Fewer H-1B visas did not mean more employment for natives (2017)

Dumb example: foreign company wants to start a branch office in the US, by sending in a bunch of people from its head office to spin it up. The branch office will be in a new building, and the branch office needs a janitor. Visa shenanigans mean the foreign company decides against doing this. Branch office is shut down, one less janitorial job.

Silly but I think it's very easy to concoct these kinds of qualitative stories to infer the theoretical possibility of some quantitative result. Probably why so many of these discussions go in circles!

The easy counter to the above is to say that the foreign company wanted to start a branch office for reasons, and that those reasons remain true with our without H-1B. My impression is that lots of company decision making is of two varieties:

- If you are ginormous: sometimes you are big enough to where you really can't find 500 IT people for your office in some mid-sized market no matter how much money you throw at the problem, so you try to hire from other places. If you aren't able to find 500 people "quickly" the idea stops being interesting

- If you are smaller than ginormous: almost every decision about hiring is actually extremely personal. New offices get opened because you have a handful of people linked to an interesting opportunity, and its through those people that it will happen or not. Denying access to even a couple of them just denies the whole opportunity.

This is just my own view of the world though

The only downside is for the countries these people come from - these countries don't have an immigration problem. Their issue is a much bigger brain drain problem.

Control group mismatch: Treats non-profits as a clean counterfactual for for-profits, but the sectors have very different occupational mixes and shocks—so the triple-diff can pick up sectoral composition, not policy.

Short pre-trend: Only a tiny non-binding window to test parallel trends → weak pre-trend evidence.

Approvals ≠ demand: Uses approved H-1Bs (not applications), so results reflect rationed outcomes and USCIS processing quirks, not employer demand.

Strong wage-equalization assumption: Identification leans on wages equalizing across very different employers; if sector premia move (e.g., recession), estimates drift.

Relative, not absolute, effect: The estimate is “for-profit vs. non-profit”; if non-profits expand (cap-exempt), the “effect” can be reallocation, not a true for-profit decline.

Recession confound: Lottery years coincide with 2008–09; macro shocks can differentially hit new vs. established hires across sectors.

Noisy worker/firm measures: Experience is imputed (age minus stylized schooling ages); employer names are inconsistently harmonized → concentration trends may be artifacts; “large firm” cutoff is arbitrary.

Wage results are shaky: Based on offer wages (not realized), trimmed, and imprecise—tail stories are fragile.

Placebo underpowered/misaligned: Native “no effect” test is noisy and not analogous to “new vs. established” H-1Bs, so it’s weak evidence on substitutability.

Ignores geography: Cells are national; regional wage floors and local cycles could drive composition shifts.

Net: clever design but brittle; treat findings as suggestive reallocation under approvals data, not clean causal effects on demand, wages, or “top talent.”