Does Growth make us Happier? A New Look at the Easterlin Paradox

Does Growth make us Happier? A New Look at the Easterlin Paradox Felix FitzRoy School of Economics and Finance University of St Andrews St Andrews, KY16 8QX, UK Michael Nolan* Centre for Economic Policy University of Hull Hull, HU6 7RX, UK Abstract June 2014 Consistent with the Easterlin Paradox commonly found in macro data, raw BHPS data on average Life Satisfaction (LS) show no evidence of an upward trend over 10 years of substantial economic growth prior to the crash of 2008, neither for the full sample nor for regional subsamples (except obviously exceptional London). When the sample is split by education into three groups (high, middle and low) the (smallest) sub-sample with high education attainment shows remarkable growth in average LS, from lowest to highest of the three groups. Pfaff and Hirata (2013) estimate LS with the micro data in the BHPS and other panels, and find no effect of regional gross value added on LS, using many controls for individual characteristics, including own household income. This is perhaps not surprising since GVA (or GDP) growth should be reflected in household incomes over time. The lack of any extra influence of GVA is certainly consistent with Easterlin. Possible explanations of the paradox include the negative effect of peer group income, and declining social capital. Whereas Pfaff and Hirata (2013) use a quadratic in age, which misses declining LS of the elderly, we utilise age dummies and find positive and significant effects of regional GVA in various specifications, even when controlling for own income. This suggests a positive externality from economic growth, or perhaps public-good expenditure which is not included in private income. Given a lack of LS growth in the aggregate, this is not what Easterlin s results lead us to expect. Comparison income also has positive instead of the expected negative effect, so lack of trend growth in average LS for the whole sample is even more surprising, and is presumably due to strong offsetting factors, such as declining social capital, though clarifying this is beyond our present scope. JEL classifications: I31, O47. Keywords: Life satisfaction, Easterlin paradox. * Corresponding author: Michael A Nolan, EMAIL: M.A.Nolan@hull.ac.uk

1. Introduction and Background Easterlin s (1974) seminal paper found no correlation between long term economic growth in rich countries, and subjective well-being (SWB) life satisfaction (LS) or happiness. With 40 years of additional data, and economic growth, there is little evidence of any generally increasing trend in SWB, (even in some of the fastest growing developing countries such China). However there is a strong cyclical relationship between real GDP per capita and SWB, with unemployment being a major cause of unhappiness that moves with the cycle, and critics have usually failed to distinguish carefully between trend growth and deviations from the trend (Easterlin, 2013). The paradox is deepened by the fact that richer people are generally happier than the poor in any one country, though many other factors such as health, family and employment are more important than income (but often also correlated with income). The well-established importance of socio-economic status or relative income is often advanced as an explanation, but studies using only macro-data on average happiness and per capita GDP obviously cannot explore this factor, while also omitting numerous important individual variables such as health, age, education and others, which do actually change in the aggregate over time. Another possible factor that could offset the benefits of growing real incomes is the widely observed decline in many components of social capital, such as community, personal and family relationships, although these issues are beyond the scope of this paper. It thus seems appropriate to use available large panel data sets, which follow individuals over time, to examine the effects of economic growth on their well-being, while controlling for both individual fixed effects and changing characteristics recorded in the survey data. A pioneering attempt by Pfaff and Hirata (2013), using German and UK panels, finds no effect of regional per capita output or growth rates on individual life satisfaction in the presence of numerous controls, including individual fixed effects. One problem is that they control for

age with the commonly-used quadratic form, but this misses the sharp downturn in LS in old age. They find the usual strong effect of households own income, on LS, though on average household income will reflect aggregate or regional growth. They conclude that Easterlin s hypothesis cannot be rejected. Here we find significant differences in the results when we use unconstrained individual age dummies instead of a quadratic in age. We also test directly for the influence of relative income, defined here as the ratio of own income to the average income of a peer group, consisting of individuals of similar age, education and region. This ratio should exclude the effects of economic growth on all individual incomes, so allowing for maximal effect of growth proxies. Alternatively, we include own income and comparison (or peer group) income separately. 2. Data and methodology Our main data 1 are taken from Waves 6-10 and 12-18 of the British Household Panel Survey, (BHPS), covering a period that runs from 1996/07 to 2008/09. We use data for 122,366 observations across 21,759 individuals, with those cases where there are missing values excluded. As usual, we note the deliberate over-sampling of the smaller nations of the UK since Wave 9 so that about half of the individuals in the BHPS are from Scotland, Wales and Northern Ireland, compared to less than 20% in the overall population. Regional Gross Value Added (GVA) data from the Office for National Statistics are also used, mainly at the NUTS2 level that splits the UK into 13 main regions 2 (10 in England, plus Scotland, Wales and Northern Ireland). As noted by Pfaff and Hirata (2013), BHPS 1 The earlier waves of the BHPS (up to Wave 10) were limited in coverage to Great Britain. The full United Kingdom (including Northern Ireland) is covered in Waves 12-18. BHPS data are available via the UK Data Service (formerly the UK Data Archive). 2 A tiny number of observations for the Channel Islands are excluded.

interviews for a given wave actually spill across two different calendar years: since GVA data refer to a particular calendar year, it would seem appropriate to consider GVA data with a one period lag (otherwise it might refer to a year that partly pre-dates the BHPS interview). In fact, this turns out to be somewhat problematic for Wave 6 (where most interviews took place in 1997) because there is a break in the GVA data series between 1996 and 1997 (so that strictly comparable GVA data for 1996 are unavailable). Our chosen measure of GVA per capita is deflated simply by way of the Consumer Prices Index (CPI) for all items (series D7BT). In the Appendix, Figure 1 demonstrates the lack of an obvious time trend in life satisfaction across the BHPS sample (over the period when life satisfaction is included). Our sample is limited to Waves 6-10 and 12-18, because of the absence of life satisfaction from the dataset outside those waves. A more detailed (region by region) plot could offer limited hints at some trends with London, Scotland and Wales as candidates for an upward trend. However, in the main, such a plot suffers from difficulties in readability to the extent that it has been omitted. Meanwhile, Figure 2 shows the expected initial upward trends in real regional GVA per capita until a downturn in 2007/08 that is in keeping with the start of events that triggered the Great Recession. The clear supremacy of Greater London on this GVA per capita measure is also totally unsurprising (as is the presence of SE England as runner up). A plausible hypothesis is that those with higher education, who generally have the best-paid and most interesting jobs, would be most likely to enjoy increasing life satisfaction with higher incomes, so we split the sample into three groups with higher (ISCED categories 5a and 6), middle (ISCED categories 3a and 5b) and low (ISCED categories primary, secondary and 3c) education. Raw plots of LS by education category are shown in Appendix 3, ignoring a relatively small number of observations from the later months of Wave 6 interviewing. The

most obvious trend is that the highly educated (top 13% or so) start with lower than average LS, but this changes to the extent that they have the highest LS by Wave 18. Our estimation approach is quite similar to that of Pfaff and Hirata (2013) certainly in that we use Fixed Effects estimation of a Life Satisfaction equation with quite a number of controls many of which would be fairly standard when using BHPS data. These include marital status (including cohabiting), number of children, health status, education, labour market status, time spent in panel, whether year of last interview, log household size, age (full set of dummies), wave number and regions. Like Pfaff and Hirata (2013), we also follow Moulton (1990) in recognising the potential (cluster-related) effect of aggregate regressors on standard errors. 3. Results and discussion Results are shown in the Appendix. See Tables 1a, 1b and 2. Only columns 1 and 3 in Table 2 use a quadratic specification for age (linear and quadratic terms), whereas all other results use a full set of age dummies. Column 4 in Table 2 is intended primarily as a comparator to column 1 in Table 1b the significance of the difference term (to capture the growth in log deflated GVA per capita), and its negative sign, place it close to its unconstrained counterpart from column in Table 1b (where the lag of log deflated GVA per capita has a positive sign, similar size and slightly more statistical significance). For the crucial test of the effects of economic growth, we calculate both trend growth in regional output (gross value added in our BHPS panel for the UK), and deviations from the trend, and include the trend to directly test Easterlin s hypothesis, and also deviations to capture cyclical effects, particularly on employment and job-security. We do also control for

unemployment, which is of course not exclusively cyclical. As an alternative, we also test the effects of the lagged level of (log) real regional GVA per capita, and the growth of this variable. The results contain several surprises. Although average LS for the whole BHPS sample shows no clear upward trend over a period when real per capita GDP grew by around 30 percent, estimation with a large set of controls shows that the lagged trend value of real regional GVA has a positive effect on LS, albeit only significant at the 10% level. Lagged deviations from this trend also have a positive effect on LS: the point estimate is about twice as large, although it is again only significant at the 10% level. Since we control for relative income and many other factors, this result is surprising it suggests there must be other, offsetting factors that prevent average LS from increasing over the period. It also differs from Pfaff and Hirata (2013), and in contrast to their results, does not appear to support the Easterlin Paradox. In the subsamples by level of education we find more surprises in the high education group. None of relative income, GVA trend or deviations from trend were significant. In the medium education sample, both relative income and the GVA trend were significant and positive, but deviations from trend were insignificant. Only in the low education group, were deviations from trend GVA significant. This is as expected because this group has the highest unemployment and lowest job-security, both influenced by cyclical movements of the economy. However, trend GVA was insignificant, and relative income only borderline significant. In other specifications including own income, or both own income and comparison income separately, the former is as usual strongly significant and positive. However, and in contrast to Pfaff and Hirata (2013), a positive and significant effect of trend GVA growth, or its

lagged level, remains robust in all versions. Thus in one sense we seem to have gone beyond just rejecting Easterlin by finding an additional positive effect of growth, even when controlling for rising household incomes. On the hand, since average LS does not increase over our time period in accord with Easterlin, our results suggest that the offsetting negative effects of economic growth or associated social change are even more important than hitherto suspected. Exploring these factors remains an important topic for future research. References Easterlin, R.A., 1974. Does Economic Growth Improve the Human Lot? In: David, P.A., Reder, M.W. (eds.), Nations and Households in Economic Growth: Essays in Honour of Moses Abramowitz. Academic Press, New York, pp. 89-125. Easterlin, R.A., 2013. Happiness and Economic Growth: The Evidence. IZA DP No. 7187. Moulton, B. R., 1990. An Illustration of a Pitfall in Estimating the Effects of Aggregate Variables on Micro Units. Review of Economics and Statistics 72(2), 334-338. Pfaff, T., Hirata, J., 2013. Testing the Easterlin hypothesis with panel data: The dynamic relationship between life satisfaction and economic growth in Germany and in the UK. Centre for Interdisciplinary Economics, Working Paper 4/2013.

Appendix 5.50 Figure 1: Life Satisfaction, BHPS, Waves 6-10 & 12-18. 5.45 5.40 5.35 5.30 5.25 GB, then UK 5.20 5.15 5.10 5.05 5.00 6 7 8 9 10 11 12 13 14 15 16 17 18

35,000 Figure 2: Regional Gross Value Added per Capita, deflated by CPI, 1997-2009. 30,000 25,000 20,000 15,000 NE England NW England YorksHum East Midlands West Midlands East of England London SE England SW England Wales Scotland N Ireland 10,000 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

5.40 Figure 3: Life Satisfaction by education, BHPS, Waves 7-10 & 12-18. 5.35 5.30 5.25 Low Education Middle Education High Education All Education Levels 5.20 5.15 5.10 7 8 9 10 11 12 13 14 15 16 17 18

Table 1a (1) (2) (3) (4) UK, BHPS, Waves 6-10, 12-18 all Low Education Medium Education High Education Fixed Effects Household income 0.034*** 0.033*** 0.038*** 0.016 (4.99) (3.09) (3.78) (0.92) Comparison income 0.104 0.096 0.104 0.197 (1.63) (0.84) (1.01) (1.16) Lagged trend, log GVA per capita (deflated) 0.159* 0.130 0.394*** 0.017 (1.80) (0.63) (2.68) (0.13) Lagged deviation from trend, deflated log GVA/capita 0.162 0.479* -0.136-0.050 (0.85) (1.75) (-0.42) (-0.11) Observations 122,366 68,366 37,876 16,124 Number of persons 21,759 13,016 7,081 2,869 Adj. R-squared 0.038 0.039 0.045 0.050 Dependent variable: Life-Satisfaction. Controls for marital status (including cohabiting), number of children, health status, education, labour market status, time in panel, year of last interview, household size, full set of age dummies, wave number and regions are included. Standard errors clustered at the individual level, robust t-statistics in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Table 1b (1) (2) (3) (4) UK, BHPS, Waves 6-10, 12-18 all Low Education Medium Education High Education Fixed Effects Household income 0.034*** 0.033*** 0.038*** 0.016 (4.99) (3.08) (3.77) (0.92) Comparison income 0.104 0.095 0.101 0.196 (1.63) (0.84) (0.98) (1.16) Lagged log of GVA per capita (deflated) 0.160** 0.238 0.326** 0.013 (1.96) (1.42) (2.40) (0.10) Observations 122,366 68,366 37,876 16,124 Number of persons 21,759 13,016 7,081 2,869 Adj. R-squared 0.038 0.039 0.045 0.050

Table 2 (1) (2) (3) (4) UK, BHPS, Waves 6-10, 12-18 all, Quadratic age all, income ratio all, income ratio and Quadratic age all, using difference (GVA growth) Fixed Effects Household income 0.033*** 0.034*** (4.99) (5.00) Comparison income -0.204*** 0.104 (-4.61) (1.62) Income ratio 0.019*** 0.022*** (2.80) 3.33 Lagged trend, log GVA per capita (deflated) 0.155* 0.159* 0.157* (1.75) (1.73) (1.76) Lagged deviation from trend, deflated log GVA/capita 0.195 0.163 0.185 (1.02) (0.85) (0.97) Difference in log GVA per capita (deflated) -0.138* (-1.78) Observations 122,366 122,366 122,366 122,366 Number of persons 21,759 21,759 21,759 21,759 Adj. R-squared 0.036 0.038 0.035 0.038