Thread by @lymanstoneky

Very cool JMP from @lydia_assouad : quantifying the effect of Ataturk randomly-happening-to-visit your town on the odds you adopt a Turkish (i.e. nationalist) name!

Leaders matter!
dropbox.com/s/93e05vg2euks…

It's a very nifty paper. Ataturk went on a political tour around Turkey promoting stronger Turkish/secular/Republican identity. Part of that was promoting the new "Pure Turkish" language. Baby names are a nice test case for this!

Also really good controlling for confounds. Paper has data on Ataturk's interaction with/co-optation of local elites, the formation of branches of his political party, etc. It can show mechanisms, complementary effects, etc. Leaders matter, but so do institutions!

The method of causal inference here seems decently good to me. Results are robust to "all visited cities" vs. "only cities along the route in-between major cities," i.e. cities which were not *main destinations* but *just happened to be along the way*.

Of course, geography is never random, so this isn't a true random treatment: what cities happen to be along routes between major cities is in fact highly endogenous to the traits of those cities and what institutions they might possess!

However, the balance tests mostly look good, with just a few problems related to minority prevalence, which the paper then addresses in several rounds of quite good looking robustness tests.

My *only* critique, which @lydia_assouad may have an answer to not in the paper, is about the administrative process of generating a birth certificate. The paper uses ~15 million birth certs across 30 years, so about 500,000 a year. That seems a little low?

Just doing some guesswork from censal populations and plausible crude birth rates, seems like Turkey should have had around 800,000 births per year.

So I wonder how complete the vital registry data is, and to what extent changes in *registration* behavior could matter?

Since the birth cert data catches a very large share of plausible births I don't expect this would eliminate the result, but one *can imagine* that Ataturk visiting could change *registration behavior* of more persuaded individuals, not just naming behavior.

But that doesn't necessarily undermine the paper. While the naming effect estimate would be inflated, a registration effect is arguably *more important*: if Ataturk got people to more directly engage with the state and participate in vital registries, that's a big deal!

So this is why I don't think my critique is a big problem: because even if it's true, it doesn't change the conclusion that leaders can be highly influential for socially important behaviors. Just changes *which behavior* from "language adoption" to "state interaction"

HT @jenniferdoleac for sharing