What is the relationship between education and wellbeing, and how does it vary across demographic groups? Marco Felici and Matthew Agarwala explore competing theories and expose the dangers of relying on average effects in wellbeing policy.
What do we know about the role of education in supporting wellbeing?
The question of ‘What drives human wellbeing?’ has long interested economists, psychologists, and philosophers. Some findings are unsurprising – having a partner, good social relationships, freedom from financial stress, and being healthy and employed are all squarely in the ‘good’ column. Commuting, loneliness, and being in your mid-life years (the crisis is real) are consistently associated with lower wellbeing. One of the more unsettling findings is that education appears to be a poor predictor of human wellbeing. The results are mixed. Studies have found everything from negative, to null, to positive effects of education on wellbeing. But given its role in people’s lives, careers, and myriad government initiatives – ‘education’ makes 139 appearances in the UK’s Levelling Up White Paper – understanding how it affects wellbeing is more important than ever.
As part of the Bennett Institute’s Many Dimensions of Wellbeing Project, we explored two competing theories on the relationship between education and life satisfaction. In one, education acts as a buffer, providing those with relatively low life satisfaction with some degree of protection from life’s challenges (bad things might happen, but at least those with a degree may be more likely to have a stable job and income). At the other end of the scale, for those with comparatively high life satisfaction, we frame education as a positional good: its value depends on how much of the good others also possess. Being the only person with a degree places a high value on degree-level education, but this value declines, in relative terms at least, as more and more people gain degree-level education. To compound this effect, individuals with higher educational attainment tend to concentrate together, most notably in urban areas. Self-selection into an environment that rewards and attracts those with degrees might fuel a phenomenon termed frustrated achievement (Graham and Pettinato, 2002), whereby those that are most upwardly mobile are also the most negative when it comes to self-assessment.
There are many channels through which education might affect wellbeing – directly, if the experience of earning and holding a degree generates wellbeing regardless of other outcomes; and indirectly, if it improves income, status, and job prospects, and if these things in turn boost wellbeing. There are also concerns over the appropriate scale of analysis. Should we focus on how an individual’s education affects their own wellbeing, or are we more interested in how a community’s education level affects overall wellbeing across society? For instance, a report by the What Works Centre for Wellbeing found that on average, in most local authorities, people with lower levels of education had lower wellbeing than those with higher education – e.g. in Blaneau Gwent and Sunderland. However, in some local authorities there was no difference at all, and in others, people with lower levels of education actually had higher wellbeing – e.g. in Waltham Forest and the Scottish islands of Eilean Siar, Orkney & Shetland.
What we did
We wanted to explore whether the relationship between education and wellbeing was different for different groups of people. Specifically, we were interested in how having a degree affected life satisfaction for those who report low versus high levels of life satisfaction. To this end, we analysed data from the Community Life Survey (CLS; waves 1-5, with 19,494 observations in England from 2012 – 2017) and the combined British Household Panel Survey and Understanding Society Survey (BHPS/US; 28 waves, with over 525,000 observations and approximately 90,000 unique individuals across the UK from 1991 – December 2019). The relationship of interest is the effect of holding a degree on life satisfaction, controlling for age, gender, marital status, social capital, health, year, and interview mode (online or face-to-face).
Our empirical strategy incorporated a range of estimation procedures including both cross-sectional and panel analysis. Across all specifications, results were qualitatively consistent, though there is some variation in the effect size and statistical significance. In this blog, we will focus on the quantile regressions and event study analysis.
Quantile regression enables us to investigate the relationship between education and life satisfaction along the full distribution of the life satisfaction scale. Figure 1 shows coefficients on the degree dummy (a variable describing whether or not the person had a degree) at the 25th, 40th, 50th, 60th and 75th percentiles of the life satisfaction distribution, using the CLS. At the low-end of the life satisfaction distribution (the left-hand side), and specifically at the 25th percentile, the coefficient on the degree dummy is positive and significant, with a point estimate around 0.2. This indicates that for those with low life satisfaction, those with a degree are likely to report slightly higher satisfaction than their peers who do not have a degree. In the middle of the distribution (between the 40th and 60th percentiles) the association is undistinguishable from 0, and even turns negative at a point estimate of about -0.1 for the 75th percentile.
Figure 2 shows consistent results using the much larger combined BHPS/US dataset. Once again, we see that at the lower end of the distribution, having a degree is statistically significantly associated with higher life satisfaction. The magnitude of the association falls as we move towards the middle of the distribution, ultimately becoming statistically indistinguishable from zero. At the high end of the distribution, the association of life satisfaction with having a degree turns negative.
Finally, we exploited the fact that our data followed the same individuals over many years, including before and after they earned a degree. This showed that completing a degree-level education is associated with different trajectories of life satisfaction over time, and that these also depend on the initial level of wellbeing.
In the graphs below, point 0 represents the moment of getting a university degree. The red line represents the difference in life satisfaction between those who will obtain a degree and a “counterfactual” level of life satisfaction, prior to obtaining a degree, while the blue line represents the same difference after obtaining the degree.
The red line suggests that the difference in life satisfaction prior to getting a degree is fairly constant (the coefficients on pre-trends are not statistically distinguishable from 0). The blue line suggests that after the degree, life satisfaction first falls but then increases in relation to those who didn’t get the degree.
What happens when we look separately at those who were very satisfied with their lives before age 23 and at those who were rather unsatisfied? For those who start with higher life satisfaction, the post-degree drop lasts longer and even in later years there is no demonstrable positive effect (Figure 4). In contrast, the drop is shorter-lived for those that start with low life satisfaction and follows a more consistent upwards trajectory in later years (Figure 5). It is important to mention that attributing these dynamics to the achievement of degree level education relies on a specific set of model assumptions, outlined in detail in our working paper.
The problem with fighter jets
Perhaps the biggest message our work has revealed is that wellbeing research and policy cannot rely solely on evidence evaluated at the mean (average). Our results point to the possibility that the relationship between key indicators – education and life satisfaction – is not constant. Evaluation at the average can mask critical insights and important nuance at the tails of the distribution. Understanding this will be key to levelling up.
Although ours is the first research to investigate this relationship on such a large dataset within the UK, using a combination of cross-sectional, panel, and event study analyses, the danger of focusing on statistical averages is not a new phenomenon.
Back in the 1940s, the US Air Force faced a crisis: pilots were crashing. As many as 17 in a single day, often in routine training exercises. It wasn’t pilot error, and it wasn’t mechanical failure. The problem was a lethal failure to understand the consequences of averaging important data. The cockpits themselves had been designed to fit the ‘average pilot,’ determined by taking the average of over 140 dimensions of size, “including thumb length, crotch height, and the distance from a pilot’s eye to his [only men in those days] ear”. But, as Harvard physical anthropologist Lt. Gilbert S. Daniels pointed out, no one is average. To prove this, he gathered data on 4,063 pilots and calculated the average of 10 key dimensions of size (height, chest circumference, sleeve length, etc). He then counted how many pilots’ measurements fell within 30% of the average across all 10 dimensions. The answer? Zero. Essentially, by designing for the average, they had designed for no one.
Our investigations of the relationship between education and wellbeing demonstrate a similar concern: when the evidence base is evaluated at the mean, it may not be representative of anyone. Researchers and policymakers have an obligation to investigate the tails of the distribution. Doing so can reveal unique relationships and support more targeted policy interventions. Ultimately, the US Air Force solved its problem by creating the adjustable seat, just like the one you may very well be sitting in now. Adjustable solutions that cater to the unique needs of end users can save lives. Education, wellbeing, and levelling up policy should take note.