SNAP soda bans cut purchases 12%. The exclusion list decides what that is worth

Run the study's two headline numbers against each other and the baseline they imply is about 283 cans a year, a little over five a week ^[1][2][1]. The restriction takes roughly one can every eleven days out of that ^[2]. It is a real change in a purchasing habit, and it is small enough that a household can undo most of it by reaching for something the state left off its list.

Which is close to what the data show. The authors report that up to 39% of the money not spent on soda went to other sugary drinks and fruit juices that the state restrictions do not cover ^[6]. On a dollar basis that leaves about 61% of the nominal reduction standing, putting the effective cut in sugary-drink spending nearer 7% than 12% ^[3]. Notowidigdo's own reading is that scope is the binding variable: "If the goal is to reduce sugar consumption, you want the ban to be more comprehensive, not less" ^[7]. The leakage economists predicted did happen ^[16]. It went sideways into juice rather than back into soda bought with cash.

The national figures inherit the same shape. Paarlberg, who called the study sound, notes it is a 12% cut for the 12% of Americans who use SNAP ^[14]. If those households buy soda at roughly the national rate, the bans move about 1.4% of US soda purchases ^[4]. The billion dollars a year in health system savings Notowidigdo cites works out to about 0.02% of the $5.3 trillion the US spent on health care in 2024 ^[4][5][5]. He calls it a drop in the bucket himself, then points out that it recurs annually ^[4].

His alternative, per Paarlberg, is a tax: Philadelphia's cut consumption 31% across all groups rather than benefit recipients alone, and raised revenue that could offset the regressivity ^[15]. Worth registering that the study was funded by a Bloomberg Philanthropies grant, and Bloomberg Philanthropies has worked to tax sugary beverages in the US and elsewhere ^[17]. That does not make a 12% estimate wrong. It does mean the comparison the finding invites is the funder's preferred instrument.

The evidence is thin in the ways early working papers are thin: six months of 2026 purchase data covering 15,000 SNAP households, of which 3,291 were in restricted states, about 22% of the sample ^[10][6]. Chrisinger, at Tufts, cautions that it may not represent SNAP recipients generally ^[11]. And the survey finding sits outside the ledger entirely. His point is that stigma cannot be compared with health indicators, which means the trade-off this policy asks for is one nobody can price ^[13].