- Click here for explanation of classification scheme.

The associations between BMI and overweight and the availability of four types of grocery food stores, including chain supermarkets, non-chain supermarkets, convenience stores, and other grocery stores, holding constant a variety of other individual- and neighborhood-level influences were examined.

Students in the 8th and 10th grade participating in the Monitoring the Future surveys.

None listed

Recruitment

This study drew on individual-level national data for 8th- and 10th-grade students from the MTF study.

Design

Repeated cross-sections of individual-level data on adolescents drawn from the Monitoring the Future surveys linked at the ZIP-code level to data on food store availability.

Blinding used

Not applicable

Intervention

Not applicable

Statistical Analysis

• The goal of the empirical work was to estimate the association between access to various types of food stores and adolescent weight, holding constant a variety of socioeconomic characteristics which may be correlated with both weight outcomes and neighborhood characteristics 
• Per capita availability of alternative types of food stores proxy the opportunity cost of the time spent acquiring healthful food
• Reduced form models of individual BMI of the form  BMI_ist =B₀ + B₁FT_ist + B₂OC_ist + B₃X_it + B₄D_it + e_ist  were estimated by ordinary least squares (OLS), where FT_ist was a vector measuring food store outlet density available to individual i in geographic area s at time t, OC_ist measured other local-area contextual factors including per capita income, food prices, and restaurant availability in area s, D_it was a vector of year dummy variables, X_it was a vector of individual and household characteristics, B were conformable vectors of parameters to be estimated, and e_ist was a disturbance term.
• The characteristics in the vector X_it, included race/ethnicity, grade, highest schooling completed by father, highest level of schooling completed by mother, a rural/urban indicator, total student income, weekly hours of work by the student, and whether the mother worked part-time or full-time. In X_it complete sets of gender-specific age dummy variables also were included to remove gender-specific differences in BMI growth. These dummies implicitly included both a constant and a gender dummy.
• The coefficients on other covariates may then be interpreted as reflecting variation around arbitrary gender-specific growth curves
• The inclusion of the year dummy variables in the model was equivalent to non-parametrically detrending each variable in the analysis such that the estimates do not reflect common trends
• Neighborhood per capita income was included in the model to account for local-area wealth effects distinct from food store availability that may affect health outcomes and variation in food store density related to local incomes
• To control for other factors that may be related to weight outcomes through food access channels, full-service and fast food restaurant outlet density measures were included in the model as well as fast food and fruit and vegetable prices
• A Huber-White covariance matrix estimate which is robust to clustering at the ZIP-code level and heteroskedasticity of unknown form was used
• Finally, the full model was also estimated with overweight as the outcome using maximum likelihood probit regression, for which the marginal effects were reported. 

Timing of Measurements

• The external outlet density and food price measures were matched to the individual-level data at the school ZIP-code level for each year 1997 through 2003
• Data on per capita income were drawn from the Census 2000.

Dependent Variables

• Body Mass Index and overweight: Formula for BMI is given in statistical analysis section. Adolescents were classified as overweight when BMI at least age-gender-specific 95^th percentile based on the Centers for Disease Control and Prevention (CDC) growth chart.

Independent Variables

Availability of four types of grocery food stores that included chain supermarkets, non-chain supermarkets, convenience stores, and other grocery stores. 

Control Variables

Controlled demographic measures available in the student surveys included:

• Gender
• Grade
• Age
• Race/ethnicity
• Highest schooling completed by father
• Highest level of schooling completed by mother
• Rural/urban area neighborhood designation
• Total student income (earned and unearned, such as allowance) in real dollars (CPI base $82-$84)
• Weekly hours of work by the student
• Whether the mother works part-time or full-time. 

• Initial N: 73,079 (47.54% male)
• Attrition (final N): Assumed the same
• Age: 14.6542 (SD 1.1640)
  • 8th grade 48.69%
  • 10th grade 51.31%
• Ethnicity:
  • White 69.66%
  • Black 10.59%
  • Hispanic 9.67%
  • Other race 10.08%
• Other relevant demographics:
  • Father less than high school 13.06%
  • Father complete high school 29.43%
  • Father college or more 57.51%
  • Mother less than high school 11.11%
  • Mother complete high school 28.00%
  • Mother college or more 60.89%
  • Live with both parents 80.02%
  • Live in rural area 24.10%
  • Students' weekly real income (in 100s)  0.2281 (SD 0.2666)
  • Hours worked by student 3.8560 (SD 7.1366)
  • Mother does not work 17.60%
  • Mother works part-time 18.28%
  • Mother works full-time 64.12%
  • Per capita income (in 10,000s) 2.2107 (SD 0.9665) 
  • Number of grocery stores 3.2835 (3.0097)
  • Number of convenience stores 2.1535 (2.2501)
  • Number of chain supermarkets 0.3037 (0.5805)
  • Number of non-chain supermarkets 0.2609 (0.5906)
  • Number of fast food restaurants 2.6009 (2.2078)
  • Number of non-fast food restaurants 11.4236 (9.2185)
  • Price of fast food 2.7127 (0.1740)
  • Price of fruit and vegetables 0.7205 (0.1046)
  • Years from 1997 through 2003 evenly represented
• Anthropometrics:
  • BMI 21.8059 (SD 4.2947)
  • Overweight 10.28%
• Location: The MTF study conducted at the University of Michigan's Institute for Social Research annually surveyed nationally representative samples of high school seniors in the coterminous United States. Since 1991, the MTF surveys also included over 30,000 8th- and 10th- grade students annually.

 Key Findings

• Table 3 in the article reports the results from the BMI and overweight regressions [large thus not entered here]. The table also provides results for three additional BMI model specifications: (1) no restaurant outlet density control variables; (2) no restaurant outlet density or food price control variables; and (3) no restaurant, food price, or local-area SES control variables.
• With the full model, it was found that availability of chain supermarkets had a statistically significant negative relationship with adolescent BMI and overweight status. Each additional chain supermarket outlet per 10,000 capita was estimated to reduce BMI by 0.11 units and to reduce the prevalence of overweight by 0.6 percentage points.
• An additional convenience store per 10,000 capita was associated with a 0.03 unit increase in BMI and a 0.15 percentage point increase in overweight
• Increased availability of grocery stores had a very small positive and statistically weak association with overweight
• The results implied that part of the positive correlation between adolescent BMI and convenience and grocery stores was attributable to greater density of these outlets in low-income neighborhoods
• The results also suggested that some of the correlation between income and adolescent body weight was attributable to differential access to food stores in low-income neighborhoods
• One additional local-area chain supermarket per 10,000 capita was associated with lower BMI among African-American students by 0.32 units whereas the associated BMI of white and Hispanic students was lower by 0.10 and 0.09 units, respectively
• The effect of a chain supermarket was slightly higher in the subsample of students whose mothers worked full-time compared to those with mothers who worked part-time and roughly four times as great compared to students whose mothers did not work.

Limitations as cited by authors:

• The outlet density data were linked to the individual-level data by the student's school's ZIP code. There might have been measurement error in the density area to the extent that students lived in different areas than their schools, which might have been a particular problem for the high school subsample.
• The estimated coefficients on food stores might only be interpreted as causal if, holding everything else in the model constant, variation in food store density came from the supply side (for example, variation in local zoning laws) or if supply was perfectly inelastic. However, if all else equal, some of the variation in food store density was because of variation in demand across ZIP codes, the estimated associations cannot be interpreted as recovering the causal effect of changes in density on adolescent weight, given that supply was not perfectly inelastic.

Economic and urban planning land use policies which increase the availability of chain supermarkets may have beneficial effects on youths' weight outcomes.

Government:	National Institute on Drug Abuse
Not-for-profit	Robert Wood Johnson Foundation Other non-profit:

• The statistics seem appropriate; however, these are procedures I am not as experienced with
• For the relevance questions on the quality rating sheet, I treated the availability of chain supermarkets as the 'intervention'. I marked unclear for number 4 as that would depend on the economic situation of a particular area. Not sure if this was the way to do these questions, so please let me know if you want me to change them.