Article Text

Download PDFPDF

Food outlet visits, physical activity and body weight: variations by gender and race–ethnicity
  1. L Frank1,
  2. J Kerr2,
  3. B Saelens3,
  4. J Sallis4,
  5. K Glanz5,
  6. J Chapman6
  1. 1
    University of British Columbia, British Columbia, Canada
  2. 2
    Health Promotion/Behavioral Sciences, School of Public Health, San Diego State University, San Diego, California, USA
  3. 3
    Pediatrics/General Pediatrics, University of Washington, Seattle, Washington, USA
  4. 4
    Department of Psychology, San Diego State University, San Diego, California, USA
  5. 5
    Department of Behavioral Science and Health Education, School of Public Health, Emory University, Atlanta, Georgia, USA
  6. 6
    Lawrence Frank and Company, Inc, Seattle, Washington, USA
  1. Dr Lawrence Frank, Associate Professor, J. Armand Bombardier Chair in Sustainable Transportation, University of British Columbia, 235-1933 West Mall, Vancouver, BC, Canada, V6T1Z2; ldfrank{at}


Purpose: Recent evidence documents significant associations between community design, physical activity and obesity when adjusting for demographic covariates. Yet it is well understood that energy imbalance and weight gain are also a function of dietary patterns, and perhaps the degree of access to healthy food choices.

Methods: The current study builds upon the Atlanta-based SMARTRAQ study of over 10 000 respondents and reports an integrated assessment of obesity impacts of physical activity and food outlet visitation. Respondents in the SMARTRAQ survey aged 25–65 provided BMI, self-reported physical activity levels (IPAQ), demographic factors, and where they went for food over a 2 day period.

Results: The relative effect of physical activity, neighbourhood walkability, and food outlet visitation on BMI differed significantly across gender and ethnicity. BMI in females increased with fast food and decreased with grocery store visitation and physical activity, but not with walkability or walking. BMI in males was not related to where they went for food but decreased with walking and overall physical activity and with walkability. Fast food visitation was associated with increased BMI in white respondents and grocery store visitation with decreased BMI in black respondents. Meeting moderate activity guidelines was associated with lower BMI in both black and white respondents, yet walking was only significant in predicting reduced BMI in white respondents.

Conclusion: Obesity influences of physical activity, walkability, and where people go for food differ significantly across gender and ethnicity and offer important policy implications and insights for future research.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

The purpose of the present study was to examine the relative contributions of both diet-related and physical activity-related behaviours in explaining BMI in population subgroups defined by sex and race. The diet-related variables seek to capture travel to fast food restaurants and grocery stores, variables that have been seldom studied but that appear to be modifiable. Multiple physical activity variables are examined, and analyses are controlled for sociodemographic characteristics and address the built environment context to which people are exposed where they live, work, and visit.

Approximately one-third of the US adult population is considered obese (BMI>30), with almost two-thirds meeting criteria for overweight (BMI>25).1 Obesity and overweight prevalence are even higher for non-Hispanic black Americans. Although obesity prevalence is comparable across race/ethnicity in men, it is much higher for black women than for white women. More than 50% of non-Hispanic black women meet BMI criteria for obesity and more than 80% are considered overweight in the US.1 Obesity prevalence in non-white people is higher among women than men.1 A better understanding of the factors that contribute to these disparities across gender and ethnicity/race is needed to address the obesity problem.

The dramatic and sustained increase in overweight/obesity prevalence over the past few decades could be in part the result of a shift toward more away-from-home food consumption.2 A large percentage of away-from-home food is purchased from fast food establishments. For instance, recent studies find that 26.5% of adults in a national 2 day dietary recall sample report eating fast food3; more than 50% of Minnesotans report eating fast food at least once per week4; 76% of African Americans in North Carolina report eating fast food at least once in the prior 3 months (with more than a quarter reporting eating fast food “often” or “usually”)5; and the CARDIA study in four US cities reports an average rate of young adults eating at fast food outlets 1.3–2.4 times per week.6

Higher frequency of fast food consumption is related to poor dietary quality,3 5 7 8 prospective weight gain, and higher BMI.4 6 Also, frequency of fast food consumption appears to differ by demographic factors. Men consume fast food more often than women,3 9 and some evidence points to black people eating fast food more frequently than white people,3 6 although not all published research finds a race/ethnic difference in frequency of visits to fast food restaurants.9

High rates of obesity, fast food consumption, and associated demographic differences have led to an interest in other locations of food procurement, particularly grocery shopping. The assumption is that grocery shopping provides the opportunity for more healthful food choices than those provided by fast food. In 2005 the average American consumer made an average of 2.1 trips to the supermarket per week.10 Handy and Clifton report that Austin (Texas) residents recall on average more than 15 trips to a food store in a month and more than half of such trips were to a supermarket.11 Another Texas-based study found that more than a third of their sample reported that food shopping was most often one big weekly trip with a few small shopping trips throughout the week. In this study, frequent shoppers were more likely to be Hispanic or Asian, but less likely to be African American.12 While a high percentage of food consumed comes from grocery stores,13 little is known about the relation between grocery store shopping and weight status or whether this association differs by individual demographic factors.

These food purchasing behaviours may contribute to BMI; however, because BMI is ultimately influenced by energy balance, it is important also to examine the role of physical activity in this equation. To our knowledge, few (if any) community-based studies examine both activity and food sources within population segments. Multiple types of physical activity and sedentary behaviours, e.g., overall activity,1416 walking for transport,17 18 and driving/commuting time1921 have been found to be related to BMI, and each may interact with the type of food sources that people use.


This cross-sectional study employs travel and activity data from the SMARTRAQ (Strategies for Metropolitan Atlanta’s Regional Transportation and Air Quality) survey in the 13 county Atlanta region in 2001–2002. Data collection from 8069 households and about 18 000 participants was stratified across four ranges of income and household size and five levels of net residential density to ensure a variation in sociodemographics and urban form across participants. A subset of these participants between the ages of 20 and 65 and who were white or black are reported on in this study. A demographically representative sample of the Atlanta region was achieved through focused recruitment in predominantly ethnic minority and lower-income neighbourhoods of the Atlanta region that were also found to be more walkable.

Travel and activity data were captured in a 2 day diary where each destination visited, and the travel mode used to get there, were recorded for all household members 5 years and older. The response rate was the ratio between completed interviews and total eligible sample called on the telephone. The response rate was calculated for recruitment and retrieval of data. The overall response rate was determined by multiplying the two resultant rates. The overall response rate was 30.4 per cent. Sociodemographic and attitudinal information, height and weight and physical activity patterns were provided by a head of household in a recruitment call through the use of a computer aided telephone interview (CATI) protocol. Subsequently, a travel diary was administered to each household member 5 years and older and data from this diary and additional information were downloaded in a retrieval call from each survey participant 15 years and older. A parent or legal guardian provided this information for those 14 years and younger. Height, weight, and physical activity patterns, type, frequency and duration were reported individually by household members 16 years and older as part of the retrieval call.


A disaggregate approach to data collection was employed, resulting in observation-specific measures of food outlet visits, physical activity, and of the built environment within a kilometre of where participants live, work, or visit.

Built environment

Observation-specific measures of the built environment where each person lived were created using Network Analyst, which is an extension to the GIS software product developed by the ESRI Corporation known as ARCVIEW. This disaggregate approach enabled a 1 km road network buffer to be developed around each of the respondents’ place of residence, employment, and other destinations visited. Several papers have since been published employing these methods to measure the built environment.16 17 19 22 A combination of county-level Tax Assessors (parcel) data, census data, and street network data were used to operationalise vectors of:

  • Net residential density (households per residential land area)

  • Mixing of uses (presence and evenness of distribution of residential, commercial, and office uses), and

  • Street connectivity (intersection density capturing the degree of route directness between destinations)

  • Retail density (retail square footage/retail land area per retail parcel)

Values for each of these measures were normalised for the sample and summed to create a measure of walkability. These are the same methods already published in two separate papers linking built environment measures with physical activity and obesity in a sample of adults from the SMARTRAQ travel survey.16 17

Travel and activity data

A paper travel diary was filled out recording destinations visited, travel mode and purpose, and time of day across two preassigned days for each member of each household. Travel data were collected for all days of the week across the sample as a whole. Household travel diaries are a standard tool used within the transportation industry to capture and assess temporal and spatial dimensions of personal and household travel.23 Primary and secondary activities that took place at different destinations were reported. Nearest cross streets and/or names of destinations visited were reported.

Distance walked was calculated using GIS software, ArcView GIS 3.2 (ESRI Inc., Redlands CA, 2000), and a street network. For each trip, the origin and destination obtained from the self-reported travel diary were placed on the street network. The shortest distance (walking) path between the origin and destination was found, and actual road network distances were calculated for each trip.

Physical activity was measured using a validated, self-reported instrument known as the International Physical Activity Questionnaire (IPAQ).24 These measures of physical activity complement the travel data activity reported above and provide an overall measure of physical activity. Self-reported heights and weights were converted to metres and kilograms, respectively. We computed BMI as weight in kilograms divided by height in metres, squared. Only participants with data available for all variables were included in the analysis.

Food locations visited

Trip end destinations were classified as either fast food or grocery-based on two primary indicators — the type of activity reported and the destination’s name. The primary and secondary activities, and name and address of each destination visited over the 2 day survey period, were provided by survey participants. Primary activity (purpose) and up to four additional (secondary) activities were also provided.

Table 1 shows the distribution of reported main activities. Two of these activities are food-related, “Eating/Preparing Meals/Dining Out/Drive-Thru” and “Incidental Shopping (Gas, Groceries, Medicine).” The second activity is not necessarily food-related.

Table 1 Primary food-related activities and destinations (unweighted, all ages)

Nearly 30 per cent of all destinations were coded into a food-location type (fast food or grocery) if one of the two food-related activities occurred as either the primary activity or as one of up to four secondary activities. This subset of destinations was coded based on the name of the location. The results were reviewed and field verified based on the authors’ local knowledge of the region. Destinations were assigned to the “fast food” category after identifying and manually reviewing all names which included the following words: burger, burrito, cafeteria, chicken, deli, food court, hot dog, pizza, sub, taco, wings. Regional and national chain names, e.g. Burger King, Kentucky Fried Chicken, Krystal, McDonald’s and Mrs. Winner’s, were identified and included in the category as well.

A similar process was used to assign destinations to the “grocery” category. Locations with the word “grocery” in the name were reviewed and included in the category. Regional and national chains, e.g. Kroger, Publix, were also identified and included. The sample was restricted to ages 25–64 to reduce variations due to level of independence and health and to have a more homogenous sample in terms of life stages. In addition, the very elderly tend to experience a lowering of BMI towards the end of the lifespan.25

Linear regression analyses were performed using demographic26 and urban form14 19 variables which have been shown to be related to obesity. Body Mass Index is the dependent variable in each model. A linear regression on the whole sample indicated that gender and race (white or black) were strong correlates of food outlet visitation, so the analyses were stratified by these demographic factors. Independent variables included age (continuous), education (degree or not), household income ($10K increments from $10K to greater than $100K), number of residents in household (continuous), number of vehicles in household (continuous), number of children in household (fewer than two vs two or more), and employment status (part or full-time vs not). These demographic variables were entered in to the first block of the stepwise linear regression.

In the second block, walkability of the home environment was entered as a continuous variable. In the third block food locations visited and physical activity behaviour variables were entered: went to a fast food restaurant (at least once over the 2 day diary vs not) and went to a grocery store (at least once over the 2 day diary vs not). In the fourth block activity behaviour variables were entered: the minutes of moderate and vigorous activity from the self-report survey were combined. Those reporting 150 minutes or more of moderate or vigorous activity over the 7 day recall period were considered to meet the ACSM guidelines (moderately active at least 5 days a week). Walked at least once over the 2 day survey was entered, as was time spent in car (less than 1 h vs 1 h or more over the 2 day period). Sample characteristics are presented as means, standard deviations and percentages for the stratified gender and race subgroups. Linear regression results are presented in the analyses with the explained variance provided for each block along with t ratio statistical significance (p values). Variables that were significant in the final models are highlighted with an asterisk (*).


Gender stratification

Measures of central tendency (means) or percentages and measures of dispersion (standard deviations) for the sample are shown in table 2.

Table 2 Sample characteristics stratified by gender

Gender-stratified linear regression results are shown in Table 3. Results are very different across genders. Only employment status and number of vehicles in household were not significantly related to BMI for women whereas income, presence of two or more children, and employment were not significant for men. Younger participants, white people, those with a degree, those with a higher income, fewer in the household overall, but two or more children in the household, were less likely to be obese. In contrast to female respondents, walkability of the local neighbourhood was significantly and inversely related to BMI for men.

Table 3 Gender-stratified linear regression

Women who reported going at least once to a fast food restaurant were significantly more likely to have a higher BMI and those who went to a grocery store had lower BMIs. These same food outlet variables were not significant explanatory factors for BMI in men. Meeting physical activity guidelines was significant and negatively correlated with BMI for men and women. MVPA was negatively correlated with BMI, but walking behaviour was additionally related to BMI in men. All the variables shown as significant at the p = 0.05 level remained significant in the final model. This suggests that, even after controlling for physical activity, neighbourhood environment is related to BMI for men. A total of 14.4% of the variance was explained for women and 6.8% for men by these variables. Most of the variance is explained by the sociodemographic variables.

Race–ethnicity stratification

Table 4 presents the characteristics of the sample stratified by race.

Table 4 Sample characteristics stratified by race

Table 5 presents the linear regression model for white and black people. The results suggest that factors explaining BMI in white and black people are quite different. In black people, only education and number in household, two or more children and employment status were significantly related to BMI. Those with a degree, fewer in the household, but two or more children, and employed were less likely to be obese. Every demographic factor except employment status was a highly significant predictor of BMI in white people. Those in less walkable neighbourhoods tended to have higher BMIs, but this was not significant. Shopping at a grocery store and meeting MVPA guidelines were negatively correlated with BMI in black people. Eating at a fast food restaurant was related to higher BMI while MVPA and walking behaviour were significantly related to lower BMI for white people.

Table 5 Race-stratified linear regression

Due to the strong influence of gender in white people and race in women, the analyses were further stratified by both groups. Table 6 shows the descriptive statistics and Table 7 the linear regression results for each of the four gender–race groups (e.g. black men).

Table 6 Sample characteristics stratified by gender and race
Table 7 Linear regression analyses to assess the relationship between the independent variables and BMI, stratified by gender and race

In black men, education was related to obesity, but this did not remain significant in the final model. Those with a degree were less likely to be obese. Black men living in a more walkable neighbourhood tended to have lower BMI scores, approaching the p = 0.05 significance level. The only factor that was significant at this level was shopping at a grocery store resulting in significantly lower BMIs. In black women, all demographic variables were related to BMI in the expected direction. In addition, meeting physical activity guidelines was negatively related to BMI.

In white men, nearly all of the demographic variables were related to BMI in the expected direction. Employment status and presence of two or more children in household were not significant. White men in more walkable environments had lower BMIs, after controlling for MVPA and walking, which were also related to BMI. In white women, age, education, number in household and income were significantly related to BMI in the expected direction. Eating at a fast food restaurant was related to higher BMI and meeting physical activity guidelines was negatively related to BMI.


The main finding of the present study was that use of specific food sources, as well as physical activity and sociodemographic factors, and neighbourhood walkability contributed modestly to explaining variance in BMI. However, most interesting is that the pattern of findings varied dramatically by gender and race. The present study is one of the few that examined both dietary and physical activity-related behaviours, so another notable finding was that both domains made independent contributions to BMI. The findings that eating at a fast food restaurant and shopping at a grocery store had expected but opposite-direction associations with BMI support the need to raise public awareness about the important role of food sources and policy efforts to ensure that everyone has ready access to healthful restaurants and grocery stores. While food outlet visitation and physical activity were significant in many cases, most of the variance explained came from demographic variables.

Eating at fast food restaurants has been linked to poorer dietary quality,3 5 7 8 so it is likely that placing oneself in such an environment could decrease the chances of making healthful choices and increase the chances of overeating. Grocery stores offer extensive variety and a more balanced selection than alternative sources like convenience stores that have few healthful choices like fruits and vegetables.27 Present findings support the concept that environments consumers encounter inside food establishments can affect food selections and health outcomes such as obesity.28 However, the current cross-sectional findings cannot provide information regarding the presence or direction of a causal relationship.

Men and women had clear differences in correlates of BMI. Both eating at fast food restaurants and shopping at grocery stores were significant correlates for women, but shopping at a grocery store was only significant for black men. The dietary-related variables explained 0.6% of the variance for women but only 0.1% of the variance for men. Though both physical activity variables were significant for men, only one was significant for women. Physical activity explained more variance in women’s BMI (1.1%) than in men’s BMI (0.7%). Further study is needed to identify why use of specific food sources is more related to women’s than to men’s BMI. Demographic variables explained more variance in women’s BMI (12.7%) than men’s (5.5%), with much of this difference accounted for by stronger contributions of race, income, and children in the household for women.

Higher BMIs among black people (mean of 27.9) than white people (mean of 26.0) in the present study are generally consistent with other studies29 and highlight the need to understand correlates of BMI for both races to inform strategies to reduce the disparities. Both the diet-related and physical activity-related variables explained slightly more variance in BMIs of black than white people. For the diet-related variables, only eating at fast food restaurants was a significant correlate for white people, and only shopping at a grocery store was significant for black people. Meeting physical activity guidelines was significant for both races, but walking for transportation was only significant for whites. Though it can be concluded that both diet-related and physical activity-related variables made independent contributions to BMI for both races, it was not clear why the pattern of correlations differed by race. Demographic variables were related in the expected directions with BMI for both races, but associations were generally stronger for white people.

Patterns of association varied by subgroups defined by race and sex. Eating at a fast food restaurant was associated with higher BMI only for white women, and shopping at a grocery store was associated with lower BMI only for black men. The latter finding was somewhat unexpected because in the sex-specific analyses shopping at a grocery store was significant only for women. Meeting physical activity guidelines was associated with lower BMI for three of the four subgroups, all except black men. Walking for transportation was a significant protective factor, but only for white men. Meeting physical activity guidelines was the most generalised behavioural correlate of BMI among those evaluated. Shopping at a grocery store was significant for one subgroup and marginally significant for two other subgroups, so there is some indication of a generalised protective effect.

Variance in BMI explained by the food source and physical activity variables was small, about 0.5% to 2%. Nevertheless, at least one behavioural variable was significant for every subgroup examined, and the prevalence of all the “risky” behaviours was large. It is well known that demographic and socioeconomic variables explain substantial proportions of variance in BMI, but, with the exception of education, many of these variables are difficult or impossible to change. Thus, incremental improvements in understanding modifiable behavioural and environmental variables are more important because such data can lead to more effective interventions. Present results indicate that the places where food is purchased may be a useful behaviour change target and confirm that increasing overall physical activity as well as walking for transportation can be recommended for weight control.

Sex differences in prevalence of behavioural variables were modest, with the only notable differences being that women were more likely to shop at grocery stores during the 2 days of monitoring, and men were more likely to meet physical activity guidelines. Race differences in behavioural variables were mostly minor and confirmed previous findings. For example, white people were more likely to meet physical activity guidelines than black people30 and to spend more time driving or riding cars,19 presumably due to vehicle ownership, employment location and higher income.

Strengths of the present study included the large sample with a large proportion of African Americans that allowed subgroup analyses, inclusion of both diet-related and physical activity-related variables, exploration of seldom-studied use of specific food sources, and adjustment for many demographic variables. Several limitations must be considered. Though use of food sources had to be defined by the combination of land use (i.e., address coded as fast food or grocery store) and “activity” code (i.e., eating or shopping), there is the possibility of misclassification of food source type. Shopping and eating-out behaviours were assumed based on where people went, whereas physical activity was captured through a direct self-report. Specific food intake patterns were not captured either. For example, another household member may have shopped at a grocery store or taken food home from a fast food restaurant, and those behaviours could affect the respondent’s eating behaviour and BMI. Women are more than three times as likely as men to do the household grocery shopping.10 The 2 day travel diary was likely an insufficient sample to estimate habitual patronage of fast food restaurants and grocery stores or usual walking for transportation. Consumers are about twice as likely to shop on a Saturday or Sunday as on weekdays,10 so the 2 day diary may not be a representative sample of grocery store use. In addition, our study included multiple observations in several households, resulting in clustering effect of data.

These limitations could potentially reduce the power to detect associations, so true associations of the behaviours with BMI may be stronger than those reported. Another limitation was that use of other food sources, such as produce markets, convenience stores, and sit-down restaurants, could not be determined from the data available. Although it is hypothesised that any effect of use of food sources on BMI would be mediated by actual eating behaviour, no dietary intake data were available for analyses. Investigators are encouraged to assess all components in the presumed causal chain in future studies. Finally, healthfulness of food environments inside fast food restaurants and grocery stores was inferred from the category, and stronger data on the role of food environments on eating behaviours and weight status require direct measures of the food environment as provided by observational measures.27 31 Future research should evaluate the presence and effect of potential interactions between neighbourhood walkability and food outlet quality.

Adults who choose to visit fast food restaurants and shop at grocery stores may do so because of strong preferences for foods available in those establishments, because their choices are constrained by access to food sources, because of considerations of food cost, or because of a combination of these factors. Use of food sources was associated with BMI in several population subgroups, indicating some generalisation of effects. Multiple strategies are required to reverse the obesity epidemic and there are at least two strategies to change use of food sources that were identified in the present study as a risk factor for higher BMI. Educational campaigns could recommend reducing visits to fast food restaurants and increasing visits to grocery stores to control weight. Policy interventions to increase access to grocery stores, limit the concentration of fast food restaurants, or improve the availability and promotion of healthful foods within fast food restaurants should be evaluated for their effects on eating behaviours and weight status. Another implication is that interventions to change multiple diet-related behaviours and multiple physical activity-related behaviours may be needed to control obesity. Although associations varied by subgroup in the present study, further investigation is needed to confirm the behavioural targets that are most important for each subgroup. Such information is needed so that interventions can be targeted to the needs of each major sociodemographic group.

Significance for clinicians and public health workers

The current study is innovative because it assesses both activity and food access patterns in relation to community design within a single analytical framework. Results indicate that systematic associations exist between community design and food purchasing patterns and body weight for certain population subgroups but not for others. More specifically, considerable differences appear to exist between physical activity and food outlet visitation, as moderators of body weight, across gender and ethnicity. Results suggest considerable complexity between residential behaviour settings and health-related outcomes and the potential for costly built environment interventions to be ineffective without careful consideration of the demographic characteristics of targeted populations. On the surface, it would appear that reductions in BMI may be most achievable through interventions that increase access to sit-down restaurants and grocery stores for women and black populations, and through increased levels of walkability and physical activity in men. The results shown here merely scratch the surface. It is likely that underneath lie a set of mechanisms constituting causal pathways between behaviour settings and body weight that are only beginning to be understood. It appears that additional programmatic actions will be required that remove constraints on time and resource availability for black people and that increase safety and security for women to walk for potential benefits in BMI from changes in neighbourhood design to be realised across a broader population.


We would like to thank the Robert Wood Johnson Foundation for their generous support of this research and the Georgia Department of Transportation and The Georgia Regional Transportation Authority for their support of the original data collection upon which this paper is based.



  • Competing interests: None.

Linked Articles