David Leonhardt's column yesterday discusses a new paper by Wolpers and Stevenson that claims to debunk the Easterlin paradox. The paradox was that despite an association between income and reported happiness in cross-sectional data, in the time series big increases in income did not increase average reported happiness. An obvious interpretation is that people care about relative, not absolute, income. I have to get the paper, but from Leonhardt's account, I wonder if it answers what to me is the larger point of Easterlin and Easterlin-spawned work: that relative income matters.
The column includes a graph that shows a positive correlation between income and happiness across countries. What I want to know - I have to get the paper - is , granting that the correlation is positive for a particular country's time series, not zero, as Easterlin found, whether the correlation in a cross-section is stronger than in the time series - whether relative income matters at all. The idea that only relative income matters, that absolute income matters not at all, was never very plausible.
Got to go: I'm bitter and angry and dodging sniper fire, can't seem to find my flag lapel pin, trying to find a weatherman to see which way the wind blows, and damn if isn't Spring in poor old Northwest Ohio!
The Easterlin paradox itself and the Wolfers and Stevenson's reply to it are good examples of the trivialization of analysis. The "happiness data" in question descends historically from market research customer satisfaction surveys by way of Paul Lazerfeld (with an intellectual nod to Max Weber). The surveys contain a battery of questions and the interesting results come from correlating responses to the different questions.
ReplyDeleteIt may be intriguing (but it's analytically pretty jejune) to try to go from those internal survey comparisons to comparing some aggregate happiness assessment to a per-capita income number. At that level it's all interpretation and very little substance.
It's not just that "correlation doesn't imply causation" (although that's worth keeping in mind). It's also that there's no obvious follow-up question or procedure so the "next step" in any argument based on happiness surveys is to jump to some unconnected anecdotes or data sets. How one makes that leap depends mainly on the prior ideological commitment of the researcher (or the researcher's projected audience).
Ultimately, the happiness/income question -- framed in terms of what "the data tells us" -- initiates an interminably inconclusive discussion that has no policy bite. On the other hand, an expansive historical investigation of the sociological precedents for happiness surveying would take us right to the heart of some very meaty policy questions (Max Weber on the psychophysics of industrial work, Karl Bucher on work and rhythm, Paul Lazerfeld and colleagues on unemployment and social distress). But nobody goes there. Why? I suspect there's no career-building traction in looking at aspects of questions that expose the sheer distractedness of so much academic wool-gathering.
Stevenson and Wolpers have to strain to find countries with risign happiness and rising income, and some of them are odd examples. Does not hold for the US, where measured happiness peaked back in 1956, or Japan, the original example of Easterlin's. The cross-section relation is much stronger than the time-series, where Easterlin's finding continues to hold for lots of countries.
ReplyDeleteThere is a curious thing about the country cross-sections. Given the little change over time in individual countries, this means the international pattern has not changed much. With a few exceptional outliers like Central America (poor but happy), richer countries tended to be happier decades ago and still are. There is an obvious interpretation of this that fully supports Easterlin. What we are seeing is an international version of Easterlin's relative income effect. People in poorer countries know that they are poorer than people in richer countries. Also, richer countries often dominate and humiliate poorer ones. There are likely to be issues of national pride or lack thereof going on here as well.
Frankly, the Leonhardt piece is ridiculous, trying to puff something up that is not there at all.
Barkley
Barkley, the idea that the international cross-section shows EP with respect to countries occured to me, too. Seems plausible.
ReplyDeleteSandwichman, are you saying that the whole happiness/income discourse is worthless for policy or only that the particular discourse based on self-reported happiness data is worthless? I have a lot of respect for people like Easterlin, Frank and Layard - is their work "distracted academic wool-gathering," do you think?
No, I'm not saying it's all worthless. I'm saying that it veers off on a trivial digression just when it could instead lead to fruitful lines of inquiry. Sort of like the Monty Python argument clinic or Harry Nilsson's song, "Joy". Good, bad, good, bad, good, bad.
ReplyDeleteI have to confess to having plunged into the happiness/income discourse with high expectations. Some of the contributions show a lot of promise. Even the self-reported happiness data could be a real eye-opener if only investigators would do a bit more conceptual archeology -- nay, a lot more archeology -- before pulling out their pocket calculators and running regressions.
That said, there is substantive analysis being done that touches on the happiness/income question. It is annoying -- but not surprising -- that what gets media attention is the man-bites-dog statistical artefacts.
wharton school i think employs j scott armstrong, who is a 'forecasting' expert (who doesn't seem to have forecasted anything i can see) who has disproven the theory of global warming, using his forecasting questionaire. Hansen, etc asked the wrong questions. All that wasted NOAA money.
ReplyDeleteso in that context, this paper looks quite good. i note the NYT column has 400 plus comments; perhaps happiness is correlated with blog hits or comments? Maybe the NSF and Pew will fund some international surveys on this important question.
the polls i think are suggestive but not totally reliable, and also culture-bound. the paper is standard style of rigor, what is called 'obfuscationism' by experts. I guess i didnt see the time series stuff well presented. They do show the cross section, but who is going to read a list of regression coefficients?
also, there are very simple and related common sense questions that could be asked which they don't do. And other interpretations. Hence, it is 'publish or perish' fodder; just another half baked pile of paper. But that is the classical approach.
regarding relative income, they could look at cross sections for ginis, and also time series. some of this has been done. alot of it is ambiguous, because of the interpretations. why don't they do this? for the sasme reason noone has heard of Minsky. If you are doing the GIGO thing you really don't have time to deal with ideas. also, bad ideas drive out good ones, and that is part of Wharton's job.
Having seen the paper, I agree for the most part with Media. The time series stuff isn't well presented. But here is something that jumped out at me. They compare cross country to within-country cross-sections, with the former meant to tell us about the relationship between happiness and absolute income(Barkley has already noted the dubiety of this). Then they look at the coefficient on income in both regressions and claim that they are more or less the same, proving that relative income doesn't matter. In fact the coefficient in the second, intra-country regression is initially less than the first, just the opposite of what one would expect if relative, as well as absolute, income matters. But they admit that it must be adjusted up to get the relationship between permanent income and happiness. At first they use %.55 as the effect on permanent income of a $1 increase in measured income. With this assumption the coefficients shopuld be increase by 80%, putting them about 37% above the coefficient from the cross-country regressions, it would appear. That is in passing,they then go on to argue that with less-than-complete smoothing the upward adjustment shoud be in the neigherhood of 30% making them more or less the same (conveniently) as the cross-country.
ReplyDeleteThe paper seems to have lots more to say about how the within-country cross-section coeffs. are over-estimates; almost no thinking about about how they could have been under-estimates- with the one exception just noted, which is pretty much rejected.
I'd appreciate hearing what others make of it.
'i dream of gini' manifesto.
ReplyDeletei glanced at this again. (i won't print out 71 pages). the scatter on the plots seems pretty extreme in many cases; i assume they are using a least squares fit (perhaps i'm wrong and they are doing something else)to find a trend, but like in behavior genetics the scatter may actually reveaL more than the trend (eg the same gini can occur for many income distributions). the polls appear to be fairly consistant.
also the time series are erratic---it suggests maybe happiness is endogenous; possibly manic depression. or, simply sketchy polling.
eastern europe's breakup also might need to be dissagregated from western europe, not to mention places like congo or afghanistan.
one graph does show for eurobarometer gdp correlates with happiness, but only for the first half of the series; later it seems to slow down in increase. Germany stays flat; belgium actually gets unhappier with wealth, is what it appears to me. (perhaps the french/flemish split was kicking in.) this may be related to the details presented above regarding intra v cross country correlations of income/happiness. the main thing it seems to me is the time series appear to have breaks so linear regression/least squares is 'wack' (with log smoothing excaberbating the problem).
only when they aggregate do you see any trend. this is what is termed in the litterture as 'data mining' (eg the big bang theory---though perhaps there they have really looked over the data).
it is interesting that US GDP/per capita increases, while average household income stays flat. that almost looks like an error---or maybe that says households are getting smaller, due to the influence of Bill Cosby and other gangsta rappers. if so, then this would explain the stasis of the usa, but not elsewhere, for their argument.
of course the other explanation is because gini has been increasing this would explain it. the gini might be explained by household size too (as is done----all those welfare queens hoarding wealth) or household size might be explained by gini (who wants to pay child support, after all?)
usa has a high gini but high happiness supposedly. (perhaps the unhappy ones self-select for homicide or suicide to keep the numbers looking good. maybe tax breaks could be given for this.) canada is there too, whose gini i forget (slightly lower than the us likely); and then there is uk.
i wonder about places like costa rica, sweden, etc.
the absolute approach would suggest the 'poor' in the usa might be happier than the 'rich' in a poor country (using PPP). so one could forget about countries altogether and just deal with real $.
there do appear to be alot of papers on gini/happiness which support easerlin (who is known on the east coast as easterly due to a gauge transformation).
in sum, this paper mostly seems to be propoganda for the wharton growth agenda. it seems 50% of 'economics' is this---because i can plot a straight line, i have shown more is better than you. I imagine wharton is fully funded by businesses, who produce tax shelters and other goods. (A canadian economist recently showed that goods produce happiness too. If only AK's can be sold in walmart, everyone wiould be happier.)