The relationship between material wealth and subjective well-being has long been an object of casual fascination. But in recent years, economists, sociologists, and psychologists have shown increased interest in assessing the money-happiness nexus formally.
While the story is far from complete, important themes have begun to take shape. One is that money affects different aspects of well-being in different ways; in particular, wealthier people are more satisfied with their lives overall, but not necessarily better off in the day-to-day emotional sense. Being precise about how we define and measure “well-being” is essential.
At the same time, it is important to recognize that money is an oft-miscast protagonist; the second lesson from the recent academic literature is that other factors, such as good health and good friends, implicate our happiness more than do dollars and cents. Money can make us happier, yes, but its single-minded pursuit can, on net, impoverish well-being.
In this issue brief, I put the existing literature to the test, using more than a decade’s worth of General Social Survey (GSS) data to examine patterns in income and happiness among Americans. In large measure, I find strong support for the emerging academic consensus, reaching two principal conclusions.
- More money does make us happier—but there’s only so far it can go. I find that more affluent households are happier ones, and that happiness rises more or less steadily with income. However, income is also correlated with many other positive influences, such as good health and ample social time. Once we account for these other factors, the impact of money on happiness diminishes considerably. To the extent money buys happiness, it appears to do so by allowing increased consumption of other goods.
- Health, family, friends, and faith matter most. Good health is far and away the best predictor of happiness. Being married, having strong religious beliefs, having positive views of others, and spending time socializing are also extremely important determinants of emotional well-being.
Although not all Americans have equal access to all the elements of the good life, it’s important to note, that, generally speaking, we’re a happy bunch. Fully fourteen in fifteen Americans report being at least “pretty happy.” Nevertheless, policy is not as attuned to well-being as it could be; the challenge lies in moving beyond financial metrics to prioritize those programs and services that bring us the biggest happiness dividend for our social downpayment.
Does Money Make Us Happy?
If you had more money, would you be happier?
It’s a question people have asked for at least as long as money has existed. Certainly, many of us act as if the answer is yes. Why else would we spend most of our waking hours working or, failing that, playing the lottery?
But it’s also true that we do a lot of wondering about whether this manner of living is actually good for us. “There’s more to life than money,” we say, taking it as self-evident that “you can’t buy happiness.” Even on its best day, money is, we readily admit, mostly a means to an end, valuable for the experiences it enables (including, yes, inflating our egos), but devoid of intrinsic value. Not that we let that get in the way when there are Benjamins to be made.
So we could be excused for being more than a little confused. Is money good? Bad? Neither? Both?
Intuitively, I suspect, many of us come down somewhere in the middle, with a sort of Goldilocks attitude towards the green stuff: like most things, it’s best consumed in moderation. Moderation is a good rule of thumb.
Recently, though, economists have begun pushing for a more rigorous answer, deploying the full force of their analytic arsenal to better understand the complicated relationship between income and subjective well-being. And their findings are starting to change how we think about happiness.
To Recap: Did You Get Your $70,000 Raise?
I recently wrote about a Seattle-based CEO, Dan Price, who raised the minimum wage at his 120-person company to $70,000 a year—and financed it, in part, by cutting his own million dollar salary. The impetus for his bold move was a paper by Princeton’s Angus Deaton and Daniel Kahneman, which found that emotional well-being increases with income up to about $75,000 a year, after which it flattens out. In other words, beyond a certain baseline standard of living, some things are more important than money for our day-to-day happiness.
In my piece, I took an in-depth look at the Deaton-Kahneman paper, as well as other recent academic analyses assessing the money-happiness connection, including the just-released 2015 World Happiness Report (WHR).
I reached two main conclusions. The first was the need to distinguish between two types of “happiness”—experiential happiness (the kind that reflects our day-to-day emotional states) and evaluative happiness (our satisfaction with our lives as a whole), because, as it turns out, the two measures respond differently to income. The former, as Dan Price noticed, doesn’t depend on income beyond middle-class levels, but the latter—life satisfaction—does, in fact, continue to rise with income, even to quite high levels (this was another finding of the Deaton-Kahneman paper). Clarifying this distinction between affect and evaluation can go a long way toward reconciling some of the contradictory findings in the existing happiness research.
So money matters, sometimes. But the second point of consensus that I found was that other things are more important for our well-being than are dollars and cents. Among the factors that loom larger are health, social supports, freedom of choice, and a society’s ethos (for example, things like generosity and freedom from corruption). This is true both at the individual level, as well as cross-nationally.
But as interesting as discussing groundbreaking research can be, I’ve always found it more interesting to do my own. So that’s exactly the agenda of this issue brief: to put the existing literature to the test. More specifically, I wanted to see if, using a different data source than Deaton-Kahneman and the WHR, I could replicate their results. Replicability, is after all, the hallmark of scientific validity: if findings are sensitive to choice of data or idiosyncratic methods, maybe they aren’t findings after all.
To preview my main findings: Deaton and Kahneman are right. Income is associated with happiness, even at quite high levels, but this association attenuates once you factor in other life circumstances, such as health and family life. To learn why, read on.
What the General Social Survey Can Tell Us About Happiness
My data source is the General Social Survey, which has been conducted annually or bi-annually since 1972 by NORC (previously the National Opinion Research Center) at the University of Chicago. The GSS collects detailed data on Americans’ attitudes and behaviors, as well as about their demographic, social, and economic characteristics. The data allow researchers to understand the composition of America’s population and its values, and how these opinions, beliefs, and socioeconomic traits change, or don’t, over time. It’s among the most cited data sources in the social sciences.
Most of the data in the GSS is cross-sectional, providing a snapshot of the attitudes and life circumstances of American adults at a particular point in time. Every household in the U.S. has an equal chance of being selected—which ensures that the results are representative of all American adults. (Within each household, one adult is interviewed, with the interviewee’s responses taken as representative of all adults in the household; my analysis adjusts for the fact that adults living in larger households are less likely to be selected.)
Happiness has long been a part of the GSS. Each year, the survey asks the following type of question: “Taken all together, how would you say things are these days—would you say that you are very happy, pretty happy, or not too happy?” The responses, and their correlates, are the focus of my analysis. To balance timeliness with having a large sample size, I pooled the bi-annual GSS cross-sections from 2000 to 2014.
The first thing to note is the wording of the happiness query. It’s a bit of a middle ground between the emotional and evaluative concepts of happiness, with a leaning towards the latter, asking respondents to assess their whole lives (clearly evaluative), though on the time scale of “these days” and in terms of “happy” (more emotional).
Given what we know from the Deaton-Kahneman study, we would expect this satisfaction-leaning measure of happiness to respond to income, with richer people happier than poorer ones. And, indeed, as the figure below shows, this is exactly what I found.
The data points are mean household income for each income vigintile (a term for percentiles grouped in five percent intervals). The black dots represent the percentage of people in that vigintile reporting being “very happy” while the red X’s show the percentage reporting “not happy.” (You can determine the “pretty happy” share by subtracting the sum of the other two categories from 100 percent.) Since the data is based on an aggregate over fourteen years, the figure also adjusts for the fact that average happiness in the U.S. was higher in some years than others; dollars are expressed in constant 2014 terms. (Note, too, that GSS respondents report income in categories, so household income is estimated to be at the midpoint of the reported interval.)
The relationship between household income and happiness is remarkably close to linear, as shown by the tight clustering of points around the best-fit lines (in statistical language, the R2’s are near 1). While less than 20 percent of people in the bottom income decile (that’s two vigintiles) say they are “very happy,” nearly half of people in the top vigintile do. Similarly, fully a fifth of the poorest Americans are “not happy,” compared with about one in thirteen people in the top half of the income distribution.
So it certainly appears money has something to do with happiness. But as tight as the relationship is, we need to be wary of spurious correlations. Simple one-by-one plots of variables can obscure as much as they reveal; we have to consider the relevance of related factors we might be leaving out. In the context of money and happiness, we should wonder: what factors are related to income that also affect happiness?
One candidate is good health; another is leisure time. If wealthy people are healthier than poor people and have more time to spend doing fun things with family and friends, it could be that income is picking up the salutary effects of health and leisure. In other words, maybe it’s not income that’s making people happy, it’s things that coincide with income. Economists refer to this type of situation as omitted variables bias—we’re measuring something other than what we think we’re measuring. (It’s also worth considering the extent to which reverse causality could be at play; some research suggests that people with happier dispositions in youth go on to earn more money as adults.)
A Model of Happiness
What can we do to better tease out the effects of income? Happiness is clearly a complicated phenomenon; at any given moment, both our emotions and our general sense of satisfaction with our lives are dependent upon a complex web of intermingled factors, some of which are mostly under our control (for example, marital status, occupation, or whether we had a nice dinner with friends last night), some of which definitely aren’t (age, race), and some of which are a mix (health). Figuring out where income comes in isn’t easy.
But it doesn’t need to be a wild guessing game either. Rather than speculate about the relative contributions of each of these factors, we can use multiple regression—the most dependable tool in an economist’s toolkit—to incorporate them all explicitly and simultaneously into our happiness model. The variant of multiple regression I use in this analysis is known as “ordered probit,” which is mostly a fancy way of acknowledging the outcome in question (happiness) is measured in categorical terms (not happy, pretty happy, very happy).
Like conventional (linear) multiple regression, an ordered probit model measures the relationship between the response variable (happiness) and each explanatory variable, after taking into account the interplay of all the other variables included in the model. The difference—and the reason why you’d use an ordered probit model—is that it relaxes the assumption that the contrast between “not happy” and “pretty happy” is precisely equal to the gap between “pretty happy” and “very happy.” In other words, while a linear model assumes ordinal responses to a question can be interpreted cardinally, an ordered probit considers the possibility that such response categories simply happen to be thresholds we’ve marked on an underlying happiness continuum. By looking at how responses to the happiness question relate, on average, to respondents’ other attributes—education, health, age, income, and so forth—the ordered probit gives us the predicted probabilities a person with particular characteristics will fall into each of the happiness buckets the structure of the question imposes on the emotional continuum. If you think about it, this is a much more realistic situation than simply assuming there are three discrete happiness levels.
Based on the existing happiness research in economics and the questions asked in the GSS, I included the following factors in my model: household income (measured as the midpoint of the respondent-reported income interval, in constant 2014 dollars), education, labor force status, self-reported health status, region of residence and geography type, age, race, sex, marital status, household size, children (both having children and living with them), religiosity, political ideology, self-assessed relative income, self-assessed recent change in financial status, attitude toward government redistribution, view on whether people get ahead through hard work or luck, assessment of whether people are mostly helpful or selfish, confidence in government, confidence in business and financial institutions, socialization patterns, and survey year. These variables are summarized in the table below.
The Results
Because the GSS uses a split ballot design in which not every respondent is asked every question, I ran two versions of my model. Most of the variables are identical in both models. Model 1, which I call the “generosity, reciprocity, and institutional confidence” model, includes the questions on confidence in government, confidence in financial institutions and big businesses, support for government redistribution, and views about whether people are helpful or selfish. Model 2, the “social” model, includes measures of frequency of time spent socializing with family and friends. There is no overlap between households included in each model, but the average characteristics of each household grouping are quite similar.
Table 2 presents the results. On average (as shown in the first row), about a third of American adults are happy, while about one in fifteen is “not very happy.” That is a fairly remarkable fact: fourteen of fifteen Americans are at least “pretty happy.”
The remaining rows in the table give the change in the probability of being “not happy” or “very happy” that is associated with having a particular characteristic, assuming a person is otherwise average in all of the dimensions described in the table.
See the full chart on page 8 in the downloadable PDF
For each characteristic, pairwise comparisons are made between a baseline category and an alternative category. The happiness probability associated with the baseline characteristic (the first one described) is listed in the “Category 1” column (for example, being in income decile 9), while the happiness probability associated with the alternative, given after the “vs.,” is listed in the “Category 2” column (for example, income decile 2). The difference between the two (Category 1 minus Category 2), is given in the “Change” column—these changes are the relationships we’re interested in. Unless otherwise noted, all results presented are statistically significant at the 10 percent significance level; comparisons with the baseline that are omitted can be assumed insignificant.
Here’s the big result: no longer is income the star of the show. While there is some evidence that household income continues to matter even after incorporating the influence of other factors, it’s impact is modest.
In Model 1, going from the first income decile to the tenth is associated with an increase in the probability of being “very happy” of 11.1 percentage points and a decrease in the probability of being “not very happy” of 4.1 percentage points. This represents a major, and fairly unlikely, change in income, however—from about $7,300 to $226,000 a year. The effect on happiness of going from income decile two to nine—a major jump in its own right—is only about half as large. In Model 2, the estimated effects of income are even smaller—and, in fact, statistically indistinguishable from zero. This suggests more typical changes in income, which are much smaller still, do not have much of an impact on happiness.
Instead, several other factors have much larger estimated associations with happiness. Topping the list is health. People who report being in excellent health are about 33 percentage points more likely to be “very happy” than those in poor health; even the difference between good and excellent health, about 14 percentage points, is fairly enormous.
Marital status also appears to have a major connection with happiness; compared to people in other relationship statuses, married people are between 10 and 27 percentage points more likely to be “very happy.” Religiosity also has a consistent impact across both models, with strongly religious people between 5 and 13 percentage points more likely to be “very happy” than those with weak or no religious beliefs. Political views may play a similar role; as with religion, those with the strongest feelings—in this case, extreme liberals—are significantly happier than moderates.
The effect of other factors are more mixed, appearing as insignificant in at least one of the models. (Note, however, that the lack of significance may not necessarily mean that they are unrelated to happiness so much as their effect is estimated imprecisely, given that other, related variables are included in the model.)
One such factor is race. Model 1 suggests that, even controlling for a range of other factors that conceivably affect happiness, minorities are about 4 percentage points more likely than whites to be “unhappy” and about 9 percentage points less likely to be “very happy.” Although the statistical significance of this result diminishes when we account for socialization patterns in Model 2, the fact that some racial groups appear to be systematically less happy is troubling.
The effect of education is ambiguous. In Model 1, higher levels of education are associated with being less happy, while in Model 2, the opposite is true. One explanation may be that education does not have a straightforward impact on happiness. High school dropouts may struggle with poor employment prospects and material deprivation, but it may also be the case that highly-educated overachievers spend much of their time stressed out by demanding jobs. Indeed, the inconsistencies between the models could drive from these other, related life factors incorporated in each regression. For example, accounting for confidence in institutions and attitudes about American society (as in Model 1) or time spent socializing (as in Model 2) may capture some of the effects of education, as would variables included in both models, such as income or marital status. Put differently, if the impact of education on happiness operates mostly through education’s influence on earnings, marriageability, or attitudes—or if it is systematically related to things like race—then the measured impact of education, in and of itself, will be smaller.
A similar pattern holds for labor force status: as we might expect, unemployed people are appreciably less happy than those working full-time, though the effect is significant only in Model 1. There is also some evidence that people who’ve had children are less happy, but here again the effect of children may well be clouded by the inclusion of marital status and separately accounting for living with children under seventeen.
As interesting as what influences happiness is what does not. Age, sex, household size, opinion of whether Americans get ahead through hard work or luck, and (for the most part) where people live have no significant relationships with their reported happiness. On one hand, some of these non-findings are good—we might hope that happiness is not systematically related to something like sex.
On the other hand, the lack, say, of an age finding may not mean that age doesn’t influence happiness, but instead may be indicative of a more nuanced pattern. Previous research, for instance, has found that happiness takes a U-shape by age., with younger and older people being more happy than those in middle age. Further, as noted previously, the lack of significance may be attributable to the relationship between these factors and others included in the models.
What are significant, however, are several of the model-specific factors. In Model 1, people who view others as mostly helpful are 4.1 percentage points more likely to be “very happy” than those who view others as selfish. In addition, those with low confidence in financial institutions and big businesses are significantly more likely to be “very happy” than those with high confidence, by 7.1 percentage points. To the extent that these attitudes are reflective of an underlying predisposition to be generous or charitable in caring for the welfare of others, they underscore the World Happiness Report’s finding that a society’s generosity and happiness go together.
However, the results also show that those who do not support government redistribution are happier than those who do. Whether this is because they think redistribution is already prevalent enough or whether this anti-generosity finding is a challenge to the WHR is a question for further research; indeed, it would not be surprising to find that people who are more satisfied with the status quo are happier. It’s also possible that the generosity effect is being captured by the other variables in the model (that is to say, what the model is telling us is that among people who are otherwise similar, including the extent to which they view others as helpful and don’t trust big businesses, those who don’t support government redistribution are happier than those that do).
In Model 2, the impact of socialization is quite dramatic; including it is entirely responsible for eliminating the effects of income and race on happiness. And not only does accounting for socializing wipe out other variables; it also has a sizeable impact on happiness itself. Those who are in the 75th percentile of evenings spent with relatives or friends are 6.6 percentage points more likely to be “very happy” than those at the 25th percentile. Once again, this is in keeping with the WHR and Deaton and Kahneman’s research. Friends and family are at the core of our well-being.
Back to Money
To recap, lots of things in our lives affect our happiness, and many of them matter more than money. But it’s also important to consider that income might matter in ways other than its absolute amount. To test this, both models include two additional income variables, one related to assessing relative income and the other measuring changes in financial status. Prior research has found that not only do people evaluate their income in terms of what those around them make and what they themselves have earned in the past, but also that these relative assessments mediate the relationship between absolute income and happiness.
My results show both propositions to be accurate. Across both models, those who say their incomes are “below average” or “far below average” are, respectively, about 7 and 13.5 percentage points less likely to be “very happy” than those who consider their income to be average. At the same time, compared to those whose finances have remained stable, those who say their financial condition has worsened are about 7 percentage points less likely to be “very happy,” while those whose finances have improved are about 9 percentage points more likely to be “very happy.”
For money to bolster our senses of well-being, we don’t necessarily need more of it; instead, we need to be satisfied with our financial position. Indeed, when I directly included a measure of financial satisfaction in my model (not shown), the impact was nearly as large as health in magnitude and it wiped out any effect of absolute income.
Figure 2 summarizes the main conclusion of my analysis. It’s a repeat of Figure 1, only now it shows the relationship between income and happiness after accounting for all the other factors we’ve discussed (the top panel corresponds to Model 1 and the bottom panel depicts Model 2). The takeaway is that the trend lines are basically flat; in Model 1, there is a small positive relationship between absolute income and happiness, while in Model 2 there is no relationship.
Of course, as I’ve tried to emphasize throughout, we should be circumspect in our interpretations of the results. Even though some of the patterns are striking and although I use terms like “effects” colloquially in describing relationships in the data, we should remember that, as powerful as multiple regression is as an analytical tool, it can’t, by itself, give the final word on cause and effect.
To definitively ascertain the effect of various life circumstances on life satisfaction, we’d have to run a controlled experiment on a randomly selected, representative group of people—picture a happiness lab. In the absence of such a sterilized environment, there are simply too many things that affect both happiness and income that we cannot be sure we are accurately disentangling causality. But it is just as clear that having researchers comprehensively manipulate people’s well-being is not possible in practice. Usually, the best we can hope for is so-called “natural experiments” that exploit organically occurring variations in life circumstances or “panel” studies that track people over time—but these are often hard to come by or expensive. So we settle for the insights of multiple regression, and bear in mind that our findings are associations, not causations.
Policy Lessons
Fortunately, a lack of unassailable quantification doesn’t mean there aren’t lessons for policy to be learned from the economics of happiness. The first lesson is that we ought to be much more comprehensive in our measures of national progress and well-being. Income and wealth are part of the story, but they are not objectives to be pursued to the exclusion of all else. The danger lies in ease of measurement: because financial variables are easier to quantify than other components of the good life, they are often elevated to a place of outsized importance. So that’s challenge number one: devise more and better measures of national welfare.
The second lesson is that there is more than one route to happiness, just as there is more than one way to become rich. Our circumstances, skills, attitudes, and desires vary, and policy must appreciate—indeed, embrace—this diversity.
But as diverse as our experiences are, the correlates of happiness do apply consistently in many cases. And that’s lesson three: happiness research gives us a compass for what policy areas we ought to be investing in more heavily.
The good news is that we are, in many ways, headed in the right direction. We already spend a great deal on health, and its strong association with happiness suggests this is a smart priority. But we can also do a better job of understanding which types of health services deliver the largest happiness dividend. Similarly, combating inequality (income, racial, and otherwise) and ensuring the gains of economic growth are more fairly shared—something that gets a lot of talk but little action—should be near the top of the list, because, as we’ve seen, people can’t help but compare themselves.
The same prioritization agenda applies, perhaps even more forcefully, to policy areas that get less attention. For example, finding meaningful work for the unemployed (perhaps through public sector jobs programs) would likely do more for happiness than being hawkish on inflation. Similarly, policies that encourage the formation and maintenance of stable, loving families is something that would not only boost happiness, but is also an area ripe for bipartisan cooperation. Finally, government should be more involved in building strong communities and fomenting social capital—think of it as a national happiness infrastructure. Few things do more for our well-being, and few things are more properly in the purview of the public sector.
Does money buy happiness?
Sometimes yes, sometimes no, and mostly conditionally. Money and happiness are both things to strive for, and they often coincide. The challenge lies when they come into conflict, and it is at those times we would do well to remember that having the former in the absence of the latter is too steep a price to pay.
Should You Buy Buying Happiness?
The relationship between material wealth and subjective well-being has long been an object of casual fascination. But in recent years, economists, sociologists, and psychologists have shown increased interest in assessing the money-happiness nexus formally.
While the story is far from complete, important themes have begun to take shape. One is that money affects different aspects of well-being in different ways; in particular, wealthier people are more satisfied with their lives overall, but not necessarily better off in the day-to-day emotional sense. Being precise about how we define and measure “well-being” is essential.
At the same time, it is important to recognize that money is an oft-miscast protagonist; the second lesson from the recent academic literature is that other factors, such as good health and good friends, implicate our happiness more than do dollars and cents. Money can make us happier, yes, but its single-minded pursuit can, on net, impoverish well-being.
In this issue brief, I put the existing literature to the test, using more than a decade’s worth of General Social Survey (GSS) data to examine patterns in income and happiness among Americans. In large measure, I find strong support for the emerging academic consensus, reaching two principal conclusions.
Although not all Americans have equal access to all the elements of the good life, it’s important to note, that, generally speaking, we’re a happy bunch. Fully fourteen in fifteen Americans report being at least “pretty happy.” Nevertheless, policy is not as attuned to well-being as it could be; the challenge lies in moving beyond financial metrics to prioritize those programs and services that bring us the biggest happiness dividend for our social downpayment.
Does Money Make Us Happy?
If you had more money, would you be happier?
It’s a question people have asked for at least as long as money has existed. Certainly, many of us act as if the answer is yes. Why else would we spend most of our waking hours working or, failing that, playing the lottery?
But it’s also true that we do a lot of wondering about whether this manner of living is actually good for us. “There’s more to life than money,” we say, taking it as self-evident that “you can’t buy happiness.” Even on its best day, money is, we readily admit, mostly a means to an end, valuable for the experiences it enables (including, yes, inflating our egos), but devoid of intrinsic value. Not that we let that get in the way when there are Benjamins to be made.
So we could be excused for being more than a little confused. Is money good? Bad? Neither? Both?
Intuitively, I suspect, many of us come down somewhere in the middle, with a sort of Goldilocks attitude towards the green stuff: like most things, it’s best consumed in moderation. Moderation is a good rule of thumb.
Recently, though, economists have begun pushing for a more rigorous answer, deploying the full force of their analytic arsenal to better understand the complicated relationship between income and subjective well-being. And their findings are starting to change how we think about happiness.
To Recap: Did You Get Your $70,000 Raise?
I recently wrote about a Seattle-based CEO, Dan Price, who raised the minimum wage at his 120-person company to $70,000 a year—and financed it, in part, by cutting his own million dollar salary. The impetus for his bold move was a paper by Princeton’s Angus Deaton and Daniel Kahneman, which found that emotional well-being increases with income up to about $75,000 a year, after which it flattens out. In other words, beyond a certain baseline standard of living, some things are more important than money for our day-to-day happiness.
In my piece, I took an in-depth look at the Deaton-Kahneman paper, as well as other recent academic analyses assessing the money-happiness connection, including the just-released 2015 World Happiness Report (WHR).
I reached two main conclusions. The first was the need to distinguish between two types of “happiness”—experiential happiness (the kind that reflects our day-to-day emotional states) and evaluative happiness (our satisfaction with our lives as a whole), because, as it turns out, the two measures respond differently to income. The former, as Dan Price noticed, doesn’t depend on income beyond middle-class levels, but the latter—life satisfaction—does, in fact, continue to rise with income, even to quite high levels (this was another finding of the Deaton-Kahneman paper). Clarifying this distinction between affect and evaluation can go a long way toward reconciling some of the contradictory findings in the existing happiness research.
So money matters, sometimes. But the second point of consensus that I found was that other things are more important for our well-being than are dollars and cents. Among the factors that loom larger are health, social supports, freedom of choice, and a society’s ethos (for example, things like generosity and freedom from corruption). This is true both at the individual level, as well as cross-nationally.
But as interesting as discussing groundbreaking research can be, I’ve always found it more interesting to do my own. So that’s exactly the agenda of this issue brief: to put the existing literature to the test. More specifically, I wanted to see if, using a different data source than Deaton-Kahneman and the WHR, I could replicate their results. Replicability, is after all, the hallmark of scientific validity: if findings are sensitive to choice of data or idiosyncratic methods, maybe they aren’t findings after all.
To preview my main findings: Deaton and Kahneman are right. Income is associated with happiness, even at quite high levels, but this association attenuates once you factor in other life circumstances, such as health and family life. To learn why, read on.
What the General Social Survey Can Tell Us About Happiness
My data source is the General Social Survey, which has been conducted annually or bi-annually since 1972 by NORC (previously the National Opinion Research Center) at the University of Chicago. The GSS collects detailed data on Americans’ attitudes and behaviors, as well as about their demographic, social, and economic characteristics. The data allow researchers to understand the composition of America’s population and its values, and how these opinions, beliefs, and socioeconomic traits change, or don’t, over time. It’s among the most cited data sources in the social sciences.
Most of the data in the GSS is cross-sectional, providing a snapshot of the attitudes and life circumstances of American adults at a particular point in time. Every household in the U.S. has an equal chance of being selected—which ensures that the results are representative of all American adults. (Within each household, one adult is interviewed, with the interviewee’s responses taken as representative of all adults in the household; my analysis adjusts for the fact that adults living in larger households are less likely to be selected.)
Happiness has long been a part of the GSS. Each year, the survey asks the following type of question: “Taken all together, how would you say things are these days—would you say that you are very happy, pretty happy, or not too happy?” The responses, and their correlates, are the focus of my analysis. To balance timeliness with having a large sample size, I pooled the bi-annual GSS cross-sections from 2000 to 2014.
The first thing to note is the wording of the happiness query. It’s a bit of a middle ground between the emotional and evaluative concepts of happiness, with a leaning towards the latter, asking respondents to assess their whole lives (clearly evaluative), though on the time scale of “these days” and in terms of “happy” (more emotional).
Given what we know from the Deaton-Kahneman study, we would expect this satisfaction-leaning measure of happiness to respond to income, with richer people happier than poorer ones. And, indeed, as the figure below shows, this is exactly what I found.
The data points are mean household income for each income vigintile (a term for percentiles grouped in five percent intervals). The black dots represent the percentage of people in that vigintile reporting being “very happy” while the red X’s show the percentage reporting “not happy.” (You can determine the “pretty happy” share by subtracting the sum of the other two categories from 100 percent.) Since the data is based on an aggregate over fourteen years, the figure also adjusts for the fact that average happiness in the U.S. was higher in some years than others; dollars are expressed in constant 2014 terms. (Note, too, that GSS respondents report income in categories, so household income is estimated to be at the midpoint of the reported interval.)
The relationship between household income and happiness is remarkably close to linear, as shown by the tight clustering of points around the best-fit lines (in statistical language, the R2’s are near 1). While less than 20 percent of people in the bottom income decile (that’s two vigintiles) say they are “very happy,” nearly half of people in the top vigintile do. Similarly, fully a fifth of the poorest Americans are “not happy,” compared with about one in thirteen people in the top half of the income distribution.
So it certainly appears money has something to do with happiness. But as tight as the relationship is, we need to be wary of spurious correlations. Simple one-by-one plots of variables can obscure as much as they reveal; we have to consider the relevance of related factors we might be leaving out. In the context of money and happiness, we should wonder: what factors are related to income that also affect happiness?
One candidate is good health; another is leisure time. If wealthy people are healthier than poor people and have more time to spend doing fun things with family and friends, it could be that income is picking up the salutary effects of health and leisure. In other words, maybe it’s not income that’s making people happy, it’s things that coincide with income. Economists refer to this type of situation as omitted variables bias—we’re measuring something other than what we think we’re measuring. (It’s also worth considering the extent to which reverse causality could be at play; some research suggests that people with happier dispositions in youth go on to earn more money as adults.)
A Model of Happiness
What can we do to better tease out the effects of income? Happiness is clearly a complicated phenomenon; at any given moment, both our emotions and our general sense of satisfaction with our lives are dependent upon a complex web of intermingled factors, some of which are mostly under our control (for example, marital status, occupation, or whether we had a nice dinner with friends last night), some of which definitely aren’t (age, race), and some of which are a mix (health). Figuring out where income comes in isn’t easy.
But it doesn’t need to be a wild guessing game either. Rather than speculate about the relative contributions of each of these factors, we can use multiple regression—the most dependable tool in an economist’s toolkit—to incorporate them all explicitly and simultaneously into our happiness model. The variant of multiple regression I use in this analysis is known as “ordered probit,” which is mostly a fancy way of acknowledging the outcome in question (happiness) is measured in categorical terms (not happy, pretty happy, very happy).
Like conventional (linear) multiple regression, an ordered probit model measures the relationship between the response variable (happiness) and each explanatory variable, after taking into account the interplay of all the other variables included in the model. The difference—and the reason why you’d use an ordered probit model—is that it relaxes the assumption that the contrast between “not happy” and “pretty happy” is precisely equal to the gap between “pretty happy” and “very happy.” In other words, while a linear model assumes ordinal responses to a question can be interpreted cardinally, an ordered probit considers the possibility that such response categories simply happen to be thresholds we’ve marked on an underlying happiness continuum. By looking at how responses to the happiness question relate, on average, to respondents’ other attributes—education, health, age, income, and so forth—the ordered probit gives us the predicted probabilities a person with particular characteristics will fall into each of the happiness buckets the structure of the question imposes on the emotional continuum. If you think about it, this is a much more realistic situation than simply assuming there are three discrete happiness levels.
Based on the existing happiness research in economics and the questions asked in the GSS, I included the following factors in my model: household income (measured as the midpoint of the respondent-reported income interval, in constant 2014 dollars), education, labor force status, self-reported health status, region of residence and geography type, age, race, sex, marital status, household size, children (both having children and living with them), religiosity, political ideology, self-assessed relative income, self-assessed recent change in financial status, attitude toward government redistribution, view on whether people get ahead through hard work or luck, assessment of whether people are mostly helpful or selfish, confidence in government, confidence in business and financial institutions, socialization patterns, and survey year. These variables are summarized in the table below.
The Results
Because the GSS uses a split ballot design in which not every respondent is asked every question, I ran two versions of my model. Most of the variables are identical in both models. Model 1, which I call the “generosity, reciprocity, and institutional confidence” model, includes the questions on confidence in government, confidence in financial institutions and big businesses, support for government redistribution, and views about whether people are helpful or selfish. Model 2, the “social” model, includes measures of frequency of time spent socializing with family and friends. There is no overlap between households included in each model, but the average characteristics of each household grouping are quite similar.
Table 2 presents the results. On average (as shown in the first row), about a third of American adults are happy, while about one in fifteen is “not very happy.” That is a fairly remarkable fact: fourteen of fifteen Americans are at least “pretty happy.”
The remaining rows in the table give the change in the probability of being “not happy” or “very happy” that is associated with having a particular characteristic, assuming a person is otherwise average in all of the dimensions described in the table.
See the full chart on page 8 in the downloadable PDF
For each characteristic, pairwise comparisons are made between a baseline category and an alternative category. The happiness probability associated with the baseline characteristic (the first one described) is listed in the “Category 1” column (for example, being in income decile 9), while the happiness probability associated with the alternative, given after the “vs.,” is listed in the “Category 2” column (for example, income decile 2). The difference between the two (Category 1 minus Category 2), is given in the “Change” column—these changes are the relationships we’re interested in. Unless otherwise noted, all results presented are statistically significant at the 10 percent significance level; comparisons with the baseline that are omitted can be assumed insignificant.
Here’s the big result: no longer is income the star of the show. While there is some evidence that household income continues to matter even after incorporating the influence of other factors, it’s impact is modest.
In Model 1, going from the first income decile to the tenth is associated with an increase in the probability of being “very happy” of 11.1 percentage points and a decrease in the probability of being “not very happy” of 4.1 percentage points. This represents a major, and fairly unlikely, change in income, however—from about $7,300 to $226,000 a year. The effect on happiness of going from income decile two to nine—a major jump in its own right—is only about half as large. In Model 2, the estimated effects of income are even smaller—and, in fact, statistically indistinguishable from zero. This suggests more typical changes in income, which are much smaller still, do not have much of an impact on happiness.
Instead, several other factors have much larger estimated associations with happiness. Topping the list is health. People who report being in excellent health are about 33 percentage points more likely to be “very happy” than those in poor health; even the difference between good and excellent health, about 14 percentage points, is fairly enormous.
Marital status also appears to have a major connection with happiness; compared to people in other relationship statuses, married people are between 10 and 27 percentage points more likely to be “very happy.” Religiosity also has a consistent impact across both models, with strongly religious people between 5 and 13 percentage points more likely to be “very happy” than those with weak or no religious beliefs. Political views may play a similar role; as with religion, those with the strongest feelings—in this case, extreme liberals—are significantly happier than moderates.
The effect of other factors are more mixed, appearing as insignificant in at least one of the models. (Note, however, that the lack of significance may not necessarily mean that they are unrelated to happiness so much as their effect is estimated imprecisely, given that other, related variables are included in the model.)
One such factor is race. Model 1 suggests that, even controlling for a range of other factors that conceivably affect happiness, minorities are about 4 percentage points more likely than whites to be “unhappy” and about 9 percentage points less likely to be “very happy.” Although the statistical significance of this result diminishes when we account for socialization patterns in Model 2, the fact that some racial groups appear to be systematically less happy is troubling.
The effect of education is ambiguous. In Model 1, higher levels of education are associated with being less happy, while in Model 2, the opposite is true. One explanation may be that education does not have a straightforward impact on happiness. High school dropouts may struggle with poor employment prospects and material deprivation, but it may also be the case that highly-educated overachievers spend much of their time stressed out by demanding jobs. Indeed, the inconsistencies between the models could drive from these other, related life factors incorporated in each regression. For example, accounting for confidence in institutions and attitudes about American society (as in Model 1) or time spent socializing (as in Model 2) may capture some of the effects of education, as would variables included in both models, such as income or marital status. Put differently, if the impact of education on happiness operates mostly through education’s influence on earnings, marriageability, or attitudes—or if it is systematically related to things like race—then the measured impact of education, in and of itself, will be smaller.
A similar pattern holds for labor force status: as we might expect, unemployed people are appreciably less happy than those working full-time, though the effect is significant only in Model 1. There is also some evidence that people who’ve had children are less happy, but here again the effect of children may well be clouded by the inclusion of marital status and separately accounting for living with children under seventeen.
As interesting as what influences happiness is what does not. Age, sex, household size, opinion of whether Americans get ahead through hard work or luck, and (for the most part) where people live have no significant relationships with their reported happiness. On one hand, some of these non-findings are good—we might hope that happiness is not systematically related to something like sex.
On the other hand, the lack, say, of an age finding may not mean that age doesn’t influence happiness, but instead may be indicative of a more nuanced pattern. Previous research, for instance, has found that happiness takes a U-shape by age., with younger and older people being more happy than those in middle age. Further, as noted previously, the lack of significance may be attributable to the relationship between these factors and others included in the models.
What are significant, however, are several of the model-specific factors. In Model 1, people who view others as mostly helpful are 4.1 percentage points more likely to be “very happy” than those who view others as selfish. In addition, those with low confidence in financial institutions and big businesses are significantly more likely to be “very happy” than those with high confidence, by 7.1 percentage points. To the extent that these attitudes are reflective of an underlying predisposition to be generous or charitable in caring for the welfare of others, they underscore the World Happiness Report’s finding that a society’s generosity and happiness go together.
However, the results also show that those who do not support government redistribution are happier than those who do. Whether this is because they think redistribution is already prevalent enough or whether this anti-generosity finding is a challenge to the WHR is a question for further research; indeed, it would not be surprising to find that people who are more satisfied with the status quo are happier. It’s also possible that the generosity effect is being captured by the other variables in the model (that is to say, what the model is telling us is that among people who are otherwise similar, including the extent to which they view others as helpful and don’t trust big businesses, those who don’t support government redistribution are happier than those that do).
In Model 2, the impact of socialization is quite dramatic; including it is entirely responsible for eliminating the effects of income and race on happiness. And not only does accounting for socializing wipe out other variables; it also has a sizeable impact on happiness itself. Those who are in the 75th percentile of evenings spent with relatives or friends are 6.6 percentage points more likely to be “very happy” than those at the 25th percentile. Once again, this is in keeping with the WHR and Deaton and Kahneman’s research. Friends and family are at the core of our well-being.
Back to Money
To recap, lots of things in our lives affect our happiness, and many of them matter more than money. But it’s also important to consider that income might matter in ways other than its absolute amount. To test this, both models include two additional income variables, one related to assessing relative income and the other measuring changes in financial status. Prior research has found that not only do people evaluate their income in terms of what those around them make and what they themselves have earned in the past, but also that these relative assessments mediate the relationship between absolute income and happiness.
My results show both propositions to be accurate. Across both models, those who say their incomes are “below average” or “far below average” are, respectively, about 7 and 13.5 percentage points less likely to be “very happy” than those who consider their income to be average. At the same time, compared to those whose finances have remained stable, those who say their financial condition has worsened are about 7 percentage points less likely to be “very happy,” while those whose finances have improved are about 9 percentage points more likely to be “very happy.”
For money to bolster our senses of well-being, we don’t necessarily need more of it; instead, we need to be satisfied with our financial position. Indeed, when I directly included a measure of financial satisfaction in my model (not shown), the impact was nearly as large as health in magnitude and it wiped out any effect of absolute income.
Figure 2 summarizes the main conclusion of my analysis. It’s a repeat of Figure 1, only now it shows the relationship between income and happiness after accounting for all the other factors we’ve discussed (the top panel corresponds to Model 1 and the bottom panel depicts Model 2). The takeaway is that the trend lines are basically flat; in Model 1, there is a small positive relationship between absolute income and happiness, while in Model 2 there is no relationship.
Of course, as I’ve tried to emphasize throughout, we should be circumspect in our interpretations of the results. Even though some of the patterns are striking and although I use terms like “effects” colloquially in describing relationships in the data, we should remember that, as powerful as multiple regression is as an analytical tool, it can’t, by itself, give the final word on cause and effect.
To definitively ascertain the effect of various life circumstances on life satisfaction, we’d have to run a controlled experiment on a randomly selected, representative group of people—picture a happiness lab. In the absence of such a sterilized environment, there are simply too many things that affect both happiness and income that we cannot be sure we are accurately disentangling causality. But it is just as clear that having researchers comprehensively manipulate people’s well-being is not possible in practice. Usually, the best we can hope for is so-called “natural experiments” that exploit organically occurring variations in life circumstances or “panel” studies that track people over time—but these are often hard to come by or expensive. So we settle for the insights of multiple regression, and bear in mind that our findings are associations, not causations.
Policy Lessons
Fortunately, a lack of unassailable quantification doesn’t mean there aren’t lessons for policy to be learned from the economics of happiness. The first lesson is that we ought to be much more comprehensive in our measures of national progress and well-being. Income and wealth are part of the story, but they are not objectives to be pursued to the exclusion of all else. The danger lies in ease of measurement: because financial variables are easier to quantify than other components of the good life, they are often elevated to a place of outsized importance. So that’s challenge number one: devise more and better measures of national welfare.
The second lesson is that there is more than one route to happiness, just as there is more than one way to become rich. Our circumstances, skills, attitudes, and desires vary, and policy must appreciate—indeed, embrace—this diversity.
But as diverse as our experiences are, the correlates of happiness do apply consistently in many cases. And that’s lesson three: happiness research gives us a compass for what policy areas we ought to be investing in more heavily.
The good news is that we are, in many ways, headed in the right direction. We already spend a great deal on health, and its strong association with happiness suggests this is a smart priority. But we can also do a better job of understanding which types of health services deliver the largest happiness dividend. Similarly, combating inequality (income, racial, and otherwise) and ensuring the gains of economic growth are more fairly shared—something that gets a lot of talk but little action—should be near the top of the list, because, as we’ve seen, people can’t help but compare themselves.
The same prioritization agenda applies, perhaps even more forcefully, to policy areas that get less attention. For example, finding meaningful work for the unemployed (perhaps through public sector jobs programs) would likely do more for happiness than being hawkish on inflation. Similarly, policies that encourage the formation and maintenance of stable, loving families is something that would not only boost happiness, but is also an area ripe for bipartisan cooperation. Finally, government should be more involved in building strong communities and fomenting social capital—think of it as a national happiness infrastructure. Few things do more for our well-being, and few things are more properly in the purview of the public sector.
Does money buy happiness?
Sometimes yes, sometimes no, and mostly conditionally. Money and happiness are both things to strive for, and they often coincide. The challenge lies when they come into conflict, and it is at those times we would do well to remember that having the former in the absence of the latter is too steep a price to pay.