Elsevier

Social Science & Medicine

Volume 74, Issue 2, January 2012, Pages 254-262
Social Science & Medicine

How does the business cycle affect eating habits?

https://doi.org/10.1016/j.socscimed.2011.10.005Get rights and content

Abstract

As economic expansions raise employment and wages, associated shifts in income and time constraints would be expected to also impact individuals’ health. This study utilizes information from the US Behavioral Risk Factor Surveillance System (1990–2009) to explore the relationship between the state unemployment rate and the consumption of various healthy and unhealthy foods in the United States. Estimates, based on fixed effects methodologies, indicate that unemployment is associated with reduced consumption of fruits and vegetables and increased consumption of “unhealthy” foods such as snacks and fast food. Heterogeneous responses are also identified through detailed sample stratifications and by isolating the effect for those predicted to be at highest risk of unemployment based on their socioeconomic characteristics. Among individuals predicted to be at highest risk of being unemployed, a one percentage point increase in the resident state’s unemployment rate is associated with a 3–6% reduction in the consumption of fruits and vegetables. The impact is somewhat higher among younger, low-educated, and married adults. Supplementary analyses also explore specific mediating pathways, and point to reduced family income and adverse mental health as significant channels underlying the procyclical nature of healthy food consumption.

Highlights

► Individual health and the business cycle are interdependent. ► Using nationally representative individual-level data, this study analyzes the effect of the business cycle on nutrition. ► Analyses generally reveal the procyclical nature of healthy food consumption. ► Accounting for confounding factors, this study finds that unemployment is associated with reduced intake of healthy food. ► Unemployment is also associated with higher consumption of snacks and fast food.

Introduction

As economic expansions raise employment and yield financial benefits, associated shifts in income and time constraints would be expected to also impact individuals’ health. While some studies suggest that health may decline during a recession (Charles and Decicca, 2008, Dee, 2001), the majority of U.S. studies point to health and healthy behaviors being countercyclical. Strengthened economies or income receipt are associated with increases in mortality, acute myocardial infarction, alcohol consumption, smoking, physical inactivity, and other outcomes related to health (Dustmann and Windeijer, 2000, Edwards, 2008a, Ettner, 1997, Evans and Moore, 2011, Gerdtham and Ruhm, 2006, Ruhm, 2000, Ruhm, 2005, Ruhm, 2007). The effects are mainly temporary, and there is some evidence that the adverse effects dissipate in the longer run. Xu and Kaestner (2010), for instance, estimate the structural effect of wages and hours worked on health behaviors among low-educated individuals in the United States, and find that an increase in working hours is associated with higher cigarette smoking, a reduction in physical activity, and fewer visits to the physician. They also find that increases in wages, due to expanded economic activity, are associated with higher levels of cigarette consumption.

Given these broad effects of economic activity on mortality and health, the challenge has been to understand the underlying mechanisms. At the micro level, individuals are adversely affected by unemployment and other negative economic shocks (Sullivan & von Wachter, 2009). However, as Miller, Page, Stevens, and Filipski (2009) note, it seems unlikely that all of the aggregate effects of the business cycle are mediated fully by the individual’s own labor force status. Beyond the individual’s own employment status, the link between the business cycle and health also reflects other potential pathways that operate individually and ecologically. Prior studies that have related aggregate macroeconomic conditions to health outcomes have generally estimated an average population effect, which can mask considerable heterogeneity and also does not inform the specific pathways that drive observed associations between macroeconomic factors and health. Some prior work has considered health behaviors such as smoking, drinking, physical activity, and preventive healthcare utilization. However, with the exception of Ruhm (2000), who analyzes fruit and vegetable servings in the United States, the literature has not considered how individuals’ eating habits respond over the business cycle – a potentially salient factor that may shed light on the mechanisms underlying the connection between the macroeconomy and health.

This study contributes to the literature by addressing these limitations. Specifically, we utilize micro level data representative of the U.S. population from the Behavioral Risk Factor Surveillance System, spanning 1990 through 2009, to explore the relationship between the economic cycle and food choices and the potential pathways underlying this link. The focus on healthy and unhealthy food consumption is policy-relevant given that caloric intake and nutrition are proximate inputs into obesity and overall population health. If the economic cycle impacts individuals’ health-related outcomes, then the effect would also be more easily identifiable in a statistical sense on health behaviors; health outcomes and obesity, on the other hand, tend to be cumulative and may not respond readily or over the short-term. We proxy the business cycle and economic activity with area (state-specific) unemployment rates in the United States to capture the effects on food consumption choices.

Whether unemployment increases or decreases healthy food consumption (and vice versa for unhealthy food consumption) depends on how it affects the marginal benefits and marginal costs of such food consumption. A higher probability of being unemployed or reduction in hours worked raises total available time but reduces income. For instance, Edwards (2008b) notes shifts in leisure time and associated time costs during recessions, and Evans and Moore (2011) point to linkages between income and economic activity and subsequent effects on mortality. This shift in time and income constraints would reduce the direct marginal cost of food consumption that is relatively more intensive in time inputs and less intensive in market inputs. Noting that healthy food consumption is generally more intensive in both market as well as time inputs relative to unhealthy snacks and fast foods, the effect of the macroeconomy on healthy food consumption is ambiguous depending on the relative intensity of market versus time inputs. Greater availability of time tends to raise the demand for home-cooked meals and healthy foods, but lower income tends to raise the demand for cheaper fast food and unhealthy food consumption.

The marginal cost of food consumption is further impacted by any changes in the relative monetary costs of food over the economic cycle. Hastings and Washington (2010) indicate, for instance, that prices for foods purchased by low-income households tend to vary positively with demand. Data from the Bureau of Labor Statistics (BLS) on the overall food price index, adjusted for inflation, reveal that food prices are generally mildly procyclical. De-trended and standardized real fruit and vegetable prices (1959Q1-2011Q1) are also weakly procyclical, as expected, decreasing during economic downturns and increasing during expansions (correlation with unemployment rate = −0.13). Interestingly, the relative price of healthy versus “unhealthy” food consumption (proxied by the price index for fruits and vegetables relative to the price of purchased meals and price of limited service restaurants, which includes fast food establishments) is similarly procyclical (correlation with unemployment rate = −0.4). Procyclical relative prices of healthy food consumption may temper any potential procyclicality of healthy food consumption and countercyclicality of unhealthy food consumption.

The marginal benefits of unhealthy food consumption may also increase during economic downturns. It is well-documented in the literature that unemployment leads to higher levels of stress, depression, and psychological distress (Dooley, Catalano, & Wilson, 1994), and depressed individuals tend to consume more calories and greater amounts of junk food (Wurtman, 1993). Loss of health insurance, as a result of job loss, may also have opposing effects on eating habits due to ex ante moral hazard and reduced contact with physicians (Dave & Kaestner, 2009).

The upshot of this discussion is that while there may be good reason to believe that individuals respond to the economic cycle by varying their food consumption due to shifts in the marginal costs and benefits, the direction of the impact is a priori ambiguous depending on the relative intensity of time and market inputs in the production process and the importance of these other potential channels of effect. There may also be considerable heterogeneity across different segments of the affected population. In this study, we assess the extent to which the economic cycle impacts healthy and unhealthy food consumption, investigate heterogeneous responses, and undertake an analysis of the potential mediators driving the connection between the macroeconomy and eating habits.

Consider the following reduced-form demand function linking measures of food consumption (HC) to the unemployment rate, a proxy for economic activity, multiple macroeconomic factors, and the overall business cycle1:HCismt=B0+B1UNEMPsmt+XismtΠ+μs+λm+νt+εismt

Specifically, the above model denotes that food consumption for the ith individual, residing in geographic area s in month m and year t, is a function of the unemployment rate (UNEMP) and other observable exogenous characteristics such as age, gender, race, education, and marital status (X), with ε representing an individual-level disturbance term.

Rather than the individual’s actual unemployment status, it is the area-specific unemployment rate that is the relevant and appropriate determinant in the demand for food consumption. First, actual unemployment only partially captures potential pathways through which the economic cycle may affect food choices. Even if an individual is not actually unemployed, the economic downturn affects the probability of becoming unemployed and would be expected to alter the marginal costs and benefits of caloric intake. For instance, psychological distress due to a higher probability of being laid-off or due to the unemployment of a spouse or family member may lead to a lower (higher) demand for healthy (unhealthy) food consumption. The decline in household wages or labor supply, even if the individual remains employed, would also be expected to shift the marginal cost of food consumption. Second, the use of area unemployment rates more proximally captures the effect of the economic cycle since within-area changes in the unemployment rate are strongly countercyclical. While the unemployment rate reflects the risk of unemployment, it is also reflective of other macroeconomic factors such as economic activity, income, and hours worked. Thus, the parameter of interest is B1, which is the reduced-form net impact of the unemployment rate on the individual’s food choices operating through all (and potentially competing) individual and ecological channels of effect.

While the area-specific unemployment rate is plausibly exogenous to the individual’s food consumption, the possibility of other confounding area-specific factors remains. To account for this “statistical endogeneity,” specifications control for area fixed effects (μs), which capture all unobservable time-invariant area-specific factors, and month and year fixed effects to capture unobserved seasonal factors and general trends. In addition, alternate specifications also control for state-specific linear trends to account for systematically-varying unobserved factors within a given state over time.

The estimation strategy proceeds in four parts. First, we estimate the impact of unemployment on healthy and unhealthy food consumption for the overall population, using the respondent’s resident state-level unemployment rate. The parameter B1 in our equation captures this average population effect of the unemployment rate on consumption choices. Since this average overall effect may mask considerable heterogeneity, next we also estimate differential effects based on models stratified across socio-demographic factors. At this stage, we also estimate models separately for healthy individuals for two reasons. First, health status may be endogenous to food choices; thus, restricting the sample to individuals in good health leads to a more homogeneous sample and bypasses this endogeneity. Second, restricting the analysis to healthy individuals isolates the direct demand for food consumption, whereas analysis on the full sample allows the models to capture the input demand for food consumption derived from the underlying demand for health (which may also be impacted over the economic cycle).

The affected population, that is individuals who are most at risk of unemployment during an economic downturn and therefore most responsive in their consumption choices, is likely to be small. In this case the overall population effect, which represents an intent-to-treat effect, substantially underestimates the response amongst the affected population. Thus, as a third step, we also modify the analysis to isolate the effect of unemployment risk among those who are most impacted by it. Here, we follow the general framework of Charles and Decicca (2008). We exploit the fact that certain socio-demographic groups (such as low-educated individuals) are much more likely to become unemployed as the unemployment rate rises in their state. Specifically, the following logit model is estimated to predict the unemployment status of an individual residing in area s at month m and year t, based on the area unemployment rate (UNEMP), predetermined or exogenous individual-specific characteristics such as age, gender, race/ethnicity, education, marital status, and health status (X), and interactions between these factors and the area unemployment rate. Indicators for state, month, and year are also included.UNEMPLOYEDismt=α0+α1UNEMPsmt+Xistθ+k=1KφUNEMPsmtXkismt+γs+τt+δm+ωismt

The parameter α1 and the vector Σφ capture the impact of area-specific unemployment rates on the individual’s actual unemployment status, allowing the effects to differ across socio-demographic cells.

The predicted probability (or propensity) of being unemployed captures variation across individuals with respect to their risk of unemployment. Note that this propensity score is clearly exogenous since it is a linear combination of the area unemployment rate and individual-specific predetermined factors (Dave & Kaestner, 2009). Thus, whether or not the individual is actually unemployed, the propensity score measures their proximal probability of being unemployed based on their socio-demographic characteristics and their surrounding rate of joblessness. We then estimate a modified version of our main model by interacting the area unemployment rate with this individual-specific propensity of being unemployed (denoted RISK).HCismt=Γ0+Γ1UNEMPsmt+Γ2(UNEMPsmtRISKismt)+XismtΠ+μs+λm+νt+εismt

The above equation is analogous to a difference-in-difference-in-differences specification, where the coefficient of the interaction term represents the differential effect of the area unemployment rate among individuals most likely to be affected, relative to individuals who are least at risk of being unemployed. Specifically, the parameter Г2 captures the added effect of the state unemployment rate on food choices among those individuals predicted to be most at risk of being unemployed, and Г1 captures the effect of unemployment risk on food choices among those individual who are at zero risk of being unemployed as predicted by the propensity score. As a falsification check, we expect Г1 to be insignificant and close to zero since individuals who are not at risk of being unemployed should not be affected by area unemployment rates. As a further specification check, we expect Г2 to be larger than B1 in absolute magnitude since the effects should be largest among the at risk population (Г2) whereas B1 captures the average effect among all affected and non-affected individuals.

Finally, we implement an analysis of potential mediators to inform on the strength of the specific mechanisms underlying the impact of unemployment risk on food choices. The baseline specifications are parsimonious and only include exogenous socio-demographic factors so as not to “over-control” for factors that may be potential pathways. In alternate analyses, we incorporate measures of actual work status, family income, food prices from ACCRA, mental and physical health, and health insurance coverage to gauge the extent to which the estimated effect of the state unemployment rate on food choices can be explained by these mediators.

Our analysis relies on the Behavioral Risk Factor Surveillance System (BRFSS), an individual-level data set representative of the population of the United States. As the largest telephone-based health survey available, the BRFSS has tracked health conditions and risk behaviors for adults 18 years of age and older in the U.S. The survey is conducted by state health departments in collaboration with the Centers for Disease Control and Prevention (CDC).

Measures of food consumption are included, although not consistently. Moreover, these variables are occasionally ‘module’ variables, asked of only a limited number of respondents, rather than ‘core’ variables, asked of all respondents. Consumption of carrots, fruit, fruit juice, green salad, and vegetables are asked consistently in years 1990–2009, with the exception of 2004, 2006, and 2008. The survey questions are generally phrased as follows: “How often do you eat (FOOD)?” Options are given for the respondent to record his/her answer in times per day, week, month, or year. Answers are converted to times per month. Month indicators are included in all models to capture seasonality in the consumption of healthy and unhealthy foods. While nutritionists caution using the terms ‘healthy’ and ‘unhealthy’ regarding foods in order to avoid classifying foods per se in preference for a focus on a balanced diet, we use the term healthy for the aforementioned foods as the food pyramid stresses their consumption. Moreover, most Continuing Surveys of Food Intakes by Individuals show consumption of fats, oils, and sweets, meant to be consumed sparingly, to be higher than recommended. Consumption of snacks, hamburgers, hot dogs, French fries, fried chicken, and doughnuts is not as frequently observed in the BRFSS, yet we also analyze these outcomes in order to compare these results with those of our healthy food outcomes.

Monthly state-level unemployment rates are obtained from the Bureau of Labor Statistics, Local Area Unemployment Statistics (LAUS) and matched to the individual records by state of residence and month and year of interview. Individuals working part-time and discouraged workers are not included among the unemployed. While more family members are likely to seek work during downturns and thus increase the unemployment rate, there has been limited empirical evidence for this added-worker hypothesis. For the subgroup analyses, we compute subgroup-specific unemployment rates (for instance, males, females, younger adults, older adults, etc.) based on the March CPS, and match these to the individual BRFSS records, in order to minimize measurement error and to account for possible differential movements in and out of aggregate unemployment statistics of different groups. While most studies in the literature continue to use aggregate unemployment rates for subgroup analyses, which confound multiple channels, Miller et al. (2009) do utilize age group-specific unemployment rates to consider differential effects on mortality.

In alternative specifications accounting for potential mediators, we also include measures of actual work status (whether the individual is currently working, either employed for wages or self-employed), real total family income, mental health (number of days in the past month that mental health was not good), physical health (number of days in the past month that physical health was not good), and current health insurance coverage from any source, all derived from the BRFSS.

Data on food prices are obtained from the Council for Community and Economic Research (formerly ACCRA), available at the city level. We first divided food prices by the ACCRA Cost of Living Index to account for differences across cities, and then aggregated by state and quarter, to form state-level estimates for each price.

We restrict the sample to individuals between the ages of 26 and 58 and further exclude retired individuals from the analyses, in order to focus on working-age adults who have completed their schooling and are still tied strongly to the labor force. Estimates and conclusions are not affected by expanding the sample to ages 21 through 64. This yields a final sample size of about 1.35 million for the analysis of healthy food consumption, and about 57,000 for the analysis of unhealthy food consumption.

Section snippets

Results

Table 1 presents estimates of the conditional impact of the state unemployment rate on measures of healthy food consumption. A higher unemployment rate in the respondent’s state of residence is associated with lower levels of consumption of fruits, juice, carrots, green salad, and vegetables. The effect is statistically significant for four out of the five outcomes, and is jointly significant across all models at the one percent level. Nevertheless, the magnitude of the impact is expectedly

Discussion

This study analyzes the effects of the business cycle, as proxied by state unemployment rates, on individuals’ food consumption choices in the United States. A variety of methodological approaches is used to assess robustness: state fixed effects to address area-specific unobserved time-invariant characteristics, state-specific trends to address systematically-varying unobservables over time within states, and seemingly unrelated regressions to account for the correlation across errors and data

References (21)

There are more references available in the full text version of this article.

Cited by (92)

  • Pay (for it) as you go: Prepaid energy meters and the heat-or-eat dilemma

    2022, Social Science and Medicine
    Citation Excerpt :

    Figure A1 (Appendix A) shows that, on average, 19.5 portions of fruit and vegetables are consumed per week by those using a PPM, compared to the sample mean of 27.4 portions consumed by post-payment customers – a difference of around one portion per day. Our analysis controls for a standard set of socio-economic and demographic covariates identified in the literature as being likely determinants of FVC (see, e.g., Devine et al., 2003; Dave and Kelly, 2012; Vinther et al., 2016; Cornelsen et al., 2019). We account for the UK's monthly food consumer price index (FOODCPI) (ONS, 2022), the (log) of annual equivalised household income (LNINCOME) and its polynomial (LNINCOME2) and for a six-group categorical variable representing their situation in the labour market (UNEMPLOYED, RETIRED, STUDENT, DISABILITY, and OTHER_JOBSTATUS).

View all citing articles on Scopus
View full text