Abstract
Objectives. We assessed whether it would be feasible to replace the standard measure of net worth with simpler measures of wealth in population-based studies examining associations between wealth and health.
Methods. We used data from the 2004 Survey of Consumer Finances (respondents aged 25–64 years) and the 2004 Health and Retirement Survey (respondents aged 50 years or older) to construct logistic regression models relating wealth to health status and smoking. For our wealth measure, we used the standard measure of net worth as well as 9 simpler measures of wealth, and we compared results among the 10 models.
Results. In both data sets and for both health indicators, models using simpler wealth measures generated conclusions about the association between wealth and health that were similar to the conclusions generated by models using net worth. The magnitude and significance of the odds ratios were similar for the covariates in multivariate models, and the model-fit statistics for models using these simpler measures were similar to those for models using net worth.
Conclusions. Our findings suggest that simpler measures of wealth may be acceptable in population-based studies of health.
In health research, the term “wealth” refers to total financial resources amassed over a lifetime, as opposed to “income,” which refers to the capital obtained during a specified period of time (e.g., annual earnings in dollars).1 Wealth may buffer the effects of temporary low income, as in the event of illness or unemployment. Compared with income, wealth may better reflect long-term family resources and hence resources available across an individual's lifetime.2 Wealth may be particularly important for the health of the elderly, whose incomes typically drop dramatically following retirement,3 and for racial/ethnic disparities in health, because differences in wealth by racial/ethnic group are far greater than are the corresponding differences in income.2
A systematic review of the literature has found that greater wealth is associated with better health, even after adjustment for other socioeconomic factors, such as income and educational attainment.1,4–11 Moreover, these findings of positive correlations between wealth and health were most consistent when studies used detailed wealth measures based on multiple questions on assets (e.g., savings, home, retirement) and debts (e.g., mortgage, loans) instead of single questions on wealth (e.g., home ownership). The review also found that racial/ethnic disparities in health generally decreased after adjustment for wealth.
Despite the conceptual and empirical rationales for including wealth in research on health, population-based health surveys (such as the National Health Interview Survey) and vital statistics data generally have either poor measures of wealth or none at all. This deficiency is not surprising, given the difficulty of collecting wealth data. The topic of wealth is considered to be sensitive, collection of reliable information is laborious, and the values of assets and debts vary over time and may require professional appraisal. Conversely, population-based surveys with detailed wealth measures typically contain little information on health. Thus, current population-based data sources present significant barriers to studies of the relationships between wealth and health.
Standard wealth measures are based on multiple detailed questions; for example, the Survey of Consumer Finances (SCF) assesses 25 different classes of assets and debts that measure net worth. Simpler approaches to wealth measurement could benefit population-based health research by making it more feasible to include such measures in important public health surveys. Building on earlier work on the measurement of socioeconomic status and position,1,2,12,13 we assessed whether simplified measures of wealth could be used in health research to reasonably approximate standard wealth measures. We used population-based data from the SCF and the Health and Retirement Survey (HRS) to assess (1) correlations between 9 simpler measures of wealth and the standard wealth measure net worth, which requires multiple detailed questions on assets and debts; and (2) the results of models relating wealth to self-reported health status and cigarette smoking, with comparisons between models that measured wealth as net worth and models using simpler measures of wealth.
We used the 2004 SCF and the 2004 HRS in our analyses because these surveys provide detailed measures of net worth as well as indicators of health. The SCF, sponsored by the US Federal Reserve Board, is intended to provide a detailed picture of the finances of noninstitutionalized families in the United States. A multistage area-probability sample is surveyed along with a supplemental sample of primarily wealthy families. In 2004, the response rate for the area-probability sample was 70% (n = 3007), and the response rate for the supplemental sample was 30% (n = 1515).
The HRS is a nationally representative data set with an overall response rate of 86% in 2004 (n = 20 129). The HRS, sponsored by the National Institute on Aging and the Social Security Administration, is intended to provide a detailed picture of health, financial, and other characteristics of the aging population in the United States. For both data sets, imputation techniques were used for missing data, and survey weights were used to adjust for sampling probabilities. SCF researchers imputed all missing data by means of a multiple imputation procedure yielding 5 values for each missing value, to approximate the distribution of the missing data. For missing data in the HRS, we relied on the RAND HRS imputations, which first imputed ownership of a particular asset or debt (using logistic regression models), then brackets (using ordered logit models), then exact amounts (using either a nearest neighbor approach or a Tobit approach; http://www.rand.org/labor/aging/dataprod.html#randhrs). Further details about sample design and methodology are available for the SCF (http://www.federalreserve.gov/pubs/oss/oss2/method.html) and the HRS (http://hrsonline.isr.umich.edu/index.php?p=sdesign).
Head-of-household respondents aged 25 to 64 years (SCF) or 50 years or older (HRS) who identified as (1) Black, non-Hispanic; (2) Hispanic; or (3) White, non-Hispanic were included in the analytic samples (n = 3310 for the SCF and n = 11 847 for the HRS). For households with a couple, the SCF defined the head of household as the male in mixed-gender households or the older individual in same-gender households. The HRS analytic sample did not include persons residing in nursing homes or surveys without a financial respondent—the person designated to answer household-level financial questions.
Definitions for each asset and debt included in each survey are listed in Appendix A (available as a supplement to the online version of this article at http://www.ajph.org). The SCF contains more assets and debts in separate categories than the HRS does, but both surveys are comprehensive. For example, the SCF asks about checking accounts, savings accounts, and money market accounts separately, but those 3 categories are combined in the HRS. Typically, researchers measure wealth as net worth; following this approach, we calculated net worth by adding the dollar value of all assets minus the value of all debts. We also calculated 9 simplified measures of net worth. The first 5 of these are based on dollar values, as follows:
Assets: the additive value of all assets; Prevalent assets/debts: items that were reported to be owned by a large percentage of the total sample (at least 25%). The SCF counted checking account, savings account, retirement funds, vehicles, and primary residence as assets, and counted mortgage, credit card balance, and installment loans as debts. The HRS asked about a similar list: checking, savings, and money market account; mutual funds and stocks; retirement funds; vehicles; and primary residence as assets and mortgage and other debt/credit card balance as debts. Adding the value of assets minus the value of debts for this subset is the “prevalent assets/debts” measure of wealth; Highest-proportion assets/debts: Although a large proportion of individuals may own a particular asset/debt, its value may be high or low with respect to a person's overall net worth. We therefore defined highest-proportion items as the assets that, on average, accounted for more than 10% of an individual's overall assets, and the debts that accounted for more than 10% of an individual's debt. In the SCF, these assets were vehicles, retirement funds, and primary residence, and the debts were mortgage, credit card balance, and installment loans. For the HRS, the highest-proportion assets were checking, savings, and money market accounts; vehicles; and primary residence, and the highest-proportion debts were mortgage and other debt/credit card balance. Adding the value of assets minus the value of debts for this subset is the “highest-proportion assets/debts” measure of wealth; Prevalent assets: the value of the most prevalent assets; Highest-proportion assets: the value of the highest-proportion assets; Respondents may be more willing to indicate ownership of an asset or debt rather than the actual dollar amounts, which may fluctuate over time. We therefore created an additional 4 indices in parallel with 4 of the dollar-based measures. If a respondent owned an asset, it was scored +1; if a respondent owned a debt, it was scored −1; if a respondent did not own an asset/debt, it was scored 0. Prevalent assets/debts index: the sum of these items (range −3 to 5 in the SCF; range −2 to 5 in the HRS); Highest-proportion assets/debts index: the sum of these items (range −3 to 3 in the SCF; range −2 to 3 in the HRS); Prevalent assets index: an index of the most prevalent assets (range 0–5); Highest-proportion assets index: a scale of the highest-proportion assets (range 0–3).
We examined 2 dependent variables in the analyses: self-reported health, which corresponds closely with objective clinical assessments of an individual's overall health14; and whether the respondent is a current smoker, given the prominent standing of smoking as a cause of disease and mortality. Health status was measured on a 4-point (SCF) or 5-point (HRS) Likert scale, dichotomized as fair or poor health versus better health. Current smoking was measured similarly in the 2 survey instruments, with the questions “Do you currently smoke?” (SCF) and “Do you smoke cigarettes now?” (HRS), with “yes” and “no” as responses.
We included age, gender, race/ethnicity (non-Hispanic Black, Hispanic, or non-Hispanic White), marital status (married or partnered, previously married, or never married), and family size as covariates in the analyses. In the HRS, census region was also included to account for regional variations in cost of living; geographic identifiers were not available in the public-use SCF data set. Educational attainment was classified into 4 categories: less than high school, high school graduate or general equivalency diploma, some college, or college graduate and above. Annual household income from all sources was determined on a pretax basis and was log-transformed.
We calculated descriptive statistics, including sociodemographic characteristics of the samples and the prevalence and median values of assets and debts. We then examined correlations between net worth and each of the simplified wealth measures. We estimated a series of logistic regression models for each dependent variable and wealth measure. For these models, each wealth measure was categorized into quartiles (for measures based on dollar values) or 4 roughly equal groups (for measures based on summary indices). The base model included only a single wealth measure. Next, the demographic model added age, age squared, gender, race/ethnicity, marital status, family size, and region (HRS only). The full model added education and income to the demographic model. To compare model fit across the full models, each with a different wealth measure, we examined the percentage differences in the Somers’ D, Akaike information criterion (AIC), and Bayesian information criterion (BIC) statistics, comparing the net-worth model to those with a simplified wealth measure.
Table 1 presents weighted prevalences and median values for each asset and debt. In both data sets, the large majority of respondents owned a checking account (in the HRS, this measure was combined with ownership of a savings or money market account), a vehicle, and a primary residence. More respondents had retirement funds in the younger SCF population than in the older HRS population. Across all assets, the highest value was contained in the respondents’ primary residence ($165 000 in the SCF, $150 000 in the HRS). Mortgage on a primary residence was the most common and largest debt for both data sets, followed by other loans and credit card balances in the SCF sample and other debt or credit card balances in the HRS sample. Debt accumulation was higher in the younger population (SCF) than in the older population (HRS).
TABLE 1 Prevalences and Median Values of Assets and Debts: Survey of Consumer Finances and Health and Retirement Survey, United States, 2004
| Survey of Consumer Finances | Health and Retirement Survey | ||||
| Prevalence, % | Median Value, $a | Prevalence, % | Median Value, $a | ||
| Assets | |||||
| Checking accountb | 83 | 1500 | Checking/savings/money market accountsbc | 87 | 7500 |
| Savings accountb | 49 | 3000 | Certificates of deposit/savings bonds | 20 | 16 000 |
| Money market account | 21 | 7000 | Mutual funds/stocksb | 31 | 60 000 |
| Call account | 2 | 13 000 | Bonds | 7 | 39 000 |
| Certificates of deposit/ | 10 | 12 000 | Retirement fundsb | 39 | 49 000 |
| Savings bonds | 20 | 800 | Vehiclesbc | 84 | 10 000 |
| Mutual funds | 15 | 35 000 | Primary residencebc | 78 | 150 000 |
| Stocks | 21 | 12 000 | Other residential real estate | 13 | 70 000 |
| Bonds | 2 | 30 000 | Nonresidential real estate | 15 | 81 000 |
| Retirement fundsbc | 56 | 35 000 | Business | 11 | 100 000 |
| Life insurance | 23 | 6000 | Other nonfinancial assets/other savings | 17 | 20 000 |
| Other managed accounts | 6 | 36 000 | |||
| Other financial assets | 11 | 3800 | |||
| Vehiclesbc | 88 | 16 000 | |||
| Primary residencebc | 69 | 165 000 | |||
| Other residential real estate | 13 | 94 500 | |||
| Nonresidential real estate | 8 | 55 000 | |||
| Business | 14 | 102 000 | |||
| Other nonfinancial assets | 8 | 15 000 | |||
| Debts | |||||
| Mortgage, primary residencebc | 57 | 99 000 | Mortgage, primary residencebc | 33 | 73 000 |
| Other residential property debt | 5 | 87 000 | Mortgage, secondary residence | 3 | 59 000 |
| Other line of credit | 2 | 3000 | Other residential property debt | 11 | 22 000 |
| Credit card balancebc | 52 | 2400 | Other debt/credit card balancebc | 31 | 5000 |
| Other debt | 9 | 4000 | |||
| Other installment loansbc | 53 | 12 000 | |||
a Among those who owned the asset/debt.
b Prevalent asset/debt: an asset/debt owned by at least 25% of the sample.
c Highest-proportion asset/debt: an asset/debt that, on average, accounted for at least 10% of an individual's overall assets/debts.
Table 2 presents descriptive characteristics of the sample populations, including median values for the various wealth measures. The low proportion of women in the SCF reflects the fact that the household respondent is specified as the man in a household that includes both a woman and a man. Although the overall median value of assets was only somewhat higher in the HRS, median debt among the HRS respondents was much lower, resulting in the net worth among the older HRS population being nearly 2 times higher than in the SCF sample. Contributing to this pattern, homeownership was higher in the HRS sample than in the SCF sample.
TABLE 2 Demographic, Socioeconomic, Wealth, and Health Characteristics of Samples: Survey of Consumer Finances and Health and Retirement Survey, United States, 2004
| SCF (n = 3310) | HRS (n = 11 847) | |
| Age, y, % | ||
| 25–49 | 66 | N/A |
| 50–64 | 34 | 54 |
| 65–74 | N/A | 23 |
| 75+ | N/A | 23 |
| Gender, % | ||
| Women | 25 | 47 |
| Men | 75 | 53 |
| Race/ethnicity, % | ||
| Black, non-Hispanic | 15 | 11 |
| Hispanic | 11 | 7 |
| White, non-Hispanic | 74 | 82 |
| Marital status, % | ||
| Never married | 19 | 6 |
| Separated/divorced/widowed | 27 | 43 |
| Married/living as married | 54 | 51 |
| Family size, median (range) | 2 (1–10) | 2 (1–15) |
| Region, % | ||
| Northeast | N/A | 18 |
| Midwest | N/A | 26 |
| South | N/A | 37 |
| West | N/A | 19 |
| Educational attainment, % | ||
| < High school | 11 | 17 |
| High school graduate/GED | 30 | 35 |
| Some college | 19 | 24 |
| College graduate | 40 | 24 |
| Annual income, $1000, median (range) | 50 (0–105 070) | 36 (0–3532) |
| Wealth, $1000, median (range) | ||
| Assets | 173 (0–744 677) | 201 (0–77 225) |
| Debts | 47 (0–43 117) | 1 (0–2400) |
| Net worth | 86 (–455–714 677) | 158 (–2246–77 225) |
| Prevalent assets/debtsa | 65 (–15 294–75 832) | 130 (–2280–77 175) |
| Highest-proportion assets/debtsb | 58 (–59 298–7563) | 100 (–2388–12 185) |
| Prevalent assetsc | 149 (0–82 142) | 169 (0–77 175) |
| Highest-proportion assetsd | 143 (0–82 028) | 133 (0–12 185) |
| Prevalent assets/debts indexe | 2 (–2–5) | 3 (–1–5) |
| Highest-proportion assets/debts indexf | 0 (–2–3) | 2 (–1–3) |
| Prevalent assets indexg | 4 (0–5) | 3 (0–5) |
| Highest-proportion assets indexh | 2 (0–3) | 3 (0–3) |
| Health, % | ||
| Fair/poor health status | 21 | 27 |
| Current smoker | 26 | 17 |
Note. GED = general equivalency diploma; HRS = Health and Retirement Survey; N/A = not applicable; SCF = Survey of Consumer Finances.
a SCF: checking account + savings account + retirement funds + vehicles + primary residence – mortgage – credit card balance – installment loans. HRS: checking/savings/money market account + mutual funds/stocks + retirement funds + vehicles + primary residence – mortgage – other debt/credit card balance.
b SCF: retirement funds + vehicles + primary residence – mortgage – credit card balance – installment loans. HRS: checking/savings/money market account + vehicles + primary residence – mortgage – other debt/credit card balance.
c SCF: checking account + savings account + retirement funds + vehicles + primary residence. HRS: checking/savings/money market account + mutual funds/stocks + retirement funds + vehicles + primary residence.
d SCF: retirement funds + vehicles + primary residence. HRS: checking/savings/money market account + vehicles + primary residence.
e Summary index of prevalent assets (1) and debts (–1); nonownership of an asset or debt contributes 0 to sum.
f Summary index of highest-proportion assets (1) and debts (–1); nonownership of an asset contributes 0 to sum.
g Summary index of prevalent assets (1); nonownership of an asset or debt contributes 0 to sum.
h Summary index of highest-proportion assets (1); nonownership of an asset contributes 0 to sum.
Correlations between net worth and the other 9 wealth measures were generally moderate (not shown, available upon request), ranging from 0.43 to 0.67 in the SCF, except for the correlation with total assets (nearly 1.00), and ranging from 0.47 to 0.73 in the HRS, except for the correlations with prevalent assets/debts (0.93), assets (nearly 1.00), and prevalent assets (0.93). The very high correlations between the assets-only measures and net worth suggest that it may not be necessary to include debts in a measure of wealth.
Table 3 presents the results of the logistic regression models for the relationships of net worth with fair/poor health status and current smoker. In the full SCF model, respondents in the lowest quartile of wealth had odds of fair/poor health that were nearly 5 times (odds ratio [OR] = 4.98) higher than those for respondents in the highest quartile, and a clear gradient was observed. Differences between Hispanics and non-Hispanic Whites in the likelihood of having fair/poor health were no longer significant in the full model. Differences between non-Hispanic Blacks and non-Hispanic Whites in the likelihood of having fair/poor health were not significantly different in either model. Lower education and income were significant predictors of having fair/poor health in the full model. We found similar results for the relationship between net worth and fair/poor health for the HRS data, although the magnitude of the association was smaller than that found for the SCF data. Racial/ethnic disparities persisted in the full HRS model, with non-Hispanic Blacks and Hispanics having higher odds of fair/poor health status than did non-Hispanic Whites.
TABLE 3 Crude and Adjusted Associations of Net Worth With Fair/Poor Health Status and Current Smoking: Survey of Consumer Finances and Health and Retirement Survey, United States, 2004
| Crude Model, OR (95% CI) | Demographic Model, OR (95% CI) | Full Model, OR (95% CI) | ||||
| SCF | HRS | SCF OR (95% CI) | HRS OR (95% CI) | SCF OR (95% CI) | HRS OR (95% CI) | |
| Fair/poor health | ||||||
| Age | 0.95 (0.88, 1.03) | 1.08** (1.03, 1.14) | 0.94 (0.87, 1.02) | 1.04 (0.99, 1.10) | ||
| Age squareda | 1.00** (1.00, 1.00) | 1.00* (1.00, 1.00) | 1.00 (1.00, 1.00) | 1.00 (1.00, 1.00) | ||
| Men | 1.12 (0.85, 1.46) | 0.97 (0.87, 1.09) | 1.11 (0.85, 1.47) | 0.94 (0.84, 1.05) | ||
| Black, non-Hispanic | 0.95 (0.72, 1.26) | 1.33*** (1.15, 1.54) | 0.85 (0.64, 1.15) | 1.16* (1.00, 1.35) | ||
| Hispanic | 1.90*** (1.41, 2.57) | 2.30*** (1.93, 2.75) | 1.37 (0.99, 1.88) | 1.59** (1.32, 1.92) | ||
| Previously married | 1.37* (1.02, 1.84) | 1.20** (1.06, 1.37) | 1.13 (0.84, 1.54) | 0.99 (0.86, 1.14) | ||
| Never married | 1.35 (0.97, 1.88) | 1.08 (0.84, 1.38) | 1.23 (0.88, 1.73) | 0.87 (0.68, 1.12) | ||
| Family size | 1.00 (0.92, 1.08) | 1.08** (1.03, 1.14) | 0.98 (0.90, 1.06) | 1.02 (0.97, 1.07) | ||
| Northeast | N/A | 1.03 (0.88, 1.20) | N/A | 0.99 (0.85, 1.17) | ||
| Midwest | N/A | 1.20* (1.04, 1.39) | N/A | 1.16 (0.99, 1.34) | ||
| South | N/A | 1.08 (0.91, 1.28) | N/A | 1.13 (0.95, 1.35) | ||
| < High school | 2.91*** (2.04, 4.15) | 2.96*** (2.46, 3.58) | ||||
| High school/GED | 2.06*** (1.58, 2.69) | 1.67*** (1.41, 1.98) | ||||
| Some college | 1.53** (1.13, 2.07) | 1.52*** (1.27, 1.81) | ||||
| Income (log) | 0.80*** (0.74, 0.87) | 0.82*** (0.77, 0.86) | ||||
| Net worth | ||||||
| Quartile 1 (lowest) | 7.53*** (5.83, 9.74) | 6.12*** (5.30, 7.08) | 12.27*** (8.74, 17.22) | 5.21*** (4.43, 6.13) | 4.98*** (3.42, 7.24) | 3.13*** (2.62, 3.73) |
| Quartile 2 | 4.19*** (3.19, 5.51) | 2.93*** (2.53, 3.39) | 6.38*** (4.63, 8.80) | 2.62*** (2.25, 3.06) | 3.23*** (2.29, 4.57) | 1.85*** (1.58, 2.17) |
| Quartile 3 | 2.29*** (1.68, 3.11) | 1.54*** (1.32, 1.80) | 2.88*** (2.08, 4.00) | 1.46*** (1.25, 1.71) | 1.64** (1.16, 2.31) | 1.20* (1.02, 1.41) |
| Current smoker | ||||||
| Age | 1.09* (1.01, 1.17) | 1.28*** (1.17, 1.40) | 1.08* (1.01, 1.16) | 1.24*** (1.13, 1.34) | ||
| Age squareda | 1.00* (1.00, 1.00) | 1.00*** (1.00, 1.00) | 1.00* (1.00, 1.00) | 1.00*** (1.00, 1.00) | ||
| Men | 1.79*** (1.39, 2.30) | 0.81** (0.71, 0.93) | 1.72*** (1.33, 2.23) | 0.77*** (0.67, 0.88) | ||
| Black, non-Hispanic | 0.71* (0.54, 0.93) | 0.84* (0.70, 1.00) | 0.65** (0.49, 0.85) | 0.77** (0.64, 0.92) | ||
| Hispanic | 0.53*** (0.39, 0.73) | 0.61*** (0.48, 0.79) | 0.39*** (0.28, 0.54) | 0.48*** (0.37, 0.62) | ||
| Previously married | 1.85*** (1.42, 2.42) | 1.75*** (1.49, 2.05) | 1.63*** (1.25, 2.15) | 1.65*** (1.40, 1.93) | ||
| Never married | 1.70*** (1.26, 2.28) | 1.08 (0.79, 1.47) | 1.68*** (1.25, 2.27) | 1.04 (0.76, 1.43) | ||
| Family size | 0.98 (0.91, 1.06) | 1.01 (0.95, 1.07) | 0.95 (0.88, 1.03) | 1.00 (0.95, 1.06) | ||
| Northeast | N/A | 0.93 (0.76, 1.12) | N/A | 0.90 (0.74, 1.10) | ||
| Midwest | N/A | 0.96 (0.81, 1.15) | N/A | 0.95 (0.79, 1.13) | ||
| South | N/A | 0.78* (0.63, 0.97) | N/A | 0.82 (0.66, 1.02) | ||
| < High school | 3.41*** (2.43, 4.78) | 2.81*** (2.21, 3.59) | ||||
| High school/GED | 3.06*** (2.41, 3.87) | 2.17*** (1.76, 2.68) | ||||
| Some college | 2.03*** (1.55, 2.65) | 1.90*** (1.53, 1.36) | ||||
| Income (log) | 0.94 (0.87, 1.00) | 0.94** (0.90, 0.98) | ||||
| Net worth | ||||||
| Quartile 1 (lowest) | 5.05*** (4.01, 6.36) | 4.21*** (3.52, 5.03) | 5.86*** (4.36, 7.88) | 3.78*** (3.08, 4.63) | 2.85*** (2.02, 4.03) | 2.58*** (2.07, 3.21) |
| Quartile 2 | 3.81*** (2.97, 4.90) | 2.33*** (1.92, 2.82) | 3.99*** (2.99, 5.32) | 2.21*** (1.83, 2.73) | 2.26*** (1.64, 3.11) | 1.67*** (1.36, 2.06) |
| Quartile 3 | 2.29*** (1.76, 2.99) | 1.66*** (1.37, 2.03) | 2.33*** (1.77, 3.07) | 1.63*** (1.33, 2.00) | 1.47* (1.09, 1.99) | 1.40*** (1.14, 1.72) |
Note. CI = confidence interval; GED = general equivalency diploma; HRS = Health and Retirement Survey; N/A = not applicable; OR = odds ratio; SCF = Survey of Consumer Finances. The demographic model is adjusted for age, age squared, gender, race/ethnicity, marital status, family size, and region (HRS only). The full model is also adjusted for educational attainment and income. Reference groups are women; White, non-Hispanic; married/living as married; West; college graduate; and quartile 4 (highest).
a Significant values indicate that the confidence interval does not include 1.00, which is apparent with > 2 decimal places.
*P < .05; **P < .01; ***P < .001.
In the models predicting current smoking, all 3 quartiles of net worth were statistically significant, and results reflected a gradient pattern in both data sets. Overall, these models revealed remarkably similar results, with the exception of gender (men had higher odds of smoking than did women in the SCF but lower odds of smoking than did women in the HRS), marital status (persons who were never married had higher odds of smoking than did married persons in the SCF but not in the HRS), and income (which was significant in the HRS but not in the SCF).
Table 4 presents the associations between health status, smoking status, and the 10 wealth measures from the full models. Similar to the models already described, these full models were adjusted for age, gender, race/ethnicity, marital status, family size, region (for the HRS), education, and income. The first row reflects the same ORs and confidence intervals (CIs) for net worth that were shown in the full models in Table 3. Each subsequent row substitutes a different wealth measure for net worth. For the SCF models predicting fair/poor health and current smoking status, all 3 categories of wealth are statistically significant in most models. In general, the wealth measures based on dollar values had roughly similar ORs compared with the net-worth model. For example, those in the lowest quartile of wealth—measured as either net worth or any of the dollar-value wealth measures—were approximately 3 to 6 times as likely as were those in the highest wealth quartile to report fair/poor health for both samples. Similarly, those in the lowest wealth quartile using any of the dollar-value measures were approximately 2 to 3 times as likely as were those in the highest wealth quartile to report current smoking in both samples.
By contrast, the measures based on summary indices tended to have lower ORs than did the net-worth SCF model (although CIs overlapped in some cases). However, the magnitude and significance of wealth–health associations appeared more consistent in HRS data, regardless of the wealth measure used. Across both samples, measures that used assets only generally appeared to have higher ORs than did measures that used assets and debts (although CIs overlapped in nearly all cases). The other covariates were generally robust in magnitude and significance across the wealth models (data available as a supplement to the online version of this article at http://www.ajph.org).
TABLE 4 Associations of Wealth Measures With Fair/Poor Health Status and Current Smoking: Survey of Consumer Finances and Health and Retirement Survey, United States, 2004
| Fair/Poor Health Status | Current Smoking | |||
| SCF OR (95% CI) | HRS OR (95% CI) | SCF OR (95% CI) | HRS OR (95% CI) | |
| Dollar-based measures | ||||
| Net worth | ||||
| Quartile 1 (lowest) | 4.98*** (3.42, 7.24) | 3.13*** (2.62, 3.73) | 2.85*** (2.02, 4.03) | 2.58*** (2.07, 3.21) |
| Quartile 2 | 3.23*** (2.29, 4.57) | 1.85*** (1.58, 2.17) | 2.26*** (1.64, 3.11) | 1.67*** (1.36, 2.06) |
| Quartile 3 | 1.64** (1.16, 2.31) | 1.20* (1.02, 1.41) | 1.47* (1.09, 1.99) | 1.40*** (1.14, 1.72) |
| Assets | ||||
| Quartile 1 (lowest) | 5.77*** (3.92, 8.49) | 3.24*** (2.70, 3.90) | 3.04*** (2.15, 4.29) | 2.80*** (2.23, 3.52) |
| Quartile 2 | 3.47*** (2.45, 4.92) | 1.96*** (1.66, 2.31) | 2.31*** (1.70, 3.14) | 1.75*** (1.41, 2.17) |
| Quartile 3 | 1.56* (1.08, 2.25) | 1.24* (1.05, 1.46) | 1.49** (1.10, 2.02) | 1.52*** (1.24, 1.87) |
| Prevalent assets/debtsa | ||||
| Quartile 1 (lowest) | 4.60*** (3.15, 6.72) | 3.16*** (2.65, 3.76) | 3.24*** (2.33, 4.51) | 2.65*** (2.13, 3.30) |
| Quartile 2 | 2.88*** (2.01, 4.14) | 1.86*** (1.59, 2.19) | 2.49*** (1.82, 3.41) | 1.66*** (1.35, 2.06) |
| Quartile 3 | 1.50* (1.05, 2.14) | 1.28** (1.09, 1.51) | 1.69*** (1.25, 2.27) | 1.60*** (1.31, 1.97) |
| Highest-proportion assets/debtsb | ||||
| Quartile 1 (lowest) | 4.26*** (2.97, 6.16) | 2.75*** (2.32, 3.25) | 2.97*** (2.13, 4.15) | 2.14*** (1.74, 2.63) |
| Quartile 2 | 2.71*** (1.90, 3.88) | 1.69*** (1.44, 1.97) | 2.45*** (1.81, 3.33) | 1.46*** (1.19, 1.78) |
| Quartile 3 | 1.47* (1.04, 2.08) | 1.18*** (1.01, 1.38) | 1.72*** (1.27, 2.32) | 1.13 (0.93, 1.39) |
| Prevalent assetsc | ||||
| Quartile 1 (lowest) | 5.86*** (3.96, 8.67) | 3.16*** (2.65, 3.78) | 3.41*** (2.39, 4.86) | 2.72*** (2.18, 3.40) |
| Quartile 2 | 3.86*** (2.68, 5.54) | 1.92*** (1.63, 2.26) | 2.40*** (1.76, 3.27) | 1.70*** (1.38, 2.12) |
| Quartile 3 | 1.57* (1.09, 2.26) | 1.33*** (1.13, 1.57) | 1.69*** (1.25, 2.29) | 1.46*** (1.19, 1.80) |
| Highest-proportion assetsd | ||||
| Quartile 1 (lowest) | 5.29*** (3.60, 7.78) | 2.80*** (2.36, 3.33) | 3.40*** (2.39, 4.84) | 2.35*** (1.90, 2.90) |
| Quartile 2 | 3.58*** (2.47, 5.19) | 1.73*** (1.47, 2.03) | 2.44*** (1.79, 3.32) | 1.59*** (1.29, 1.96) |
| Quartile 3 | 1.47* (1.00, 2.15) | 1.30*** (1.11, 1.53) | 1.80*** (1.33, 2.44) | 1.21 (0.99, 1.48) |
| Index-based measures | ||||
| Prevalent assets/debts indexe | ||||
| Low | 2.12*** (1.52, 2.95) | 2.55*** (2.16, 3.01) | 1.52** (1.11, 2.08) | 1.94*** (1.58, 2.38) |
| Low/moderate | 1.61** (1.21, 2.14) | 1.53*** (1.31, 1.80) | 1.53** (1.18, 1.98) | 1.72*** (1.41, 2.09) |
| Moderate/high | 1.31 (1.00, 1.71) | 1.35*** (1.16, 1.57) | 1.28* (1.01, 1.62) | 1.42*** (1.16, 1.74) |
| Highest-proportion assets/debts indexf | ||||
| Low | 1.89*** (1.30, 2.75) | 2.41*** (1.96, 2.97) | 1.27*** (0.91, 1.78) | 1.43** (1.12, 1.83) |
| Low/moderate | 1.61** (1.19, 2.18) | 1.56*** (1.35, 1.81) | 1.39* (1.06, 1.83) | 1.40*** (1.17, 1.68) |
| Moderate/high | 1.35 (0.97, 1.84) | 1.15* (1.01, 1.31) | 1.26 (0.96, 1.66) | 1.22* (1.03, 1.45) |
| Prevalent assets indexg | ||||
| Low | 2.49*** (1.77, 3.50) | 3.31*** (2.69, 4.08) | 1.95*** (1.44, 2.66) | 3.00*** (2.32, 3.89) |
| Low/moderate | 1.69** (1.22, 2.32) | 1.87*** (1.55, 2.27) | 1.73*** (1.31, 2.29) | 1.87*** (1.47, 2.39) |
| Moderate/high | 1.22 (0.89, 1.67) | 1.46*** (1.20, 1.77) | 1.39* (1.06, 1.82) | 1.53*** (1.20, 1.95) |
| Highest-proportion assets indexh | ||||
| Low | 3.47*** (2.27, 5.31) | 2.67*** (2.04, 3.48) | 1.80** (1.20, 2.70) | 2.06*** (1.53, 2.80) |
| Low/moderate | 2.34*** (1.73, 3.18) | 2.26*** (1.90, 2.70) | 1.78*** (1.36, 2.35) | 1.80*** (1.45, 2.24) |
| Moderate/high | 1.66** (1.28, 2.16) | 1.61*** (1.42, 1.82) | 1.44** (1.15, 1.82) | 1.52*** (1.30, 1.78) |
Note. CI = confidence interval; HRS = Health and Retirement Survey; OR = odds ratio; SCF = Survey of Consumer Finances. All models are adjusted for age, age squared, gender, race/ethnicity, marital status, family size, region (HRS only), educational attainment, and income. Reference group is quartile 4 for quartile measures or high for index measures.
a SCF: checking account + savings account + retirement funds + vehicles + primary residence – mortgage – credit card balance – installment loans. HRS: checking/savings/money market account + mutual funds/stocks + retirement funds + vehicles + primary residence – mortgage – other debt/credit card balance.
b SCF: retirement funds + vehicles + primary residence – mortgage – credit card balance – installment loans. HRS: checking/savings/money market account + vehicles + primary residence – mortgage – other debt/credit card balance.
c SCF: checking account + savings account + retirement funds + vehicles + primary residence. HRS: checking/savings/money market account + mutual funds/stocks + retirement funds + vehicles + primary residence.
d SCF: retirement funds + vehicles + primary residence. HRS: checking/savings/money market account + vehicles + primary residence.
e Summary index of prevalent assets (1) and debts (–1); nonownership of an asset or debt contributes 0 to sum.
f Summary index of highest-proportion assets (1) and debts (–1); nonownership of an asset contributes 0 to sum.
g Summary index of prevalent assets (1); nonownership of an asset or debt contributes 0 to sum.
h Summary index of highest-proportion assets (1); nonownership of an asset contributes 0 to sum.
*P < .05; **P < .01; ***P < .001.
Comparing the percentage differences in model-fit statistics (Somers’ D, AIC, BIC) from the net-worth full SCF model to the other 9 full wealth measure models yielded comparable fits for both health indicators (not shown, available upon request). To illustrate, the Somers’ D statistic was 0.570 in the net-worth model for fair/poor health status; Somers’ D statistics were all within 4% of that number for the other models, with the measures based on summary indices having higher percentage differences (3%–4%) than did the measures based on dollar values (1%–2%). The AIC and BIC statistics for fair/poor health were within 3% of the AIC and 2% of the BIC statistics in the net-worth model (AIC = 2606 and BIC = 2704 in the net-worth model). Again, the measures based on summary indices had higher percentage differences than did the measures based on dollar values. A similar pattern in fit statistics was found for smoking in SCF, with Somers’ D statistics within 4%, AIC statistics within 3%, and BIC statistics within 1.5% of those found in the net-worth model. Patterns were similar in the HRS data, with Somers’ D statistics within 4%, AIC statistics within 1.5%, and BIC statistics within 0.5% for both fair/poor health status and smoking.
Wealth has repeatedly been shown to be a strong and robust predictor of health, after controlling for both income and education.1 Many studies measure wealth by using the sum of the value in dollars of all financial assets minus debts. Multiple detailed items are required for this measure, so wealth has not been included in most health surveys because of space limitations and respondent burden. Our findings suggest that simpler measures of wealth may be useful for assessing this important variable when it is not feasible to use the standard approach.
In both samples, when we used any single simpler measure we reached conclusions about the association between wealth and health that were similar to those we reached when we used net worth. Furthermore, the magnitude and significance of the ORs estimating effects on health were similar not only for the wealth measure but also for the covariates in multivariate models. Finally, when we used any of the simpler measures, the model-fit statistics were very close to those in the net-worth model (i.e., within 4%). Findings were more consistent in the HRS than in the SCF.
One straightforward approach to simplifying wealth measures is to assess assets only, as opposed to assets and debts, given that asset-only measures produced similar point estimates and model fit to those produced by the net-worth model. Inclusion of most prevalent or highest debts in addition to assets increases the response burden and may produce a more conservative point estimate. If one were to focus on assets only, taking the SCF as an example, the simpler measure of most prevalent assets would reduce the number of asset classes from 19 to 5. Using the highest-proportion assets would reduce the classes further to only those assessing retirement funds, primary residence, and vehicles.
It may be appealing to use even simpler measures consisting of indices based on whether a respondent owns a particular asset type, but our findings suggest that using indices may underestimate the wealth effect on health, at least for nonelderly adults. However, the results using indices in the HRS data were more consistent, indicating that the simpler yes or no summary index measure might be adequate. Although home ownership (without estimated value) produced similar covariate estimates and model fit, this measure is limited because it precludes examination of a gradient effect between wealth and health. We also tested a measure of housing value that compared those in the highest quartile of housing values with nonhomeowners and those in the lowest 2 quartiles of housing values, and we did not find significant gradient effects (data not shown, available upon request).
One might have expected a stronger relationship between wealth and health in the older age group because of the accumulation of wealth over the lifetime and the commonly experienced loss of income during retirement. However, the magnitude of the association between wealth and income with respect to health was similar for the 2 age groups and tended to be higher in the SCF than it was in the HRS. This finding highlights the importance of measuring wealth when examining social disparities in health throughout the life span. Cohort effects and mortality selection at the older ages in the HRS likely play a role in the smaller gradient. Some evidence exists for the mortality-selection argument: In sensitivity analyses among the HRS respondents aged 50 to 64 years only, we found that the wealth associations for both health indicators were stronger than for all respondents aged 50 years or older, although the confidence intervals overlapped.
We recommend replicating these analyses using a range of different health indicators, within and across different population subgroups. For example, wealth is known to vary greatly according to gender15 and race/ethnicity.2,16,17 In addition, further work should examine the issue of thresholds for selecting assets/debts. The thresholds for the most prevalent assets/debts (25%) and highest-proportion assets/debts (10%) were chosen on the basis of the samples’ distributions, in an effort to decrease response burden. Different results may possibly be obtained with different cut points. However, taking the SCF as an example, the same assets and debts would be chosen even with a cut point of 50%. Alternatively, choosing 15% as the threshold would add 5 additional assets measures that would significantly increase respondent burden. Careful consideration should be given to choosing a threshold that is meaningful in relation to the data source while also minimizing respondent burden.
This study has multiple limitations. First, it is based on cross-sectional data; therefore, we are unable to determine causality in the association between wealth and health. Second, wealth is time-varying (although likely more stable than income). Because we measured wealth at only 1 moment in time, we are unable to fully examine life-course socioeconomic status. However, the application of this work would probably be most useful in cross-sectional surveillance systems. Third, we used net worth as the gold standard even though it suffers from unknown measurement error and may not be optimally measured by a simple sum of assets minus debts. Fourth, we classified wealth measures into quartiles to allow for known nonlinearities in the relationship between wealth and health. Using different functional forms may reveal different results. We generated simplified measures empirically; however, there may be policy-relevant reasons for including specific measures in particular studies. Finally, the measure of current smoking is simplistic; a measure based on more detailed smoking behaviors (e.g., taking into account dependence, duration, cessation, or frequency) would have been preferable.
Our findings suggest that simpler measures of wealth can be used in health research when it is not feasible to use net worth, which has been the standard. Our results suggest that for nonelderly adults, the simplest measure is the value of the primary residence, vehicles, and retirement funds. Similarly, for older adults, the simplest measure is the value of the primary residence, vehicles, and checking, savings, and money market accounts. These simpler measures significantly reduce the number of items to be assessed, which thereby reduces respondent burden, data-collection time, and associated costs. We recommend that additional research be conducted to examine other indicators, thresholds, life stages, and demographic subgroups to determine whether these results have broader generalizability.
Acknowledgments
This article was made possible by an award from the American Legacy Foundation.
The authors wish to acknowledge institutional support from the Population Research Center (grant 5 R24 HD042849) and the Center for Social Work Research at the University of Texas at Austin. We also thank Lou Mariano for statistical advice and Adriane Clomax and Tara Powell for their assistance with preparation of the article.
Human Participant Protection
The institutional review board at the University of Texas at Austin designated this study as exempt from the requirement for protocol approval because it used publicly available secondary data.
