Sunday, April 24, 2022

Food Desert Mirage (redux)

A recent story in Eater resurfaces a tired argument about the observed correlation between certain retail stores and food deserts and nutritional equity. In this particular story, it is asserted that there is a direct correlation between their presence and food deserts. The policy proposal is to limit and regulate the expansion of dollar stores and similar retailers in these areas. This policy is biased from a couple different perspectives, one is that good intentions don't constitute an identification strategy.  It is hard work to move from correlation to causation, and as noted below, the nuance involved in generating meaningful insights that lead to meaningful policies don't generate the kinds of sensational headlines that media and special interests often seek. That leads to the second bias, as Daniel Kahneman states in Thinking Fast and Slow, System 1 likes to find quick answers to difficult questions by substituting easier questions and creating coherence where there is none. Policy advocates here have substituted an easier correlational question for a much more difficult causal question that builds a coherent story about food deserts, inequity, and Dollar stores. 

The more difficult question is, if you build a new supermarket in a food desert, will low income households go there to buy healthier food? Are Dollar Stores cornering the market in poor neighborhoods reducing options for healthy food choices? Is there a causal relationship between Dollar Stores and food inequity?

There is a misconception, a mirage if you will, related to the relationship between proximity of super markets that sell healthy foods and actual consumption and health effects. As discussed in this New Food Economy article 'Is it time to retire the term food desert':

"The idea that supermarkets enter into food deserts and all of a sudden provide access to healthy food is a little bit of a misconception"

Public Health literature provides evidence that households in lower income neighborhoods tend to eat less healthy food. These neighborhoods are often characterized as being food deserts due to the lack of access to healthy groceries for a given geography. Policy and discussion involving food deserts is often colored by an implicit or assumed causal relationship between food deserts (lack of supply of healthy food options) and nutrition and health outcomes. Failure to better understand this causal relationship can lead to potentially bad policy decisions. According to this City Journal article 'Unjust Deserts'  some communities have essentially banned or greatly restricted Dollar General from operating their stores which provide a variety of low priced products. However, some research questions a relationship between food choices and the presence or absence of a Dollar General store.

In a Health Economics Review article (Drichoutis, 2015), using a combination of difference-in-difference and propensity score matched analysis authors looked at the relationship between BMI in children and the proximity of Dollar General Stores and failed to find a relationship.

The authors conclude:

"Combatting the ill effects of a bad diet involves educating people to change their eating habits. That’s a more complicated project than banning dollar stores. Subsidizing the purchase of fresh fruits and vegetables through the federal food-stamp program and working harder to encourage kids to eat better—as Michelle Obama tried to do with her Let’s Move! campaign—are among the economists’ suggestions for improving the nation’s diet. That’s not the kind of thing that generates sensational headlines. But it makes a lot more sense than banning dollar stores."

A paper from the National Bureau of Economic Research this past year took a very exhaustive look at the relationship between food deserts, poverty, and nutrition. "THE GEOGRAPHY OF POVERTY AND NUTRITION: FOOD DESERTS AND FOOD CHOICES ACROSS THE UNITED STATES." Working Paper 24094 (http://www.nber.org/papers/w24094).

This paper helps provide a very rigorous empirical understanding of these relationships that can be leveraged for more effective policy and interventions to improve nutrition and health.

They used a very rich dataset consisting of:

1) Nielsen Homescan data - 60,000-household panel survey of grocery store purchases

2) Nielsen’s Retail Measurement Services (RMS) data - 35,000-store panel of UPC-level sales data (this covers 40% of all U.S. grocery store purchases)

3) Nielsen panelist survey data on nutrition knowledge

4) Entry and location data for 1,914 new supermarkets by zip code

Among the many findings uncovered in this data source was the following:

"over the full 2004-2015 sample, households with income above $70,000 purchase approximately one additional gram of fiber and 3.5 fewer grams of sugar per 1000 calories relative to households with income below $25,000."

Their data reflects what has been found in the public health literature in relation to low income households and nutritional health. In addition, household food purchase data was transformed using a modified version of the USDA's Healthy Eating Index (HEI) based on dietary recommendations. These various sources were brought together to give a very rich picture of household choice sets, retail environment, consumption patterns, and nutritional quality.

Using a regression based event study analysis and a structural demand model they examine the impact of supermarket entry on the nutritional quality of changes in food purchases. They also are able to separate the main drivers explaining the differences in the measured nutritional quality index (HEI) of food purchases between low and high income groups.

They model household and income group preferences using both constant elasticity of subsitution (CES) and Cobb-Douglass utility specifications. They apply this model to the rich data sources mentioned above using a Generalized Method of Moments (GMM) framework and use the model estimates to simulate policies that allow households of different incomes to be exposed to similar prices and product availability. (i.e. to make apples to apples comparisons and determine what's driving healthy vs. unhealthy food choices among low income households in food deserts vs. wealthier households).

Key Findings:

1) When new supermarkets open in what was formally a food desert, they find most of the changes in consumption are related to shifting purchases from more distant super markets to the new local super market. The change in the healthy eating index or substitutions away from unhealthy purchases from convenience and drug stores to more healthy food was minimal. This is because even in food deserts among low income households, willingness to travel was quite substantial and mitigated the lack of access to local healthy food.

" households in food deserts spend only slightly less in supermarkets. Households with income below $25,000 spend about 87 percent of their grocery dollars at supermarkets, while households with incomes above $70,000 spend 91 percent. For households in our “food deserts,” the supermarket expenditure share is only a fraction of a percentage point lower"

"one supermarket entry increases Health Index by no more than 0.036 standard deviations for low-income household"

They conclude that access to supply of healthy food or lack thereof explains only about 5% of the difference in the healthy eating index between low and high income households. Access does not appear to be driving the nutrition-income relationship.

2) Most of the differences in healthy vs unhealthy food choices by income group are driven by demand factors...i.e. preferences. When faced with the same choices and same prices, lower income households simply made purchases with a lower HEI.

"The lowest-income group is willing to pay $0.62 per day to consume the healthy bundle instead of the unhealthy bundle, while the highest-income group is willing to pay $1.18 per day."

They find that wealthier households value fruit three times the rate of lower income households and twice the rate for vegetables compared to lower income households.

Policy Implications

The authors reference studies by Montonen et al (2003) and Yang et al (2014):

"consuming one additional gram of fiber per 1000 calories is conditionally associated with a 9.4 percent decrease in type-2 diabetes" and consuming "3.5 fewer grams of sugar per 1000 calories is conditionally associated with a ten percent decrease in death rates from cardiovascular disease."

Improvements of the HEI definitely could be a driver for better health. However focusing on access may not be the greatest way to lever change. Certainly the correlations between income, food deserts, and healthy eating hold in this study and can be great flags to predict or identify which populations may need intervention. However, as this study points out the intervention should be based on theoretical and causal relationships that go beyond the supply of healthy foods and focus on aspects related to food preferences and demand. The authors conclude:

"For a policymaker who wants to help low-income families to eat more healthfully, the analyses in this paper suggest an opportunity for future research to explore the demand-side benefits of improving health education—if possible through elective interventions—rather than changing local supply."

References:

Drichoutis, A.C., Nayga, R.M., Rouse, H.L. et al. Food environment and childhood obesity: the effect of dollar stores. Health Econ Rev 5, 37 (2015). https://doi.org/10.1186/s13561-015-0074-2

NBER. "THE GEOGRAPHY OF POVERTY AND NUTRITION: FOOD DESERTS AND FOOD CHOICES ACROSS THE UNITED STATES." Working Paper 24094 (http://www.nber.org/papers/w24094)