Erase food insecurity in a model and poorer older Americans gain the most. The main reason looks like head count: 40.8% of them went short, against 2.4% of the best-off

In 2013 a mailed questionnaire asked American adults, most of them over 50, some blunt things. Did the food run out before there was money to buy more? Did you cut or skip meals because there wasn't enough money? Were you ever hungry and didn't eat? A new analysis in Frontiers in Nutrition, published on 3 November, took 6,695 people who answered and followed their reports of depressive symptoms through 2020. It's hardly news that money, hunger and low mood travel together. The authors wanted to pull apart something harder. If poorer people stand to gain more from food security, is that because more of them lack it, or because going without hurts them more?

The head count comes first. The team scored each person on wealth, income and education, then split them into three groups of similar size. Food insecurity, meaning two or more points on the six-point US government short form, touched 40.8% of the low group, 14.2% of the middle group and 2.4% of the high group. Very low food security, the more severe band, was 18.2% against 1.2%. By our calculation, 69% of all the food-insecure people in the study (869 of 1,262) sat in the low group.

Those people also reported more symptoms. After adjusting for earlier symptoms, age, sex, race and ethnicity, smoking, alcohol, BMI, health and insurance, food insecurity went with a score 0.286 standard deviations worse in 2014. Using the spread in the paper's baseline table, we make that about half a point on an eight-item yes-or-no symptom checklist. That checklist isn't a diagnosis. The gap then shrank by 0.065 per two-year wave, which by our arithmetic leaves about 0.09 by 2020. It would be pleasant to read this as recovery, but the authors warn against it: food security was measured only once. Someone who went hungry in 2013 might be fine by 2018, so part of the fading may just be a label going out of date. The study also can't separate brief hardship from years of it.

The headline number comes from a simulation. Nobody was given food. The researchers used a statistical model to predict everyone's scores with food insecurity switched off, then compared that with what was observed. Averaged over the follow-up, the gain was 0.097 standard deviations in the low group (95% CI 0.068 to 0.126), 0.023 in the middle group and 0.003 in the high group, whose interval includes zero. On the checklist, 0.097 is roughly a fifth of a point by our conversion, spread across a whole group that includes the many people who already had enough to eat. The authors say plainly that this isn't the expected effect of any particular programme, whether cash, vouchers or food boxes, and that those options are not interchangeable.

So is it exposure or vulnerability? The authors went in expecting vulnerability: people with less of a cushion would be hit harder. The evidence for that was thin. A second measure, which compares everyone food-secure with everyone food-insecure, favoured a larger gap in the low group over the high group in 2014 (a difference of 0.288, 95% CI 0.057 to 0.519), and that difference wasn't significant at later waves, though a joint test across all four waves did signal variation (P = 0.024). A separate test of how the gap changed over time singled out middle against high (P = 0.016), but low against high didn't reach significance (P = 0.185), so there was no neat ladder from poor to rich. Their conclusion is that the bigger benefit mainly reflects the bigger burden. Our own crude check supports 'mainly' without making it 'entirely'. Prevalence was 2.9 times higher in the low group than the middle one, while the modelled gain was 4.2 times higher. That is a back-of-envelope ratio and not a finding of the paper, and the high group contributes only 51 food-insecure people, so it can't carry much weight.

Direction is the other open question. Depression can make it harder to work, shop, cook and manage money, so it can produce food problems as well as follow them. Adjusting for 2012 symptoms helps but doesn't settle that, and the authors say so. The people dropped for missing data were older, poorer and in worse health, and the results carry no national weighting, so they describe this sample more than older Americans at large. With those limits in mind, the practical point is a modest one. If the gain in a group comes mainly from how many people are affected, then how many people a food programme reaches matters, whatever anyone's personal resilience turns out to be.