Results

Some observations were deleted to eliminate likely errors in the reports of income. The GHWBI asks individuals to report their monthly family income in 11 categories. The three lowest categories—0, <$60, and $60–$499—cannot be treated as serious estimates of household income. We deleted these three categories (a total of 14,425 observations out of 709,183), as well as those respondents for whom income is missing (172,677 observations). We then regressed log income on indicators for the congressional district in which the respondent lived, educational categories, sex, age, age squared, race categories, marital status categories, and height. Thus, we predict the log of each individual's income by the mean of log incomes in his or her congressional district, modified by personal characteristics. This regression explains 37% of the variance, with a root mean square error (RMSE) of 0.67852. To eliminate outliers and implausible income reports, we dropped observations in which the absolute value of the difference between log income and its prediction exceeded 2.5 times the RMSE. This trimming lost 14,510 observations out of 450,417, or 3.22%. In all, we lost 28.4% of the original sample. In comparison, the US Census Bureau imputed income for 27.5% of households in the 2008 wave of the American Community Survey (ACS). As a check that our exclusions do not systematically bias income estimates compared with Census Bureau procedures, we compared the mean of the logarithm of income in each congressional district from the GHWBI with the logarithm of median income from the ACS. If income is approximately lognormal, then these should be close. The correlation was 0.961, with the GHWBI estimates about 6% lower, possibly attributable to the fact that the GHWBI data cover both 2008 and 2009.

We defined positive affect by the average of three dichotomous items (reports of happiness, enjoyment, and frequent smiling and laughter) and what we refer to as “blue affect”—the average of worry and sadness. Reports of stress (also dichotomous) were analyzed separately (as was anger, for which the results were similar but not shown) and life evaluation was measured using the Cantril ladder. The correlations between the emotional well-being measures and the ladder values had the expected sign but were modest in size (all <0.31). Positive affect, blue affect, and stress also were weakly correlated (positive and blue affect correlated –0.38, and –0.28, and 0.52 with stress.) The results shown here are similar when the constituents of positive and blue affect are analyzed separately.

As in other studies of well-being, we found that most people were quite happy and satisfied with their lives. About 85% of respondents experienced much positive affect (the average of smiling, enjoyment, and happiness) each day. Blue affect (sadness and worry) was reported by 24%, and stress was reported by 39%. The average of the Cantril ladder score was 6.76. Compared with about 150 other countries for which we have corresponding data from the Gallup World Poll, these results indicate that the US population ranks high on the ladder (ninth after the Scandinavian countries, Canada, The Netherlands, Switzerland, and New Zealand), and also does well in terms of happiness (fifth), smiling (33rd), and enjoyment (10th), but much less well on worry (89th from best), sadness (69th from best), and anger (75th). Americans report very high levels of stress (fifth among 151 countries).

Table 1 presents regressions of the four well-being measures on a set of demographic variables, which provide context for interpreting these measures. All of the predictors are dichotomous. The first row of the table shows the regression coefficient for an indicator of high income, defined as reporting a monthly income of at least $4,000, which corresponds to the top 58% of the population. These coefficients cannot be compared across the row, because the outcomes have different scales. The entries in other rows are ratios normalized by the coefficient on the high-income indicator, thus representing the estimated effect relative to the effect of increasing income by approximately 4-fold. The sign of each ratio is positive if its regression coefficient has the same sign as the coefficient for income (positive for positive affect, negative for blue affect, etc.). A coefficient >1 indicates an effect larger than that of the income dichotomy. Because higher incomes are always associated with better outcomes, positive ratios indicate that the predictor is associated with better outcomes, and negative ratios indicate the opposite.

With few exceptions, the various predictors have the same sign for all four well-being measures, but their relative sizes vary considerably. As might be expected, weekends are associated with improved affect, especially with reduced stress. Physical illness, headaches, spending a day alone, and caring for an adult all have relatively larger adverse effects on emotional well-being than on life evaluation. Headaches and being alone, like emotional well-being, are measured for yesterday, which may enhance their importance in the regressions. At the other extreme, being a college graduate is associated with high life evaluation but has only a small association with positive and blue affect and a (perhaps) counterintuitive relation with stress; all other factors being equal, college graduates report more stress than nongraduates. The Gallup World Poll found high levels of stress in high-GDP countries (16).

Religion has a substantial influence on improving positive affect and reducing reports of stress, but no effect on reducing sadness or worry. Females report slightly higher positive affect and life evaluation, but also more blue affect and much more stress. The presence of children at home is associated with significant increases in stress, sadness, and worry (6). As reported recently, older people enjoy greater emotional well-being, most notably a pronounced reduction in the experience of stress and anger (17). Smoking is an impressively strong predictor of low well-being—especially its emotional dimensions—even when income and education are controlled for. A propensity to smoke is in part genetically determined (18) and is a known indicator of a tense personality (19, 20).

Fig. 1 and Table 2 characterize the relationship between the dimensions of subjective well-being and household income. Fig. 1 presents averages over eight income groups for the three aspects of emotional well-being and for the Cantril ladder measure of life evaluation. Here blue affect and stress are converted to their complements, not blue and stress-free, so that higher values in the figure always refer to better psychological outcomes. Income is converted to an annual basis and plotted on a log scale. (The midpoints of each income range, used only in the figure, are imputed assuming that the underlying distribution of income is lognormal; the figure shows vertical lines for the top three interval limits.) Stress is the average of a yes/no response to the question: “Did you experience a lot of stress yesterday?” Thus, Fig. 1 shows the percentage of the population in each income group who did not report experiencing this emotion on the previous day. Not blue is 1 minus the average of the percentage reporting sadness and worry. The right-hand axis shows the average score on the ladder, with values ranging from 0 to 11.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Positive affect, blue affect, stress, and life evaluation in relation to household income. Positive affect is the average of the fractions of the population reporting happiness, smiling, and enjoyment. “Not blue” is 1 minus the average of the fractions of the population reporting worry and sadness. “Stress free” is the fraction of the population who did not report stress for the previous day. These three hedonic measures are marked on the left-hand scale. The ladder is the average reported number on a scale of 0–10, marked on the right-hand scale.

Fig. 1 shows that for all measures of experienced well-being, individuals in the lower- income groups do worse on average than those above them, but that those in the top two groups do not differ. For the two top categories to be equal, the entire range of the second category must lie above the satiation point. This observation implies that emotional well-being satiates somewhere in the third category of income from the top. We infer that beyond about $75,000/y, there is no improvement whatever in any of the three measures of emotional well-being. In contrast, the figure shows a fairly steady rise in life evaluation with log income over the entire range; the effects of income on individuals’ life evaluations show no satiation, at least to an amount well over $120,000.

Table 2 reports a formal test of satiation for the four measures, showing how the second-to-top income group (annual income $90,000–$120,000) differs from the group immediately below it ($60,000–$90,000) and from the group immediately above it (> $120,000). Positive affect, blue affect, and Cantril ladder score are all significantly improved in the first comparison with the exception of stress, which appears to satiate at a lower income level, roughly $60,000. In comparisons of the top two categories, only the ladder score shows a significant improvement with higher income. The small t values are remarkable in these very large samples. We conclude that lack of money brings both emotional misery and low life evaluation; similar results were found for anger. Beyond ~$75,000 in the contemporary United States, however, higher income is neither the road to experienced happiness nor the road to the relief of unhappiness or stress, although higher income continues to improve individuals’ life evaluations.

Below $75,000, many factors become gradually worse, at least on average. For example, the emotional pain associated with ill health depends on income; for those reporting a monthly income of at least $3,000 (about two-thirds of households), the fractions reporting blue affect with and without headaches are 38% and 19%, respectively, a difference of 19 percentage points. The corresponding values for those with a monthly income of <$1,000 (about 10% of households) are 70% and 38%, a difference of 32%. Table 3 shows that the pain of some of life's misfortunes, including asthma, divorce, and being alone, is significantly exacerbated by poverty; even the benefits of the weekend are less for the poor. Similar results apply to stress and positive affect.