Are restaurants responsible for the obesity epidemic?

It has become almost the conventional wisdom that the obesity epidemic is at least partially attributable to people eating out. I for one really try and avoid eating out because I always feel like I end up eating junk food. But does this really matter? Do people actually eat more overall when they eat out more?

Economists Michael Anderson and David A. Matsa from UC Berkeley and Northwestern University respectively say no. In a working paper published last year, they argue that when people eat out they eat less at home -- resulting in only a tiny net gain in caloric intake.

To analyze the effects of restaurant access on obesity, the authors look at data from the Interstate Highway system. There are a lot of "highway towns" in America -- towns that grew up around a major highway and usually have large numbers of restaurants. The authors compared people living in these highway towns to their more rural neighbors who were located away from town. Though they found that their sample set contained two groups of people with substantially different access to restaurants, the obesity rates in the two samples were identical. Here is the key charts (Figures 1 and 2 from the paper):

Figure 1 shows the distributions of distances to the nearest restaurant for people living near or far away from the highway. You can see that people far away from the highway are not as close to a restaurant as those near a highway. (Presumably this has to do largely with the ability to get on the highway if you are not in a highway town and go to one.) Figure 2 shows the distribution of body mass indexes (a measure of obesity) for these groups. Note how the distributions are essentially identical.

(As an aside, I thought of one confound for this data, and I can't figure out whether they controlled for it. Maybe someone can read the paper more thoroughly and explain it to me. The confound is income. Maybe people who live closer to the highway have higher income. Given that people with low income have higher rates of obesity, the income effect may be masking the restaurant access effect.)

How can we explain this failure to see a difference in obesity with increased restaurant access? The authors then examine food intake data collected by the U.S. Department of Agriculture. The food intake data come from the Continuing Survey of Food Intake by Individuals, conducted from 1994 to 1996.

The authors wanted to establish how food intake in restaurants relates to food intake at home. They found that people substitute calories consumed at restaurants for calories consumed at home. This makes sense. If you go out to a restaurant, you aren't going to be hungry to eat at home:

If calories consumed throughout the day are substitutes, then our theoretical model suggests that people will compensate for larger portions at restaurants by consuming less throughout the rest of the day. Consistent with this prediction, the coefficient in the daily-level fixed effects regression is substantially less than the corresponding estimate at the meal-level. In fact, eating out increases daily intake by only 24 calories -- compared to an average caloric intake of 1,944 calories per day. This effect is statistically insignificant and represents a decline of almost 90 percent from the corresponding meal-level estimate. The result suggests that, although individuals tend to eat more at restaurants, they compensate to a substantial degree by eating less throughout the rest of the day. Meal-level estimates therefore overestimate the net effect of restaurants on total caloric intake.(Emphasis mine.)

How do we resolve this finding with the observation that restaurant food consumption correlates with obesity? The authors note that individuals who often eat at restaurants consume more both at home and at restaurants.

The between-individual coefficient is significantly larger than the fixed effects coefficient (235 versus 24), implying that individuals who frequent restaurants also eat more at home. This difference suggests that selection may explain why a number of observational studies have found a link between caloric intake and food away from home. Of course, even with individual fixed effects, the decision to eat at a restaurant is not exogenous. Given the size of restaurant portions, we suspect that consumers tend to eat at restaurants on days when they are hungrier. The 24 calorie per meal estimate therefore represents an upper bound and suggests that restaurant meals do not have a substantive causal effect on total caloric intake. (Emphasis mine.)

These findings if true have a great deal to say about obesity abatement policy. If it is indeed true that people will substitute calories consumed in restaurants for those at home and vice versa then sin taxes such as a restaurants tax will have little effect at reducing obesity rates. The whole issue has to do with the elasticity of demand for restaurant food. If demand for restaurant food then increases in prices will reduce quantity consumed but people will just get their food elsewhere -- likely in the same quantity.

The authors illustrate this problem with the calculation of the likely effects of a 50% restaurant tax. They point out to important effects. First, the tax will only moderately decrease the rates of obesity. Second, the cost of imposing such a tax, in terms of deadweight loss (the reduction in economic activity coming from the tax raising prices) and the loss of government revenue, would be significantly higher than the cost of paying for health care at the higher rates of obesity.

Next we compute a conservative estimate of the potential benefits of the restaurant tax to compare to the deadweight loss. The results from our natural experiment (presented in Section 5) suggest that taxing restaurant consumption would provide minimal benefit to public health. The point estimates, reported in Table 6, are close to zero and precisely estimated. To be conservative, assume that the effect of restaurant prices on body mass is one standard error greater than our point estimate (this corresponds to approximately the 85th percentile). In that scenario, a 50 percent tax would reduce the prevalence of overweight individuals by 1.3 percentage points -- compared to the 66 percent of Americans who were overweight in 2004. Using results from Finkelstein et al. (2003), a 1.3 percentage point decrease in the prevalence of overweight individuals reduces covered medical expenditures by $2.3 billion (reported in the second-to-last column of Table 8). The last column of Table 8 combines this estimate of the benefits with estimates of the deadweight loss from the previous paragraph to compute the ratio of the welfare costs to the potential benefits associated with a 50 percent restaurant tax. In all cases, the costs dominate the benefits, and the cost-benefit ratio ranges from 5.3-to-1 to 14.4-to-1.

While the deadweight loss associated with a tax policy is substantial, the deadweight loss associated with a zoning policy against restaurants, such has those proposed in New York City and Los Angeles, is likely even greater. With a tax policy, the government recaptures all of the out-of-pocket price increase from consumers. But with zoning regulations, only part of the effective price increase is recaptured by nearby firms while the rest is dissipated in increased time and fuel expenditures by consumers who travel further to access their nearest restaurant and wait in longer lines when they arrive.

I think the authors make a good point in emphasizing that regulators responsible for obesity abatement policy often have difficulty getting the public to comply with how their regulations are intended to work. Ban this food, and people will eat some other (and probably just as bad food). Limit access, and people will drive farther. Tax restaurants, and people will eat junk food at home. This is tough stuff.

I am not posting this to suggest that the government can have no effect on obesity through public health policy, but I do think that a bit of reticence and humility will be required in projecting the positive consequences of such policies.

The authors state their conclusions here:

At this point, many policymakers and public health advocates have shifted to designing policies intended to reduce the impact of restaurants on obesity, even while they acknowledge that convincing evidence of such a link has proven elusive. For example, the FDA recently organized a forum in which participants focused on proposing implementable solutions to the challenge of obesity in the context of away-from-home foods, even while the organizers cautioned that "there does not exist a conclusive body of evidence establishing a causal link between the availability or consumption of away-from-home foods and obesity" (Keystone 2006).

Our findings indicate that the causal link between the availability of restaurant foods and obesity is minimal at best. Manipulating the distance to the nearest restaurant using Interstate Highway proximity as an instrument demonstrates that restaurants have no significant effect on BMI or overweight status. These results are precisely estimated and robust to different specifications and samples. Translating the distance measure into an economic cost, point estimates imply that a 50 percent reduction in restaurant prices would have no positive effect on the prevalence of overweight individuals. Even the extreme tail of a 95 percent confidence interval implies that this policy would not reduce the overweight prevalence by more than 4.6 percentage points (while 66 percent of Americans were overweight in 2004). Similar conclusions hold when using BMI or an obese indicator as the outcome of interest.

...

These results, combined with work in the context of traffic safety (Peltzman 1975) and tobacco (Adda and Cornaglia 2006), suggest that regulating specific inputs into the health and safety production functions can be ineffective when optimizing consumers can compensate along other margins. Although restaurants conveniently deliver calories at a low marginal cost, they are only one source among many. The same principle may apply to other targeted obesity interventions as well. For example, two recent large-scale, multi-state randomized trials of school-based programs that improved the nutritional content of cafeteria meals found no effect on student weight (Nader et al. 1999; Caballero et al. 2003). One principal investigator notes, in retrospect, that the intervention could not control what the children ate outside of school (Kolata 2006). Future research and policy proposals may find greater success if it is designed to account for the optimizing behavior of the targeted subjects.(Emphasis mine.)

Hat-tip: Portfolio.com and Marginal Revolution