Methods. Published observational models of the relationship of income to health were applied to predict improvements in health outcomes associated with proposed wage increases in San Francisco.

Results. With adoption of a living wage of $11.00 per hour, we predict decreases in premature death from all causes for adults aged 24 to 44 years working full-time in families whose current annual income is $20 000 (for men, relative hazard [RH] = 0.94, 95% confidence interval [CI] = 0.92, 0.97; for women, RH = 0.96, 95% CI = 0.95, 0.98). Improvements in subjectively rated health and reductions in the number of days sick in bed, in limitations of work and activities of daily living, and in depressive symptoms were also predicted, as were increases in daily alcohol consumption. For the offspring of full-time workers currently earning $20 000, a living wage predicts an increase of 0.25 years (95% CI = 0.20, 0.30) of completed education, increased odds of completing high school (odds ratio = 1.34, 95% CI = 1.20, 1.49), and a reduced risk of early childbirth (RH = 0.78, 95% CI = 0.69, 0.86).

Conclusions. A living wage in San Francisco is associated with substantial health improvement.

San Francisco's legislative board recently considered adopting a living wage of $11 per hour for workers of the city's contractors and property leaseholders. We estimated the magnitude of the anticipated health improvement associated with this legislation.

Because the San Francisco State University analysis did not directly assess the social or economic characteristics of the affected workers, we used 3 years of Bureau of Labor Statistics data for the San Francisco Bay area (1997–1999 Annual March Current Population Survey) to characterize workers aged 18 to 64 years who earned $5.75 to $11 per hour and currently worked in occupational and industry categories likely to be affected by the city ordinance. We adjusted income data to current dollars by using the urban consumer price index. Estimated proportions were pooled across the 3 survey years, and standard errors were calculated by methods supplied by the Bureau of Labor Statistics.

Estimates were based on peer-reviewed published studies of income's effect on health. Health outcomes of interest were premature mortality, preventable hospitalizations, and emergency room visits. We identified relevant literature on health outcomes by using Melvyn Medline (available at: http://www.library.ucsf.edu/db/medline/medframe) and by searching for English language articles that matched the subjectheading search terms “income” and “United States” (and “mortality,” “hospitalization,” or “health status indicators”) and that were published between 1990 and 1998. A priori, we developed the following 6 criteria for study inclusion: (1) subjects representative of the US general population; (2) income measured at the individual, family, or household level; (3) longitudinal design; (4) statistical adjustment for age and sex; (5) year of income ascertainment provided; and (6) income applied as a continuous variable. When several analytic models were used in a single study, we selected those models that assessed nonlinear effects of income and adjusted for other correlates of social position, such as education.

We identified 8 general-population studies of income's effects on all-cause mortality. All of these studies observed an inverse association between income and premature mortality. Four of the prospective national studies categorized income.17^–20 Two analyses were cross-sectional,21^,22 and one used a ratio of income to the poverty level as the independent variable and limited the analyses to Whites and African Americans.23 Only one study of income and mortality, a reanalysis of the Current Population Survey data, met all 6 of our criteria.24 The investigators stratified the analysis by 3 age categories and by sex and additionally adjusted for age, household size, education, and marital status. The model that used a logarithmic transformation of income resulted in the best fit to the risk of mortality. One nationally representative study of income and hospital utilization was identified; however, income was assessed at the zip code level, and this predictor was not available in our analysis.

We identified 4 studies of the relationship between individual income and health status indicators in representative US populations.13^,25^–27 All 4 studies were crosssectional; however, one study, by Ettner,27 used a 2-stage instrumental variable approach that allowed assessment of temporal relationships, so we included it in our analysis. The Ettner study assessed several health status indicators by using 3 data sets: the 1987 National Survey of Families and Households, the 1986–1987 panels of the Survey of Income and Program Participation, and the 1988 National Health Interview Survey. Outcomes were modeled as a function of log-transformed income, and the analyses were adjusted for sex, household size, marital status, race/ethnicity, age, education, and metropolitan area of residence. The analysis demonstrated a statistically significant exogenous relationship between income and 3 continuous health outcomes—the Center for Epidemiologic Studies scale of depressive symptoms, the number of days sick in bed in the past 4 months, and average daily alcohol consumption—as well as 3 discrete outcomes—self-rated health, work limitations, and limitations in activities of daily living.

We were also interested in the relationship between income and childhood development because of the importance of child development to lifelong social position and because of its potential intervening role in the relationship of income to health. We estimated the effect of increased wages on educational attainment and on early childbearing out of marriage by using an analysis from the Panel Study of Income Dynamics.28 We selected this study because it illustrated the contribution of family income to childhood educational achievement and met all of our a priori inclusion criteria. For our analysis, we used the coefficients derived from models that used a log transformation of income and that adjusted for race/ethnicity, sex, number of siblings, family structure, and maternal age, schooling, and employment.

The San Francisco State University economic analysis estimated that 42 118 full-time and part-time earners working in 4 economic sectors would be affected by the proposed $11-per-hour living wage.16 Estimated annual income gains varied by sector but averaged (in current dollars) $2668 for affected part-time workers and $4822 for full-time workers.

Table 1 describes selected characteristics of currently employed workers in the San Francisco Bay area aged 18 to 64 who worked at least 26 weeks a year as well as characteristics of those whose wages, industries, and occupations were most similar to those affected by the living wage. Of those affected by the living wage, 32.1% (90% confidence interval [CI] = 27.0, 37.1) were members of families with annual incomes less than $25 000. Compared with all current workers, workers targeted by the ordinance were more likely to be female, young, less educated, unmarried, without children, and working part-time.

TABLE 1— Selected Characteristics of Workers: San Francisco Bay Regiona, California, 1997–1999

Wage gains predicted mortality risk reductions and improvements in health status for both men and women and for both part-time and full-time workers. The average magnitudes of these benefits for adult workers aged 24 to 44 with a current family income of $20 000 are presented in Table 2 .

TABLE 2— Estimated Health and Educational Effects on Workers and Their Children Resulting From Adoption of a Living Wage for Families With Incomes of $20 000: San Francisco Bay Region, California, 1997–1999

The estimated reduction in mortality risk (relative hazard) for a full-time worker decreases with increasing current income, from 0.93 (95% CI = 0.90, 0.96) for men and 0.95 (95% CI = 0.93, 0.97) for women with a family annual income of $15 000 to 0.98 (95% CI = 0.977, 0.990) for men and 0.99 (95% CI = 0.985, 0.994) for women with a family income of $75 000 (Figure 1 ).

FIGURE 1— Estimated mortality risk reduction among full-time workers aged 24 to 44 years benefiting from the proposed San Francisco, Calif, living wage ordinance.

The number of days sick in bed, depressive symptoms, the risks of limitations in work or activities of daily living, and being in the poorest subjective health would all be expected to be modestly reduced for full-time workers with current family incomes of $20 000; however, daily alcohol consumption would modestly increase (Table 2 ).

For the children of workers benefiting from a living wage, the chances of completing high school would increase (Figure 2 ), as would the number of years of completed education. For girls, the risk of childbirth outside of marriage would be expected to fall.

FIGURE 2— Estimated change in the likelihood of high school graduation among children from birth to 15 years of age in families with full-time workers benefiting from the proposed San Francisco, Calif, living wage ordinance.

Estimating the magnitude of societal benefits resulting from a living wage is crucial because of the sizable costs of implementing this policy. Policymakers must be able to weigh the relative benefits and costs of a living wage compared with alternative means of achieving similar benefits.

Our analysis demonstrates that a modest gain in income resulting from a living wage would be associated with substantial health benefits. In addition, the educational attainment of workers' children would be improved and the risk of premarital childbirth among offspring would be lower with these modest income gains. Although our analysis predicted an increase in alcohol consumption, which may negatively affect health, the higher consumption of alcohol predicted by the applied study was attributed to a greater prevalence of drinking among wealthier persons.27

The major limitation of our analysis is the assumption of both a causal and a dynamic relationship between income and health. Since all available studies of the influence of income on health are observational, the apparent association could be due to confounding. Although all of the studies we applied adjusted for age, sex, race/ethnicity, education, and marital status, other unmeasured individual factors may explain the relationship between income and health. We were not able to account for neighborhood poverty, institutional racism, and inequalities in regional income distributions, which may also influence health outcomes independently of individual income.8^,11^,29^–33

Reverse causality (i.e., poor health leads to poverty) is commonly raised as an alternative explanation of the association between SES and health. However, the evidence from prospective studies and the evidence for relationships between education and health and between spousal SES and health refute this hypothesis.6 As childhood development is unlikely to influence parental income, reverse causality should not be an issue for these outcomes. Recent experience with welfare reform also provides compelling experimental evidence for the causal effect of income supplementation on childhood educational performance.34

Even if a causal relationship between income and health exists, we cannot be certain that an increase in income during adulthood will result in a prospective change in adult health. SES in childhood has been shown to predict health status in adult life, indicating that socioeconomic influences may be cumulative, have latent effects, or set an individual on a particular health trajectory.9^,35^–38 However, longitudinal studies have demonstrated higher mortality rates among individuals in the middle income range whose incomes drop by more than 50%.19 Also, significant effects of changes in family income on early childhood IQ and young adult achievement within families have been demonstrated.28^,39

The application of observational studies in this policy analysis was constrained by the way the study data were reported and analyzed. While many of the reviewed mortality studies were prospective and statistically adjusted for potential confounders, few used continuous measures of income. For studies to be useful for estimating the health benefits accruing from modest income gains, researchers should retain income as a continuous measure and model nonlinear effects.

Our analysis was not intended to capture all of the possible economic effects, and their implications for health, of a living wage ordinance. Secondary economic benefits of a living wage would be “wage push” (resulting in increasing wages for persons just above a living wage), “wage ripples” (increases in prevailing wages for persons doing similar work on noncity contracts), and local “multiplier” effects (due to the workforce spending additional income in the local economy). A potential negative effect of the living wage would be displacement of workers on city contracts due to competition from higher-paid or higher-skilled workers. Over the short term, the program would not be expected to result in displacement. However, even if displacements occurred, the ordinance would still increase the number of jobs in the community that paid a living wage.

Our study demonstrates that a more egalitarian distribution of income may have long-term positive effects on individual and community health. However, attempts to modify the distribution of wealth are likely to face significant social, scientific, and economic challenges.7^,40^–43

Jennifer Mann and Randy Reiter shared in developing the original concept for this project. Susan Ettner provided invaluable assistance in applying the data from her original research.