World Happiness Report 2023 71 shown in Table 13 of Statistical Appendix 1 in World Happiness Report 2018. The Table 13 results are very similar to the split-sample results shown in Tables 11 and 12, and all three tables give effect sizes very similar to those in Table 2.1 in the main text. Because the samples change only slightly from year to year, there was no need to repeat these tests with this year’s sample. 18 Actual and predicted national and regional average 2020-2022 life evaluations are plotted in Figure 37 of Statistical Appendix 1. The 45-degree line in each part of the Figure shows a situation where the actual and predicted values are equal. A predominance of country dots below the 45-degree line shows a region where actual values are below those predicted by the model, and vice versa. Southeast Asia provides the largest current example of the former case, and Latin America of the latter. 19 See Rojas (2018). 20 If special variables for Latin America and East Asia are added to the equation in column 1 of Table 2.1, the Latin American coefficient is +0.51 (t=5.4) while that for East Asia is -0.17 (t=1.7). 21 See Chen et al. (1995) for differences in response style, and Chapter 6 of World Happiness Report 2022 for data on regional differences in variables thought to be of special importance in East Asian cultures. Those data do not explain the slightly lower rankings for East Asian countries, as the key variables, including especially feeling one’s life is in balance and feeling at peace with life, are more prevalent in the ten happiest countries, and especially the top-ranking Nordic countries, than they are in East Asia. However, as also shown in Chapter 6 of World Happiness Report 2022, balance, but not peace, is correlated more closely with life evaluations in East Asia than elsewhere, so that the low actual values may help to partially explain the negative residuals for East Asia. 22 One slight exception is that the negative effect of corruption is estimated to be slightly larger (0.86 rather than 0.71), although not significantly so, if we include a separate regional variable for Latin America. This is because perceived corruption is worse than average in Latin America, and its happiness effects there are offset by stronger close-knit social networks, as described in Rojas (2018). The inclusion of a special Latin American variable thereby permits the corruption coefficient to take a higher value. 23 As represented by Western European countries, the United States, Australia, New Zealand and Canada. 24 More precisely, the test vehicle is the equation in column 1 with no year fixed effects, given our wish to compare the three COVID-19 years to the three preceding years. 25 These results are presented and explained on pages 26-34 of World Happiness Report 2022. 26 Standard errors for happiness gaps (and the associated rank confidence intervals) in Figure 2.2 are computed by nonparametric bootstrap with 500 replications. 27 Allison and Foster (2004) show that even if life evaluations are interpreted as containing ordinal information only, a distribution of responses is more “spread-out” than a second distribution if and only if the gap in top/bottom means in the first distribution is greater than of the second distribution, for any assignment of values to the categories. Thus when the ranking of distributions by top-minus-bottom mean spread is unambiguous, it represents the correct ranking of inequality. 28 See Goff et al. (2018) for evidence that equality of happiness is correlated with happiness levels, even using a purely ordinal measure of equality. Grimes et al. (2023) report further evidence on this front, specifically that a concentration of individuals at the unhappy end of the ladder creates a negative externality that brings down happiness levels overall. 29 WEIRD=Western Educated Industrial Rich Democracies, represented in our data by Western Europe and the mixed group including the United States, Australia, New Zealand, and Canada. 30 T he latter measure was the focus of chapter 5 of World Happiness Report 2015, on the sources of happiness and misery. 31 Splitting a country into more and less happy halves requires a rule to assign survey respondents at the country’s median ladder rung to one or the other half. To calculate means for life evaluations in each half, we simply split the median respondents in the proportions necessary to produce two halves of equal size. To calculate top- and bottom-half means of emotions, social pillars of well-being, and benevolent behaviours, we use predicted life evaluations for each respondent to split the respondents at a country’s median based on how they rank by these predicted values. The regression used to fit the predictions is an individual- level analogue of the specification in the first column of Table 2.1 with a specification akin to that used in Table 2.2 of World Happiness Report 2022. We run this regression on the entire global sample of individual responses from 2005 through 2022, with country and year fixed effects, and use the estimated coefficients to calculate predicted life evaluations for each respondent. Those at a country’s median are assigned to the more or less happy half of their country on the basis of this ranking in the proportions necessary to achieve equal halves. This means that among respondents at the median, the social pillars of well-being are higher for those assigned to the top half than for those assigned to the bottom half, by design. Respondents at values other than the country’s median are assigned to the top or bottom half on the basis of their actual life evaluation, regardless of the life evaluation predicted by their other survey responses. 32 We included individuals in all countries where there was at least one survey in 2017-2019 and in every year 2020-2022, producing a sample of 563,543 individuals in 128 countries. The structure of the equation matched very closely that in column 3 of Table 2.4 in World Happiness Report 2022, with the addition this year of an interaction between age and gender. We eliminated this year all respondents who reported zero household income, which substantially raised the income effect and also removed any significant change to the income effect during COVID-19. 33 The pre-pandemic effect of having a health problem was -0.459 (t=15.9), and the additional effect during 2020-2022 was -0.055 (t=2.2).
RkJQdWJsaXNoZXIy NzQwMjQ=