Imagine you're running an A/B test on a website, but instead of randomly assigning users to button colors, the U.S. government randomly assigned young men to military service in the 1960s and 70s via a birthday lottery. That's the instrument this paper uses. The Vietnam-era draft lottery is one of the cleanest natural experiments in social science — random assignment to a life-altering treatment, with decades of follow-up data. Hogan-Hennessy and colleagues exploit it to ask whether military service causes food insecurity later in life. The committed claim: military service does not adversely affect food insecurity. In fact, veteran-headed households show lower food insecurity rates than non-veteran households, and they participate less in federal nutrition assistance programs like SNAP. The authors use the Panel Study of Income Dynamics (PSID), which gives them longitudinal household-level data stretching across decades — a meaningful upgrade over the cross-sectional CPS data that prior studies relied on. The identification strategy is textbook instrumental variables. Draft lottery numbers are random conditional on birth year, so they satisfy the exclusion restriction: the lottery number affects food security only through its effect on military service. The first stage is strong — lottery-eligible men were substantially more likely to serve. The IV estimates confirm the OLS direction: no evidence that service increases food insecurity. If anything, the causal estimates trend protective, though confidence intervals are wide enough that you shouldn't over-interpret the point estimates. Here's where the paper gets honest in a way that matters. The PSID tracks household heads, which means the sample skews toward veterans who formed and maintained traditional households. Veterans who are homeless, incarcerated, living in group quarters, or otherwise not heading a surveyed household are invisible in this data. The authors flag this directly: the veterans most likely to experience food insecurity are precisely the ones least likely to appear in the PSID. This is a selection-on-outcomes problem baked into the data source, not a flaw the authors introduced. The policy implication lands with specificity. Currently enlisted service members show elevated food insecurity — the paper cites existing evidence on active-duty food needs. And veterans who don't head traditional households (the homeless, the institutionalized, the transient) are where the real vulnerability concentrates. The authors argue that targeting these populations, rather than veterans broadly, would be more effective policy. This is a narrowing move: the paper doesn't find a crisis among veterans generally, but it identifies where the crisis hides. The main limitation is statistical power. IV estimates are inherently noisier than OLS, and the PSID sample of Vietnam-era veterans, while longitudinal, isn't enormous. The paper can reject large adverse effects but cannot precisely estimate small ones. The confidence intervals on the IV estimates include both modest protective effects and modest harmful effects. The authors are transparent about this, but a reader should understand that 'no evidence of adverse effect' is not the same as 'evidence of no effect' — classic null-result interpretation territory. Methodologically, this is clean work in a well-established tradition. The Vietnam draft lottery IV has been used extensively (Angrist 1990 is the grandfather paper), and the contribution here is applying it to a specific outcome — food insecurity — with better longitudinal data than previously available. It's not opening new methodological territory; it's deploying a trusted tool on an under-examined question. The 33-page paper with 20 pages of appendix suggests the robustness checks are thorough, which is what you want from an IV paper where the identification strategy does most of the heavy lifting.
