Somewhere in an American living room, between 2008 and 2010, an interviewer asked a woman to take off her heavy outer clothing and empty her pockets. She stood against a wall with her heels and shoulders touching it. The interviewer marked her height with a rafter angle square and measured it to the nearest quarter inch, weighed her on a digital floor scale, then pricked her finger and collected a few drops of blood onto specimen paper.
The interviewer did not know which group the woman had been assigned to fourteen years earlier. That blindness is the point. These measurements are the closest thing anyone has to an experimental answer to a question that gets asserted constantly and tested almost never: does the place you live change your body?
Why the usual evidence cannot settle it
Hundreds of studies report that people in poorer, less walkable, more fast food saturated neighborhoods carry more weight. Every one of them hits the same wall. Nobody is assigned a neighborhood. People sort themselves. Someone who enjoys walking finds a walkable street. Someone with more money finds better groceries. The place and the person arrive together, and no amount of statistical adjustment fully pulls them apart.
The Artful Age
A weekly letter on aging well, family across generations, and the creative life after the kids leave home.
Which is why a housing program built to study employment ended up producing the best available evidence on obesity.
The lottery
From 1994 to 1998 the US Department of Housing and Urban Development ran Moving to Opportunity in Baltimore, Boston, Chicago, Los Angeles and New York. Families with children in selected public housing, in census tracts where at least 40 percent of residents lived below the poverty line, were invited to enter a lottery for a rent voucher. About a quarter of eligible families applied. Of 5,301 volunteers, 4,498 families were randomized.
One group was offered a voucher redeemable only in a tract where fewer than 10 percent of residents were poor, plus short-term counseling on the move. A second group was offered a standard voucher with no restriction on destination and no counseling. A third was offered nothing new.
That second group is what makes the study unusual. It separates the act of moving from the act of moving somewhere less poor.
The measurements described above came an average of 12.6 years after randomization. Height and weight were taken rather than asked. Blood spots went to a certified laboratory for glycated hemoglobin, which reflects average blood sugar over the preceding months. About 84 percent of participants provided height and weight; about 71 percent provided blood.
The split result
Among women in the control group, 58.6 percent had a body mass index of 30 or more, 35.5 percent were at 35 or more, 17.7 percent at 40 or more, and 20.0 percent had a glycated hemoglobin level of 6.5 percent or higher, the American Diabetes Association’s threshold for diabetes.
The group offered the low-poverty voucher sat 4.61 percentage points lower at a BMI of 35 or more, 3.38 points lower at 40 or more, and 4.31 points lower above the diabetes threshold. In relative terms, reductions of 13.0, 19.1 and 21.6 percent.
At a BMI of 30 or more, the standard definition of obesity, there was nothing to report. The estimate was 1.19 percentage points, with a confidence interval running from minus 5.41 to plus 3.02.
The group handed an unrestricted voucher showed no significant difference from the control group on any of the four measures. That is suggestive rather than settled. The authors could not rule out that the unrestricted voucher did as much as the restricted one, and on severe obesity the unrestricted group’s own estimate fell just short of significance. But the gap between the two voucher groups on diabetes came close to significance in its own right, pointing the same way. Moving may not have been the active ingredient. Moving somewhere less poor may have been.
The caveats the authors put in writing
Jens Ludwig and his co-authors were more careful with this result than most of the coverage that followed it.
Fewer than half the families offered the low-poverty voucher, 48 percent, actually used it. The control group did not stand still either. Their average tract poverty rate fell from 53.1 percent at the start to 33.0 percent a decade later, as families moved under their own steam. The gap in neighborhood poverty between the groups was 17.1 percentage points after one year and 4.9 points after ten. The experiment was fading while it ran.
The authors tested several BMI cutoffs without adjusting for multiple comparisons, and said so, noting that their estimates on extreme obesity “may be marginally significant.” That sentence is theirs, in their discussion section. Their diabetes threshold also misses anyone whose diabetes was being successfully treated. And the baseline surveys contained almost no health information, so the study cannot say whether it prevented conditions from starting or helped existing ones ease.
They were equally plain about who was in the room. Participants volunteered. More than 90 percent of the households were headed by a black or Hispanic woman with children, and the sample was heavier than national samples. Care should be taken, they wrote, in applying the findings to populations with different characteristics.
They could not say what caused the effect either. Access to routine medical care was no different between the groups at any follow-up. What did differ was tract poverty, the share of neighbors with college degrees, and how safe and cohesive the women said their streets felt. The mechanism, they concluded, “remains unclear.”
A candidate mechanism, from a very different study
If the experiment cannot name the ingredient, a separate line of work offers one, and it is the opposite of the feature that has dominated policy.
Kristen Cooksey-Stowers, Marlene Schwartz and Kelly Brownell examined all 3,141 US counties using the 2009 USDA Food Environment Atlas. They set two ideas against each other. A food desert is an absence: the share of a county’s population that is both low income and more than a mile from a supermarket, or ten miles in rural areas. A food swamp is an excess: the ratio of fast food outlets and convenience stores to grocery stores.
The two measures turned out to be weakly and negatively correlated, meaning they describe different places. In the correlations, the food desert measure was not significantly associated with obesity. In the ordinary regressions, once food swamps were accounted for, no food desert measure survived. When the authors used an instrumental-variable approach to deal with self-selection, food deserts did come back as significant, but with an effect much smaller than food swamps.
The size of the food swamp effect depended entirely on the method. In the plain regressions it was tiny: a 1 percent increase in fast food restaurants went with roughly a 0.125 percent increase in obesity. Using highway exits per county as an instrument, the coefficients grew several times over. The authors’ own translation is that the plain estimates would imply reducing obesity by around a tenth of a percent, while the instrumented estimates imply something nearer 3 percent.
More interesting than the size is where the effect showed up. It held in counties where people were less likely to drive or use public transit to get to work, and was not significant in counties with above-average driving or transit use. The authors read this as a finding about who is stuck with their surroundings. If you can drive past the convenience store, its presence matters less.
Their limits are real and they list them. The data are a single cross-section, the obesity figures are self-reported, and the highway-exit instrument is imperfect by their own admission, with 1,672 counties having no highway exits at all. Gas stations sell convenience food and cluster near highways but were not counted. They state directly that the design “limited our ability to make causal statements.”
Adding versus subtracting
These two studies were done by different teams, in different decades, with designs that share nothing. Cooksey-Stowers and colleagues do not cite Ludwig and colleagues anywhere in their paper. They nonetheless point the same way, and they point away from the intuition that has driven most of the spending.
That intuition is additive: open a supermarket where there is none. The published abstract of a 2014 pilot study by Steven Cummins, Ellen Flint and Stephen Matthews, which followed the opening of a new supermarket in a Philadelphia food desert, reports that residents’ perceptions of food access improved while their reported fruit and vegetable intake and body mass index did not. That is the paper’s own summary.
What to take from it
None of this is advice you can act on before dinner. Nobody rezones their own street.
What it is good for is calibration, in both directions. The claim that your address determines your weight is too strong. The randomized test found nothing at all at ordinary obesity, and its authors described their severe-obesity finding as possibly marginal. The opposite claim, that body weight is purely a matter of individual willpower operating independently of surroundings, is also too strong, and the housing lottery is why. The women who received a low-poverty voucher were not given a nutrition course, a gym membership or a lecture. Something about where they ended up living did a small amount of work at the far end of the distribution over twelve years.
If there is a usable version, it is narrow, and it is about proximity rather than absence. What sits within easy reach of your door appears to matter more than what is missing from your area, and it matters most to people who cannot easily get past it. Which is a finding about streets and transport. It is not a verdict on anyone’s character, and it is not a reason to think less of your own.