) in a log can be interpreted as
d ln y = d/dy (ln y) * dy = 1/y * dy = dy/y or a local percent change.This is wrong, as the paper shows
WTF??? You need a paper to show you that this is wrong? What on earth did you learn in calculus and undergrad econometrics?
You may be misunderstanding. It's not about the approximation .. it is anerror made in many aer papers and actually a ton of diff in diff papers. The really key thing is you can't even sign the error that they make!

I dont understand it so much noise on it . If the coef is small, say less than 0.20 or 0.10, this adjustment does not matter much at al. If the cof is very large, yes the log approximation would fail.
Nobody is talking about the log approximation except the dummies that aren't reading ..and also think they know everything even though they spend all day making s**tty comments on ejmr instead of publishing

I dont understand it so much noise on it . If the coef is small, say less than 0.20 or 0.10, this adjustment does not matter much at al. If the cof is very large, yes the log approximation would fail.
Nobody is talking about the log approximation except the dummies that aren't reading ..and also think they know everything even though they spend all day making s**tty comments on ejmr instead of publishing
i still don;t get it. what is new after that AER 1980 paper? do you have the new working paper?

Not really a new point, unless there's more to it than the abstract shows. The fundamental issue is that percent changes are not even a useful concept when you're talking about substantial differences. Increase a variable by 100%, then decrease it by 50%, and you're back where you started.