If you’re gonna use this framing you also need to publish “Trumps fails to deliver on his promise to lock up Hillary Clinton”

Feel free to get in touch with any questions (and definitely if you notice any errors!)

Here's an overview of the paper's findings.

We're planning to release the Key Officers dataset soon, so keep an eye out for that, and feel free to get in touch if you have any questions about it.

Another finding: there seems to be an implicit "quota" system in effect, preventing multiple women from serving in leadership roles in the same embassy.

(As one female FSO describes: "I have never heard anyone complain about too many men in the embassy.")

We also find, at all levels, women are assigned to smaller and less important posts.

We find no relationship between host-country gender equality conditions and the gender distribution of US diplomats assigned there.

Women face a large disadvantage in promotion to the low-visibility DCM position, but not the higher-visibility CG. Then, among DCMs, women are more likely to reach ambassador (high viz), but no more likely to reach DAS (low viz).

But our data allow us to look more closely at individual appointment decisions, focusing on 1997-2017.

We compare appointments to two pairs of positions that are equal in status, but differ in visibility: Deputy Chief of Mission (DCM) vs. CG, and Ambassador vs. Deputy Assistant Secretary (DAS)

Top-line finding: women face substantial penalties in promotion into the middle and upper ranks, but only in low-visibility positions.

Looking only at representation in the aggregate, things seem to be trending in a positive direction:

Just curious, could you say a bit more? I can see simulations revealing e.g. you’re underpowered or you’re using the wrong SEs, but not obvious to me what sort of “mistake” you would make in the analysis that you wouldn’t also make when simulating the analysis

If the causal chain is X -> Y -> Z, then a regression of Z on X while conditioning on Y is problematic because of collider bias / conditioning on a post-treatment variable. Lots of relevant cites, eg onlinelibrary.wiley.com/doi/abs/10.1...

But have you considered that this creates disadvantages for certain groups of scholars (people who like to p-hack)