I’ve?written here before about my?research with?Betsey Stevenson showing that?economic development is associated with rising life satisfaction.?Some people find this result surprising, but?it’s the cleanest interpretation of the available data.? Yet over the past few days, I’ve received calls from several journalists asking whether Richard?Easterlin had somehow debunked these findings.? He tried.? But he?failed.
Rather than challenge our careful statistical tests, he’s simply offered?a new mishmash of?statistics?that appear to make things murkier.
For those of you?new?to the debate, the?story begins with?a series of papers that Richard Easterlin wrote between 1973 and 2005, claiming that economic growth is unrelated to life satisfaction.? In fact, these papers?simply?show?he?failed to definitively establish?such a relationship.? In?our 2008 Brookings Paper, Betsey and I?systematically examined all of the available happiness data, finding?that the relationship was there all along:?rising?GDP?yields rising life satisfaction.? More?recent data reinforces our findings. Subsequently, Easterlin?responded in?a pair of papers circulated in early 2009.??That’s the research?journalists are now asking me about.? But?in?a paper released?several weeks?ago,?Betsey,?Dan Sacks and?I assessed?Easterlin’s?latest claims, and found little evidence for them.
Let’s examine Easterlin’s three main claims.
1. GDP and life satisfaction rise together in the short-run, but not the long-run. False.? Here’s an illustrative graph.? We take the main international dataset – the World Values Survey – and in order to focus only on the long-run, compare the change in life satisfaction for each country from the first time it was surveyed until the last, the corresponding growth in GDP per capita.? Typically, this is a difference taken over 18 years (although it ranges from 8 to 26 years). The graph shows that long-run rises in GDP are positively associated with growth in life satisfaction.
This graph includes the latest data, and Dan generated it just for this blog post.? In fact, Easterlin was responding to our earlier work, which showed each of the comparisons one could make between various waves of this survey: Wave 1 was taken in the early ’80s; Wave 2 in?the early ’90s; Wave 3 in the mid-late ’90s; Wave 4 mostly in the early 2000s.? And in each of these comparisons, you see a positive association – sometimes statistically significant, sometimes not.
What should we conclude from this second graph?? Given the typically-significant positive slopes, you might conclude that rising GDP is associated with rising life satisfaction.? It’s also reasonable to say that these data are too noisy to be entirely convincing.? But the one thing you can’t conclude is that these data yield robust proof that long-run economic growth won’t yield rising life satisfaction.? Yet that’s what Easterlin claims.
2. The income-happiness link that we document is no longer apparent when one omits the transition economies. Also false.? One simple way to see this is to note that in the first graph the transition countries are shown in gray.? Even when you look only at the other countries, it’s hard to be convinced that economic growth and life satisfaction are unrelated.? To see the formal regressions showing this, read Table 3 of our response.? (Aside: Why eliminate these countries from the sample?)
Or we could just look to another data source which omits the transition economies.? For instance, the graph below shows the relationship between life satisfaction and GDP for the big nine European nations that were the members of the EU when the Eurobarometer survey started.?? Over the period 1973-2007, economic growth yielded higher satisfaction in?eight of these nine countries.? And while we’re puzzled by the ninth – the increasingly unhappy Belgians – we’re not going to drop them from the data! And if you think Belgium is puzzling, too, then we’ve done our job.
3. Surveys show that financial satisfaction in Latin American countries has declined as their economies have grown. Perhaps true.? But how are surveys of financial satisfaction relevant to a debate about life satisfaction?? And why focus on Latin America, rather than the whole world?? In fact, when you turn to the question we are actually debating – life satisfaction -these same surveys suggest that those Latin American countries which have had the strongest growth have seen the largest rise in life satisfaction.? This finding isn’t statistically significant, but that’s simply because there’s not a lot of data on life satisfaction in Latin America!? (Given how sparse these data are, we didn’t report them in our paper.)
What’s going on here?
Now it’s reasonable to ask how it is that others arrived at a different conclusion.? Easterlin’s Paradox is a non-finding.? His paradox simply describes the failure of some researchers (not us!) to isolate a clear relationship between GDP and life satisfaction.
But you should never confuse absence of evidence with evidence of absence.? Easterlin’s mistake is to conclude that when a correlation is statistically insignificant, it must be zero.? But if you put together a dataset with only a few countries in it – or in Easterlin’s analysis, take a dataset with lots of countries, but throw away a bunch of it, and discard inconvenient observations – then you’ll typically find statistically insignificant results.? This is even more problematic when you employ statistical techniques that don’t extract all of the information from your data.? Think about it this way: if you flip a coin three times, and it comes up heads all three times, you still don’t have much reason to think that the coin is biased. But it would be silly to say, “there’s no compelling evidence that the coin is biased, so it must be fair.” Yet that’s Easterlin’s logic.
There’s a deeper problem, too.? The results I’ve shown you are all based on analyzing data only from comparable surveys.? And when you do this, you find rising incomes associated with rising satisfaction.? Instead, Easterlin and co-authors lump together data from very different surveys, asking very different questions.? It’s not even clear how one should make comparisons between a survey (in the US) asking about happiness, a survey (in Japan) asking about “circumstances at home,” surveys of life satisfaction in Europe based on a four-point scale, and global surveys based on a ten-point scale.? Easterlin’s non-result appears only when comparing non-comparable data.
If you want to advocate against economic growth – and to argue that it won’t help even in the world’s poorest nations – then you should surely base such radical conclusions on findings rather than non-findings, and on the basis of robust evidence.
A final thought
Why not look at the levels of economic development and satisfaction?? The following graph does this, displaying amazing new data coming from the Gallup World Poll.? There’s no longer any doubt that people in richer countries report being more satisfied with their lives.
Is this relevant?? Easterlin argues it isn’t – that he’s only concerned with changes in GDP.? But the two are inextricably linked.? If rich countries are happier countries, this begs the question: How did they get that way?? We think it’s because as their economies developed, their people got more satisfied.? While we don’t have centuries’ worth of well-being data to test our conjecture, it’s hard to think of a compelling alternative.
This post was co-authored with Dan Sacks.
Update: The media reports about this paper are citing a paper published this week in PNAS. I actually linked to earlier work instead, because of a press embargo on that paper. But the PNAS paper just re-hashes results from the (non-embargoed) Easterlin papers I linked to and commented on, above.
and 
. You can also watch it on 
I may have missed it but where in the data does infer the cause of the correlation? That a higher GDP results in a greater level of statistfaction in the culture or is it as a country is more satisfied it’s more productive. I’m sure it’s a spiral that one influences the other but it’s hard to trace out a cause.
What is also found is that a higher GDP, while correlated with greater levels of self-reported happiness, is not a prerequistite for high happiness levels – eg, Argentina in the fist graph.
Further, GDP is a very rough measure. There is a massive chance of ommitted variable bias in such analysis.
Wouldn’t a better analysis take a wide selection of potential variables that influence happiness and use some statistical test to see which are most significant. There are numerous ways to approach this. If it has been done, I would be interested in a link to the paper.
To make any visual inference you need to present 99% CI error bars around each point in both the Per Capita GDP dimension, and the well-being index dimension. Without that data it is impossible to calibrate the systematics in the data.
The SE you have presented is the statistical error regression, not the range of systematics in the dataset.
There is a fundamental mathematical flaw in your analysis.
Through the Cauchy-Schwarz Inequality and the Riesz Representation Theorem, the average probit satisfaction index is maximized when the responses are distributed closest to the first derivative of normal distribution.
By calculating the average of the probit scores it can be mathematically proven that this is equivalent to taking the L2 inner product between the first derivative of the normal distribution and the empirically observed percentiles.
west gerrmany and ireland appear to have close to no correlation between happiness and economic growth. in the bottom chart a number of south american economies (e.g. venzuela) appear to be far happier than they should be based on the regression…..segmenting your data by region of the globe may actually provide better correlation than the economic development. in your first set of charts i believe you go draw a line with 0 slope though almost every one of them and be as statistically significant.
i have no idea if easterlin is right, not having read his material but your conclusions appear somewhat dubious given how you are presenting the data. you lead credence to anyone disputing your theory by not highlighting issues in your own analysis and instead call it “robust”.
I’m still confused by how we can correlate GDP, which is essentially measured on an infinite objective scale, with “happiness” or “subjective well-being”, which is measured on a finite (e.g. 7-point scale) subjective scale. This doesn’t make sense to me. Until we can find a “happiness” assessment that can increase limitlessly, like GDP, and can be assessed objectively, like GDP, then I’m still unconvinced.
Justin Wolfers’ column dismisses Richard Easterlin’s work as just plain wrong. I argue here, as I have elsewhere, that where you come out on the Easterlin paradox depends on the happiness question (and therefore the definition of happiness) that you use, as well as the sample of countries and the period of time.
Richard Easterlin finds no clear country-by-country relationship between average per capita GDP and life satisfaction (among wealthy countries), despite a clear relationship between income and happiness at the individual level within countries. Easterlin also found – and continues to find, based on methods different from Wolfers’ – an absence of a relationship between life satisfaction and long-term changes in GDP per capita.
Different well-being questions measure different dimensions of “happiness”, and, in turn, they correlate differently with income (something they themselves show at the end of their last paper, and admit that the relationship between income and well-being is complex). The best possible life question – which the authors primarily use in the first (2008) work, and also in the second – asks respondents to compare their life today to the best possible life they can imagine for themselves. This introduces a relative component, and, not surprisingly, the question correlates most closely with income of all of the available subjective well-being questions. Life satisfaction, which they use in the second work, also correlates with income more than open ended happiness, life purpose, or affect questions, but not as closely as the best possible life question.
In an earlier work, Justin Wolfers, with Betsey Stevenson, (2008) – used the most recent and extensive sample of countries available – from the Gallup World Poll, and, as the measure of “happiness”, the best possible life question therein, and challenged the Easterlin paradox. In more recent work, with Stevenson and Dan Sacks (2010), referenced in this blog, the authors look at the relationship between life satisfaction and economic growth, based on the World Values survey and GDP levels and the best possible life question, based on the Gallup World Poll. They isolate a clear relationship between life satisfaction and GDP levels, and their statistical analysis is spot on.
Recent studies by Kahneman and Deaton (2010), and Diener and colleagues (2010), for example, find that happiness in a life evaluation sense (as measured by the best possible life question) correlates much more closely with income than does happiness in a life experience sense (as measured by affect or more open ended happiness questions). This holds within the United States (Kahneman and Deaton) and across countries (Diener et al.).
My own work on Latin America, with Soumya Chattopadhyay and Mario Picon, tested various questions against each other and finds a similar difference in correlation, with affect and life purpose questions having the least correlation with income and the best possible life question the most. My work on happiness in Afghanistan (2009) found that Afghans were happier than the world average (on par with Latin Americans) as measured by an open ended happiness question, and 20 percent more likely to smile in a day than Cubans. Yet they scored much lower than the world average on the best possible life question. This is not a surprise. While naturally cheerful and able to make the best of their lot, the Afghans also know that the best possible life is outside Afghanistan.
Thus the conclusions one draws on whether there is an Easterlin paradox or not in part rest on the definition of happiness, and therefore the question that is used as the basis of analysis. Wolfers and co-authors find a clear relationship between GDP levels and life satisfaction and best possible life – clearly important dimensions of well-being. Yet in the same paper they find much less clear relationships when they use happiness, affect, and life purpose questions.
There is also the question of the sample of countries, and whether one is examining cross section or time series data. The most recent debate with Easterlin is about the trends over time rather than cross-sectional patterns. Dropping the transition economies, as Easterlin does, may be a mistake, as Wolfers contends. But it is also important to recognize the extent to which including a large sample of countries that experienced unprecedented economic collapse and associated drops in happiness alters the slope in the cross-country income-happiness relationship (making it steeper). Wolfers also criticizes Easterlin for relying on financial satisfaction data for his Latin American time series sample (because there is not enough life satisfaction data); financial satisfaction correlates closely, but not perfectly, with life satisfaction. Easterlin’s technique allows for the inclusion of a much larger sample of middle income developing countries, a sample of countries that one can imagine is very important to the growth and happiness debate. Wolfers and co-authors use far fewer Latin American countries because comparable life satisfaction data is limited. Either approach is plausible and, as with all work with limited data, is not perfect. But I would not go as far as calling one or the other “plain wrong”.
Finally, there is the simpler question of giving credit where credit is due. We would not be having this debate, nor would we have a host of analysis on well-being beyond what is measured by income, had Easterlin not triggered our thinking on this with his original study of happiness and income over three decades ago (and his patient and thoughtful mentoring of many economists since then). In the big picture of things, Easterlin had the idea.
Your last statement “We think it’s because as their economies developed, their people got more satisfied. While we don’t have centuries’ worth of well-being data to test our conjecture, it’s hard to think of a compelling alternative.” – Isn’t that precisely what you denounce and accuse Easterlin of doing – a non-finding? You state to “never confuse absence of evidence with evidence of absence.”. But you end your entire analysis with an overall “non-finding” or stating that because there is an absence of evidence, that there is evidence of absence.