Happiness & Public Policy

Thanks to commenter Conchis for bringing this important paper to my attention.

Heterogeneity in Reported Well-being: Evidence from Twelve European Countries by Andrew Clark, Fabrice Etilé, Fabien Postel-Vinay, Claudia Senik, Karine Van der Straeten

CONCLUSION. This paper modelled the relationship between income and self-reported well-being using random-effect techniques applied to panel data from twelve European countries. We show that people are different, and in more complicated ways than just having different intercepts. We are not able to distinguish between heterogeneity in the utility function (translating income into utility) and heterogeneity in the expression function (turning utility into reported well-being). We can, however, strongly reject the hypothesis that individuals carry out these joint transformations in the same way.

We identify four classes of individuals, and show that the “marginal well-being effect of income” is very different across these classes. In particular one class is satisfied and has a high marginal well-being effect, while another is dissatisfied and has a low marginal wellbeing effect. Descriptive statistics reveal demographic and country patterns between classes. This has at least two important implications. First, in a political economy sense, as the effect of income differs sharply across classes (and classes are not independently distributed between countries), we would expect average opinion regarding economic policies to differ across countries. . . This is a subject for ongoing research.

Perhaps more importantly, our results suggests that aggregating data across diverse populations may be a dangerous practice. Individuals, who seem to fall naturally into a number of different classes, differ in ways that are far more complicated than those picked up by a simple fixed effect. The trend towards comparative research in social science, whereby data from different countries are compared, is laudable. Nonetheless, our results suggest that the blind aggregation of diverse populations risks producing empirical results that are false for everybody.

The econometrics in the happiness literature is clearly edging its way toward adequacy. What we have had is not quite useless. But now, minimally credible statistics that take into account individual variation throws a huge wrench into attempts to ground highly specified policy on happiness studies. Heterogeneity is very problematic for would-be benevolent technocrats. First, they’re stuck with a big information problem. They’d need to know about dynamiclly shifting individual differences at a finer grain of detail than is possible. Second, heterogeneity prevents you from basing policy on the welfare of an “representative” agent. If it is true, and it is true, for example, that income has a big positive effect on some people’s happiness, or that simply keeping very busy has a big positive effect, then tax policy aimed at incentivizing the consumption of leisure against the production of wealth is going to cause a lot of immediate micro-harm, not just the delayed macro-harm of slower growth. “Truces” in income/status “arms races” are very unlikely to be pareto improvements. (Which means the situation is not accurately characterized as an “arms race.”) This may not bother a Benthamite technocrat willing to overlook individual harms if the net utility is positive. But the Benthamite is still stuck with the informational problem. And ignoring the separateness of persons is terribly wrong, which is why Benthamism is false. Morally legitimate policy is left having to take pluralism and the separateness of persons into account, which generally means worrying primarily about the structure of the basic framework of interaction, trying to ensure that it is sufficiently general and neutral between reasonable sets of beliefs and preferences.

Also, back to the paper, note the inability to “distinguish between heterogeneity in the utility function (translating income into utility) and heterogeneity in the expression function (turning utility into reported well-being).” The recognition that there is a distinction between real well-being and self-reported well-being has big implications. We don’t know to what extent talk tracks fact unless we know how talk relates to fact. The recognition of variation in the expression function throws in a further wrinkle. If the expression function was homogenous, discovery of that function would allow us to infer real well-being from talk. But if there is variation in the reporting function, and that variation cannot be accounted for by a general law-like principle, then aren’t we getting pretty close to empirical proof of the near-uselessness of self-reports as a proxy for real well-being?

For Bob, income has a big positive effect on well-being, but, despite growing happiness from growing riches, Bob in his modesty won’t ever put himself in the highest category of reported well-being. Karen is sort of miserable, and higher income means almost nothing to her. But she’s a shallow, ideological, free-market economist, and worries that there must something wrong with her for not being happier given her wealth. So she says she’s “pretty happy” even though she’s not. Bob and Karen report themselves as being in the same category of self-reported well-being. What do we know about well-being? What do we know about the relationship between income and well-being?

Money Does Matter! Evidence from Increasing Real Incomes and Life Satisfaction in East Germany Following Reunification by Paul Frijters, John P. Haisken-DeNew and Michael A. Shields, American Economic Review, Vol. 94, No. 3, June 2004

ABSTRACT. In this paper we investigate how life satisfaction (or happiness) is affected by a substantial increase in real income. Our context is East Germany in the decade following reunification, and we implement a new fixed-effect estimator for ordinal life satisfaction and develop a decomposition approach that accounts for new entrants and panel attrition. Using data from the German Socio-Economic Panel we find that average life satisfaction in East Germany increased by around 20% between 1991 and 2001, leading to a clear convergence with West Germany. Importantly, increased real household incomes in East Germany accounted for around 35-40% of this increase, which corresponds to the economists’ view that money surely matters.

The great thing about this paper is not that it tells us that money matters, but that it is better statistics, doing something to take into account individual fixed effects.

I also highly recommend Frijter’s and Ferrer-Carbonell’s paper comparing various statistical methods for analyzing happiness data. The thing that really sticks out in their comparisons is that methods that take into account individual differences, like personality, produce quite different, but more likely accurate, results. “We call for more research into the determinants of the personality traits making up these fixed effects.” Me too! It’s also fun to note that the various methods disagreed a lot about whether kids were good or bad for self-reported happiness. Almost every method shows that money is good for SWB, by the way, albeit weakly so (better than being married, or having kids, not as good as being healthy.)

I liked the concluding thought in the methodology paper:

Finally, a note on the unimportance of income for happiness. The coefficient of 0.11 of log-income in the OLS individual fixed-effect model, implies that an individual would need an income increase of over 800000% to achieve an increase of one for general satisfaction on a (0,10) scale. This in itself raises the question of why individuals expend so much effort on obtaining more income to the extent that most economists since Jevons (1871) have taken this as the main human motivation. The psychologists Brickman and Cambell (1971) long ago answered this question by proposing that humans can be on an ‘hedonic treadmill’ in which they are constantly chasing objectives that cease to be satisfying once reached. This often repeated argument would fit the finding that average satisfaction hardly increases in countries where incomes increase (Diener and Suh, 1999; Kenny, 1999), but it would seem to need a high degree of imperfect forecasting and self-delusion on the side of individuals to be true. Is there perhaps more to individual choice than happiness?

Now, as you know, I think the self-report data is quite unlikely to accurately track individual or average increases in objective well-being, so I think the correlation between the self-report and income understates the correlation between real well-being and income. That said, I like their concluding thought. Even if folks do have some problems with forecasting how they will feel, surely you’d need to be pretty silly to have the experience of something failing to satisfy you hedonically over and over again and yet keep doing it because you’re looking for hedonic satisfaction. Donald Davidson would not accept this! The best explanation is that people keep doing it because they’re looking for something else.

Can People Feel Happy and Sad at the Same Time?, by Jeff T. Larsen, A. Peter McGraw, John T. Cacioppo

ABSTRACT. The authors investigated whether people can feel happy and sad at the same time. J. A. Russell and J. M. Carroll’s (1999) circumplex model holds that happiness and sadness are polar opposites and, thus, mutually exclusive. In contrast, the evaluative space model ( J. T. Cacioppo & G. G. Berntson, 1994 ) proposes that positive and negative affect are separable and that mixed feelings of happiness and sadness can co-occur. The authors both replicated and extended past research by showing that whereas most participants surveyed in typical situations felt either happy or sad, many participants surveyed immediately after watching the film Life Is Beautiful , moving out of their dormitories, or graduating from college felt both happy and sad. Results suggest that although affective experience may typically be bipolar, the underlying processes, and occasionally the resulting experience of emotion, are better characterized as bivariate.

This is a cool paper. It suggests lots of interesting hypotheses. They mention that co-activation of positive and negative affect is likely to be dissonant, unpleasant, and short-lived. It strikes me that good literature, for example, may turn on the artful co-activation of negative and positive affect systems. Might an “interesting” life have to do with relatively often occupying complex precincts of the evaluative space? Worthwhile.

I nominate Andrew Oswald of Warwick University as the new champion of the egregiously bad happiness policy op-ed. In this Financial Times piece (unfortunately behind paywall) Oswald touches on every worn trope of politics-masquerading-as-social-science happiness studies, and in the process reveals himself to be less than honest.

Oswald writes:

The hippies, the Greens, the road protesters, the downshifters, the slow-food movement–all are having their quiet revenge. Routinely derided, the ideas of these down-to-earth philosophers are being confirmed by new statistical work by psychologists and economists.

He is clearly excited by this.

Oswald goes on unreflectively to accept the quality of the survey data, and then gives badly undersupported interpretations of that data in order to reach his desired conclusion. This is a problem given things Oswald has said elsewhere, as we shall see. For now, let me stress, as I’ve pointed out on this blog, the happiness surveys and the depression data prima facie contradict one another. Either this hasn’t occurred to Oswald, or it doesn’t bother him, for he jumps, in the usual style, straight from the survey data to the depression data:

Using more formal measures of mental health, rates of depression in countries such as the UK have increased.

As I’ve also emphasized, the depression data are so corrupted by the arbitrary boundaries of the diagnostic category with which these “formal measures” are made, and by the structure of perverse incentives encompassing pharmaceutical companies who would like to sell anti-depressant drugs, people who may or may not be depressed who would like to get them, and doctors who would like to be reimbursed by insurance companies for prescribing them. And, again, allegedly rising rates of depression do not show up in the percentages of people who report themselves in the lowest category of happiness on surveys.

Fourth, suicide statistics paint a picture that is often consistent with such patterns.

Check that sentence out. I like the “often consistent.” In the hedge there is a glimmer of conscience. But here we go anyway:

In the US, even though real income levels have risen sixfold, the per-capita suicide rate is the same as in the year 1900.

I was just looking at the suicide rates the other day. First, it is true that they are remarkably stable. The only big spike detectable in the 20th C. US data came at the Great Depression. Speaking of depression, wouldn’t you think that severe depression was a contributing cause of suicide? And so wouldn’t a stable trend in suicide be prima facie evidence against an explosion of depression? Well, you would think. And isn’t the assumption of Oswald’s proposition simply baffling? That increases in real income ought to decrease the suicide rate? Has anyone ever thought that not enough money was the main cause of suicide? Might it not be just as likely that that the stability of the trend has to do with the stability in the trend of jilted lovers, frustrated dreamers, number of people with organic mental disorders, sudden losses of status, etc. The statistic in isolation means exactly nothing. “Often consistent.” And so very often not! Why mention it at all, as if you know something? The data is also “often consistent” with the hypothesis that the suicide rate remains steady at a very low proportion of overall deaths during times of high economic growth, but spikes during times of no growth, or constriction, such as the Great Depression. I don’t know that that’s not true. Does Oswald?

Toward the end Oswald writes:

Economists faith in the value of growth is diminishing. That is a good thing and will slowly make its way down into the minds of tomorrow’s politicians.

And thus it is clear what Oswald’s aim is. I know a number of economists read this blog? Tell me, is your “faith” in the value of growth diminishing? Do you think the world would be a more happy or a less happy place if world GDP growth slowed, such that the incomes of all the billions of people living in abject poverty (for whom the effect of income on happiness appears to be profound) grew at slower rate? I wonder Benjamin Friedman thinks!

The real kicker here is that it is quite clear that Oswald knows that the survey data are too ambiguous to actually sustain the interpretation he puts on them, and therefore too weak to actually support his politics.

In this short, stimulating paper, Oswald makes an excellent point about the self-report surveys:

The key point is that we do not know the shape of the function relating ‘reported happiness’ to actual happiness. This is a serious problem when researchers try to make statements about the curvature of relationships — though not as serious when we talk, as most of the happiness literature does, about the direction of relationships.

To put this in a different way, happiness survey answers tell us which way is up or down. They do not persuasively tell us the speed of the rise or fall.

It seems reasonable — given only mild assumptions — to argue that we have established, say, that greater income buys greater happiness, ceteris paribus. But in my judgment we have not done sufficiently more than this to allow us to be confident about rates of change. [emphasis mine]

What is Oswald saying? He’s saying that the survey data are an incredibly blunt instrument, and may be useful for detecting gross relationships, but otherwise not so useful because we don’t even know if there is a lawlike relationship (the kind necessary to support valid scientific generalizations) between the way people talk about happiness and the way that we actually feel.

(Oh, and it is reasonable to say that we have established that other things equal greater income buys greater happiness. I don’t think the readers of the FT were informed of this.)

Later in the same short paper, Oswald throws out a quite plausible hypothesis about why rates of self-reported happiness might not increase, even if the objective quality of subjective experience was improving. While he doesn’t draw this out, if true, the surveys would under some conditions obscure even the direction of the relationship between talk and fact by obscuring that there is a direction of the relationship.

Imagine, for example, that there is constant marginal utility of income, but that people, as they get happier, mark themselves happier on a questionnaire scale but do so in a way in which they are intrinsically reluctant to approach the upper possible level on the questionnaire form (the 5 on a 1-5 scale, say). Then the reporting function itself is curved, and we will have the illusion that true diminishing marginal utility of income has been shown.

Which is just to say, Oswald knows perfectly well that happiness surveys may systematically fail to track increases in well-being. And so we must conclude that he is pretending in public to know things that he does not know; that he is representing to the public mere conjecture as “scientifically proven” fact in order to serve his already settled political convictions. It is no doubt inconvenient to be fully honest about your own methodological objections to data you would like to use to foist the politics of “the hippies, the Greens,” etc. upon the unwilling. But, in the end, it is also inconvenient to sully your reputation by torturing data to fit your dogmas.

This article in the NBER Digest presents a nice summary of Oswald and Branchflower’s paper Happiness and the Human Development Index: The Paradox of Australia in which they criticize the UN’s Human Development Index for failing to incorporate SWB data.

Happiness measures, Blanchflower and Oswald add, “can tell politicians and others how citizens value the different effects upon well-being of diverse influences such as unemployment, the divorce rate, real income, friendship, traffic jams, crime, health, and much else. If we can learn to exploit the power of statistical happiness equations, it should be possible to make public policy choices in a more coherent way than before.”

Branchflower and Oswald seem to make the apparently common error of seeing averages of the way lots of people filled out happiness self-report surveys with “how citizens value” this or that. It just doesn’t follow.

For example, divorce rates. Do people really value the effects of the divorce rate? (”Why so sad, Barry?” “Oh, the divorce rate went up this year.” WTF!?) What people want to know is whether their divorce (or their sister’s or friend’s) will make them and/or their kids happier. And, of course, it depends. For some, divorce is a huge relief, a new chapter, etc. For other’s it’s the end of the world. If we toted up the self-reported SWB of all divorced people to tell whether, on average, divorce makes you happier or not, we’d know almost nothing worth knowing, given the likely high variance in the sample. What’s worth knowing is whether YOU are the sort of person who will gain or lose. We couldn’t care less if the extremes of misery are more acute than the extremes of elation, producing a mildly negative average.

And, anyway, how happy a divorce is going to turn out for you just isn’t that important. Maybe it’ll make you miserable, but unless you get a divorce from the controlling creep, you’ll never become the person you need to become. Maybe the problem is precisely that you like the lack of responsibility that comes from being controlled. The fact that you’re NOT going to like leaving is why you should. Or maybe you’ll be miserable if you stay in the marriage, but you should just suck it up, because the kids need you there, and you made a promise that you should keep. Knowing the happiness averages helps us frame divorce policy how? And what does the self-reported SWB of everybody—divorced, affected by divorce, totally untouched by it—correlated to the divorce rate tell us. Probably nothing, and certainly not how “citizens value” the effects of the divorce rate. Definitely nothing that enhances our ability to make more coherent policy choices.

Will I ever tire of noting that happiness and value can come apart? Probably not. Will I ever tire of noting that self-reported well-being, which is a poor proxy for happiness, may have only an extremely tenuous relationship to value? Probably not. Will you tire of being reminded that poor proxies for things that may or may not be morally relevant in particular cases are unlikely to be extremely useful tools in our efforts to engineer a better society. Probably. But until you do . . .

This article from the Christian Science Monitor explores new census data that shows that the poor in America own computers, dishwashers, and other appliances of convenience and amusement at historic rates. Naturally, the happiness question arises, and shows how journalists have been effectively propagandized into repeating a misleading, ideological happpiness talking point whenever good economic news arrives…

. . . by almost all measures, the data show rising well-being for all of society. And while the wealth gap may not be narrowing, the rich-poor gap in lifestyles has narrowed substantially since 1992 when measured in many of these tangible items.

“In terms of the items people have … it amazes me the number of people who are at or near the poverty line that have color TVs, cable, washer, dryer, microwave,” says Michael Cosgrove, an economist at the University of Dallas in Irving, Texas. That’s not to ignore the hardships of poverty, he adds, “but the conveniences they have are in fact pretty good.”

Poor, but more comfortable
The study doesn’t explore the happiness factor — whether the growing material prosperity is actually making people feel more satisfied with their lives. While economists tend to focus on things that can be measured in dollars and cents, the spiritual side of the economy has begun to garner more attention. That’s partly because some research has found that once people gain a modest sufficiency in goods, further increases in income don’t result in rising happiness.

This happiness talking point is extremely misleading. First, “don’t result in rising happiness” obscures the fact that a large majority of poor Americans already report themselves as being pretty or very happy. I think many readers would interpet this as “doesn’t do anything to pull people up from misery or unhappiness.” It would be less misleading to say “further increases in income don’t result in happy people becoming even happier.”

Second, the talking point makes it sound as if there is some general finding that implies that a doubling of my salary would have no effect on my happiness. Which, of course, is total rubbish. We’re talking about wealth within the United States, here. So, while the correlation between average income and average happiness is weak (though positive) over time or between countries–meaning that the average happiness at any given point in the distribution over the critical absolute threshold is likely to hold pretty steady over generations or between societies–the correlation within a given country at a given time is strong, and that’s the relevant measure. Since everyone lives in a particular country at a particular time, what the research has found is that if your income increases, you are likely to get an increase in self-reported happiness. At any time and place, wealthier individuals will tend to be happier than less wealthy individuals.

Now, a rise in income sufficient only to maintain your position in the distribution will be unlikely make you much happier than you already are (unless your aspirations were low). However, through their prime working years, individual’s incomes generally rise much faster than the economy grows (my income, to take a very typical example, has increased well more than 200% since I first entered the labor market). And so you can expect your rising income to have a very positive effect on your happiness through your working years, and throughout your life, if you have invested well. And this is just what the life-cycle happiness breakdown shows. It’s our growing wealth that keeps our total satisfaction with life more or less steady as our satisfaction with health and family starts to declines in our our middle age.

Journalists need to understand that the “more money won’t make you happier” talking point is, in fact, a piece of propaganda designed to weaken public support for wealth-enhancing policies. Once the data is framed correctly, there is really no reason to use the talking points. It does nothing to “balance” the story. Even if it is true that the microwaves and dishwashers of those beneath the poverty line aren’t making them happier than they already were, they are freeing up time that would otherwise be spent on cooking and doing dishes. And that’s just good.