Reply to M Bishop

Center for Human Nutrition
Bloomberg School of Public Health
Johns Hopkins University
615 North Wolfe Street
Baltimore, MD 21205
E-mail: ywang{at}jhsph.edu

Qi Zhang

School of Community and Environmental Health
Old Dominion University
Norfolk, VA
E-mail: qzhang{at}odu.edu

Dear Sir:

We appreciate Dr Bishop's interest in and comments on our recent study (1), and we share her viewpoint that readers should be aware of the potential limitations of using poverty income ratio (PIR) as an indicator of socioeconomic status (SES), as well as of the limitations of other variables, such as education and occupation. As stated in our study, each of the commonly used variables of SES has its own strengths and limitations for studying the relation between SES and health outcomes. We agree that readers need to know how the PIR was constructed. It is the ratio of income to the family's appropriate poverty threshold set by the US Census Bureau in a given calendar year. The Census Bureau, working in accordance with the Statistical Policy Directive of the Office of Management and Budget, uses a set of money income thresholds that vary by family size and composition to determine who is poor. Thresholds are updated annually for inflation by using the Consumer Price Index. The official poverty definition counts pretax money income and excludes capital gains and noncash benefits such as public housing, Medicaid, and food stamps (2). PIR values for the National Health and Nutrition Examination Survey (NHANES) participants were computed by the National Center for Health Statistics by using family income data (3, 4). As pointed out by Bishop, a major limitation of the PIR is that the same thresholds are used throughout the United States—ie, they do not vary geographically.

Bishop also draws readers' attention to the facts that changes in several factors, such as the cost of living, happen over time and that family structure may vary across geographic regions and ethnic groups, and she points out that these changes may affect the meaning of income or PIR values. Ideally, if available, such variables should be accounted for in an analysis of the relation between income and the other health outcome variables. In our analysis, ethnicity was considered in stratified models; however, good geographic variables are not available.

Furthermore, previous studies indicated the trade-off between the feasibility and the desirability of adjusting the poverty thresholds (5). A lack of adequate data on the cost of living is a major barrier. The cost-of-living index published by the American Chamber of Commerce Researchers Association may be the best available data source, but it covers only a limited number of regions. Merging NHANES and American Chamber of Commerce Researchers Association data by using geographic indicators can cause 60% data attrition (SD Lakdawalla et al, unpublished observations, 2005). We conducted additional logistic regression analysis by using the NHANES I, II, and III data after control for variables of the 4 census regions that were made available in the NHANES data (ie, North, South, Midwest, and West) and urban or rural residence as proxies of differences in the cost of living. These further adjusted odds ratios (OR) and 95% CIs were changed only slightly from those that were not further adjusted. For example, for NHANES III, the OR and 95% CIs without and with control for regions and urban or rural residence, respectively, were 1.190 (0.856, 1.654) and 1.189 (0.857, 1.650) for low-income groups and 0.768 (0.530, 1.112) and 0.777 (0.539, 1.120) for high-income groups. We could not conduct this sensitivity analysis for the NHANES 1999–2002 data, because, since 1999, such geographic indicators have not been released in the public domain, in an effort to protect survey participants' privacy. Further research is needed to study the effect of the cost of living on our conclusions if better data are available.

In addition, although it is true that another limitation of the poverty threshold is that it does not account for family expenses, such as childcare costs for working and nonworking families, we suspect that the distribution of such families is likely to be random across different SES groups. Thus, not accounting for expenses such as childcare costs for working and nonworking families is unlikely to bias our results regarding the SES disparities in obesity or the trends in the relation between SES and the prevalence of obesity. Furthermore, our results for young children and adolescents are consistent, although families with young children are much more likely to have childcare costs than are families with adolescent children. Limited information was collected in NHANES about family expenses, which prevents our use of an expenditure-based poverty threshold, although an expenditure-adjusted poverty index is theoretically more appropriate than an income-based poverty threshold (6).

To address another concern about cash income in contrast to comprehensive income—ie, income that includes noncash public assistance—we carried out an additional sensitivity analysis after control for food stamp program participation among NHANES III subjects, but we found no material difference in the results. We did not include other variables for public assistance, and, because of the large percentage of missing information, we also did not conduct a similar analysis for other waves of NHANES data.

Therefore, although we agree that PIR has its limitations, we think these limitations should not be overinterpreted and that they are unlikely to change our main conclusions if a better indicator is available and can be used in our analysis of the NHANES data. PIR has been widely used in the United States to measure poverty as well as SES, including in many government aid programs, which either use it directly or use a simplified version of the federal poverty threshold, such as the US Department of Health and Human Services' Poverty Guidelines (7). PIR is one of the best available indicators of SES for studies based on nationally representative data, such as ours. Nevertheless, the concerns raised by Bishop do suggest that further research is needed to help fully understand the complex relations among SES, ethnicity, and obesity and the time trends.

ACKNOWLEDGMENTS

Neither of the authors had a personal or financial conflict of interest.

REFERENCES