Does structural breakage occur in the World Happiness Report model during a recession?

Your name: Sonny de Bok

Your student number: 5754887

Specialization: Economics and Finance

Field: Mint, effect of recessions

Number of credits thesis: 12

Title of your research: Does structural breakage occur in the World Happiness Report model during a recession?

Statement of Originality

This document is written by Student Sonny de Bok who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Abstract

There is an increasing desire to both measure and understand well-being as a global phenomenon. This has given rise to multiple indicators, attempting to scale well-being, of which the World Happiness Report is the most popular. A decent amount of literature has been dedicated to testing both the internal and external validity of this model, most of them focusing on a single country or region. Some studies have tested the usefulness of this model during a recession. However none of these studies tested whether or not structural breakage, or regime switching, occurs on a global scale during a global recession. Structural breakage occurs when a significant change in the coefficients of the six independent variables explaining life evaluation scores, takes place. This paper attempts to answer that question by analyzing available data during the most recent recession, as well as a multitude of other time frames to account for potential lag. The method consists of statistical tests on both the significance of the dummy variable for a recession as well as the interaction terms with the explanatory variables. This paper showed no indicators of structural breakage in 27 of the 28 time frames. The one time frame that showed any significance was 2012-2013, outside of the time frame generally assumed to contain the most recent recession. No significance of the dummy variable recession was observed, however an F-test on joint significance of the interaction terms showed a significant P-value of 0.022. Overall this paper cautiously concludes to not reject the null hypothesis of no structural breakage occurring.

Introduction

Compared to a few decades ago there is a substantial larger amount of indices, ratios and indicators to choose from when evaluating a country. These range from the development index to the freedom index. These vast amounts of new scales seems especially important in economics, since economists seem to have shifted some focus towards monitoring and understanding well-being. (Diener & Seligman, 2004, p.2). The authors state that as a consequence these quality-of-life indicators are dominated by economics, both in the policy area as in media coverage. However, not all economists are in favor of these new indicators as Nordberg (2010) calls them the enemies of the GDP. While acknowledging the potential downsides that arise when using GDP as a measure of a country’s performance, the author argues that when using GDP as a measurement at least it’s clear what it measures and what it fails to measure. This does not always seem to be the case with the more subjective indices, ratio’s and indicators.

So why this sudden interest to quantify subjective factors like happiness or a country’s development? An answer to this question can be found in the potential advantages of successfully monitoring and understanding these factors. Mapping development for example will increase understanding of the matter and will help us center development towards improving individuals capabilities, achievements and freedom (Human Development Report, 1994).

This paper starts by discussing relevant literature. This consists of a short overview on why understanding and measuring well-being is important, some general distinctions made, as well as an overview of findings in literature on the explanatory variables. The next chapter covers the methodology, followed by the results. Lastly, limitations will be discussed and all of the above will be summarized in a brief conclusion.

Literature review

Happiness, and to similar extent well-being, appear to have an important role both in current indices, ratio’s and indicators as well as in literature. Multiple authors describe it as a necessary condition that allows individuals, groups and societies to both flourish and thrive (Diener, 2009; Huppert et all, 2005; Huppert, 2009; Seligman, 2011). According to Sanz et all. (2017, p.1) it’s the basis of a meaningful life for most people and thus there is an increasing intention to measure it. Diener & Seligman (2009, p.1) argues for the importance for both nations and organizations to monitor the well-being of their workers accompanied by steps to improve it. The most recent recession has also been accompanied by an increasing desire to measure well-being globally (Boffo, Brown & Spencer, 2016, p.459). Another reason why understanding well-being specifically has become increasingly important is because current attempts of globally measuring it have shown significant variations of well-being between countries (Michaelson et all, 2009, p. 4).

Current literature makes a distinction between two types of subjective well-being namely; emotional well-being -the quality of emotions felt on a daily basis- and life evaluations. The latter refers to the way people evaluate the quality of their life when asked (Kahneman & Deaton, 2010, p.1). Emotional well-being can be viewed as the sum of positive and negative emotions and can be challenging to measure (Diener & ng, 2010, p.52). Correlations found with emotional well-being are mostly variables of social nature like health and loneliness. Correlations found with life evaluations however are mostly economic such as income and education. Income seems to have a special status within these studies, since it often affects both life evaluations as well as emotional well-being significantly, however it has been shown to occasionally only affect life evaluations (Kahneman & Deaton, 2010, p.2). This paper will focus almost entirely on life evaluations.

To determine which variables influence life evaluations, a distinction has to be made. From a macro perspective, variables like income, education and perceived freedom of the country of residence are arguably the biggest indicators (World happiness report, 2013). However these are not the best indicators when predicting the level of life evaluations of an individual. On an individual level a factor like marital status would be a better indicator than any of the six macro variables (Kahneman & Deaton, 2010, p2). This paper will focus on life evaluations from a macro view, which tries to explain collective life evaluations as a global phenomenon in a country or region.

The Gallup world poll started in 2005 with measuring collective life evaluations on a global scale by continually surveying citizens in 160 countries, representing over 98% of the world’s population. Seven

years later the World Happiness Report (henceforth WHR) analyzed this now available data (Helliwell Layard & Sachs, 2013). The first WHR ranged a multitude of variables influencing life evaluations ranging from “values and religion” to “type of work”. The following 2013 report showed a total of six variables that were statistically significant influencers of life evaluations, these six variables are still being used in their latest 2018 report. These variables include: Income, freedom, generosity, perception of corruption, healthy life expectancy, and social support. It appears that the WHR went from a somewhat individual perspective to a more macro perspective in explaining well-being as a global phenomenon. The WHR score is the arithmetic sum of these six values plus a constant, meaning every explanatory variable has an equal weight in this model. This paper will use the same six variables, some of which have been discussed extensively in the current literature, some of them less so.

Income is by far the most discussed variable affecting life evaluations in literature. Indeed “does money buy happiness?” is an age-old question that has no easy answer. An answer which, according to Kahneman & Deaton (2010, p.4) can never be captured in a single article. Carlsen (2017, p.5-6) showed that GDP had the single biggest correlation with reported life evaluations and the fourth biggest indicator when normalized. GDP, or a derivative thereof, appears to be used as a proxy for income in most studies. While it doesn’t capture every change in income in a country or region, it appears to be the best available tool both for estimating current income levels in a country or region as well as estimating changes in income over time. Richer countries are indeed, on average, happier than poorer countries so a correlation between GDP and well-being exists, this however does not imply causality. (Hagerty & Veenhoven, 2003, p.24). The rationale behind this correlation is the presumption that the ability for individuals to pursue their choices comes with a higher disposable income (Applasamy et all, 2014; Boffo et al., 2017).

However other research has shown that relatively big changes -a magnification of income of up to three times- showed virtually no increase in well-being, the so called Easterlin-paradox (Diener & Biswas-Diener, 2002; Easterbrook, 2003; Easterlin, 1995; Oswald, 1997; Clark, Frijters & Shields, 2008). Frey and Stutzer (2002) have shown that the correlation between economic growth and individual happiness has broken down, or at least diminished.

While a relatively high amount of studies have studied the correlation between well-being and income, very few have tried to explain it (Diener & Ng, 2010). Attempts at explaining this phenomenon range from a yearly income cap of $75,000 above which utility doesn’t increase any more, to adaptation; the joy of any increase in income is transient and thus short-lived (Kahneman & Deaton, 2010, p.1-2). The author also arguesthat incorrect analysis of data also explains a part of the observed paradox since a decent amount of studies have been found to use raw GDP data instead of its log. Diener & NG (2010, p.2)

conclude income is the strongest indicator of happiness when basic needs -like water, food and shelter- are not yet fully met, indicating a convex relationship between the two.

If the relationship between happiness and income is indeed both of causal nature as well as convex, a decrease in income should result in a decrease of the same magnitude or bigger -depending on the starting point- compared to when income increases. This side of the relationship however has been researched less extensively than an increase has, most studies are one-directional (Gudmundsdottir, 2011, p.1). Since a recession is defined in terms of GDP, the occurrence of one is likely accompanied by a decrease in income. This is expected to result in a negative effect on life evaluations.

Inglehart & Klingemann (2000) found a substantial relation between freedom and a nation’s well-being. Carlsen (2017, p.5-6) ranked freedom the fourth and third correlation and normalized indicator respectively. A distinction can be made between economic and political freedom, the first mentioned appears to be more important in poorer countries, while the latter is a bigger factor in developed countries (Diener & Seligman, 2004, p.6). The authors state that, during the most recent financial crisis, the most affected out of the six dependent variables was freedom. The authors argue that the financial crisis seems to have both limited individuals available services through cutbacks as well as a decline in expected future opportunities. Therefor a recession is expected to have a negative effect on freedom, resulting in a negative effect on life evaluations.

Generosity has the lowest correlation with life evaluations and was the second lowest indicator when normalized (Carlsen, 2017, p.5-6) The author also states that some countries, like Finland and Switzerland, show discrepancies in generosity with a value significantly lower than expected. According to Boffo et al. (2017) the latest financial crisis has lowered the overall generosity. It is therefore expected that a recession has a negative effect on generosity and thus a negative effect on life evaluations, although this effect is expected to be relatively small.

Perception of corruption had the second lowest correlation with life evaluation scores and was founded to be the lowest indicator when normalized (Carlsen, 2017, p.5-6). The index used to scale perception of corruption as a global phenomenon is reversed, meaning a higher score indicates a lower perception of corruption.It is thus expected that a recession brings forth a decrease in the perception of corruption score and thus a, relatively small, decrease in reported life evaluations.

Healthy life expectancy was the single biggest indicator on reported life evaluations when normalized and reported the second biggest correlation (Carlsen, 2018, p.5-6). It makes intuitive sense that healthy people will in fact score higher regarding life evaluations, however health seems to correlate more with emotional well-being than it does with life evaluations (Kahneman & Deaton, 2010, p.1). The

author also argues that the causality is expected to be stronger when reversed; the effect of life evaluations on health is larger than the effect of health on life evaluations. Most of the current literature about the effect of a crisis on health seems to focus on disorders of mental health, instead of well-being (Gudmundsdottir 2011), so more work is required. It is expected that a recession has a negative effect on health and thus a negative effect on life evaluations.

Social support is arguably one of the most important variables. Carlsen (2018, p.5-6) ranked it the third and second biggest factors in correlation and when normalized respectively. Gudmundsdottir (2011, p.1086) claims social support is probably the single biggest factor influencing well-being claiming the current tools do not capture its effect fully. However just like it is the case with health its effect on emotional well-being is expected to be larger than its effect on life evaluations. Boffo et al. (2017, p.465) claims that overall social support has dropped during the most recent financial crisis, which negatively influenced on both emotional well-being and life evaluations. This paper expects the same relationship between a recession and social support, resulting in a negative effect on life evaluations.

It appears current literature describes a recession negatively affecting all six, positively correlated, explanatory variables of life evaluation, predicting a negative effect of a recession on life evaluation scores. When changes in reported well-being in Eurozone countries during the most recent financial crisis were analyzed the biggest declines found appear to be found in countries that have beenhit the most by the crisis, like Portugal, Italy, Spain and Greece (Boffo et al., 2016, p.456). Their decreases were quite significant, equal to that of dropping twenty ranks in the WHR or halving their GDP (Helliwell et al., 2013, p.15).

Contradicting this however is a slight increase in reported global well-being during the most recent recession as well as anomalies like the United Kingdom that, despite being hit hard by this recession, showed a steady reported well-being (Boffo et al., 2017, p.456). However the authors do not argue that there was no decline in well-being, current measurements just seem to have failed to capture it. This is in part explained by a change in expectations and norms on survey responses. To evaluate life evaluation survey responses a benchmark is used, as an alternative option. Groups that score low on the life evaluation scale have a benchmark that is arguably too low, resulting in reported scores that are too high. (Boffo et al., 2017, p.451)

When data on Iceland, a country also hit hard by the recession, was analyzed an under expected small decline of reported well-being was observed, again suggesting a change in the relationships. The first explanation for this occurrence given by the author is closer social relationships that have reduced the negative effects, with the model failing to capture this increase. The second explanation is an

overvaluation of the importance of economic indicator such as GDP -a critical variable used to define a recession- with economist worldwide overestimating its relationship with well-being (Gudmundsdottir 2011, p.1089).

Another explanation is offered by Boffo et al. (2017, p.459) arguing that this paradox can be explained by reinterpreting the reported well-being data through a different perspective, namely a political economic perspective mapping how individuals respond to social surveys. This philosophy would put a bigger emphasis on the hedonic treadmill: the tendency for individuals to rapidly return to a happiness level that is relatively stable, despite large negative or positive shocks, events or life changes. If this is in fact true and applies to this scenario, this phenomena could potentially be a cause of structural breakage, assuming a recession produces a higher frequency of large negative shocks in individuals’ life. The reasoning for this is the following, if on average the aggregated life evaluations tends to shift back to previous levels relatively easy, the model would overestimate an effect of a recession on life evaluations. However the reverse, the model underestimating the effect of a recession, is also a possible way for structural breakage to occur. This would suggest a recession brings forth a negative effect uncaptured in the decline of the six explanatory variables.

Overall a recession should have a substantial negative effect on reported well-being, both trough variables like perception of corruption or GDP directly as well as indirectly through variables like life expectancy that suffered a drop as a result of the decline in GDP. Carlsen (2018, p.5-6) showed that the correlation between GDP and the six remaining independent variables were all positive. This indicates that a decline in the values of the remaining six explanatory variables, as a consequence of a recession, is at least in part explained by the decline in GDP. This effect one of the explanatory variables has on the other explanatory variables could create multicollinearity. However since this study attempts to test whether or not structural breakage occurs in current life evaluation models during a recession, the exact amount of variation explained by our variables isn’t of great importance as long as the explanatory variables capture the same amount of variation outside of a recession compared to during a recession.

Since there appears to be no available literature on the testing of structural breakage of coefficients on models of life evaluations -or other subjective indicators- it’s impossible to outline a framework of current literature on this topic. This however implies all the more necessity for this type of research.

Methodology

Data on the six explanatory variables as well as the dependent variable was retrieved from multiple reports and databases. As a consequence of the WHR using data that isn’t publicly available, two of the variables had to be proxied through different datasets, namely social support and healthy life expectancy.

Virtually every study uses gross domestic product as a proxy for income, with this study being no exception. However a decent amount of studies have used raw GDP instead of its log. This study uses the latter, as did the WHR. Since the results of the survey of the world Gallup Poll used for social support in the WHR is not publicly available, a different proxy for social support had to be used. This proxy was the total social contribution transferred from companies to governments to cover social securities, as a percentage of company profits. It is expected that this -as is the case with life expectancy- won’t exactly capture the same variation of social support as the WHR did, however as long as it captures the same explanatory variation outside of a recession as it does within, the variable can be used to assess if structural breakage in its coefficient occurred. To account for perception of corruption this study also follows the WHR, the proxy used is the most commonly used proxy for estimating corruption globally; namely the Corruption Perception Index. The proxy for freedom was also in line with the proxy the WHR used, namely the freedom index. This index is the most commonly used index to assess both economic and human freedom and consists of the equally weighted sum of both types. Since Gallup’s healthy life expectancy – which is currently being used by the WHR- is not publicly available as well, the closest proxy had to be used, life expectancy in years. As makes intuitive sense a high correlation between healthy life expectancy and life expectancy is expected. The proxy used for generosity was also in line with the WHR, namely the World Giving index; the most commonly used index to estimate global tendencies in generosity. With data on life evaluation scores and all six explanatory variables a panel-data regression analysis was performed to test for structural breakage, meaning a significant change in the six coefficients during a recession.

To test this the following model is introduced.

Yi,t = (Xi,t)’ β + (dt Xi,t)’ ϒ + αi + λt + Ui,t

With Xi,t = X⃗⃗ = [LNgdp ̂i,t, Freedom ̂i,t, Life expectation ̂i,t, Corruption ̂i,t, Health ̂i,t, Social Support ̂i,t] or the column-vector of the six explanatory variables. Yi,t corresponds to the estimated life evaluation score for country i and period t. β consist of the estimated coefficients that are being tested for structural breakage. dt is the dummy variable for a recession. (dt Xi,t)’ is the transpose matrix of the six applied interaction terms. The null-hypothesis states ϒ = 0, meaning no structural breakage occurred, which will

be tested against the P-value of the corresponding F-test. Ui,t is the error term. αi is included as a country specific intercept, meaning αi captures the leftover variation in the dependent variable which the six explanatory variables can’t explain, or the country specific effect. λt captures the year specific effects, estimated by year dummies.

Regarding the time frame chosen for the dummy variable literature seem to suggest the most recent recession initiated in 2008 and ended in 2011. However since a potential lag between the moment a recession occurs and the moment individual life evaluation levels adjust could exist, structural breakage is possible to occur at a different time. By applying multiple time frames this paper looked for multiple structural breakage points. Combining this dummy variable with all six explanatory variables, X1 through X6, creating equally many interaction terms, upon which these terms were utilized to estimate if structural breakage, or regime switching, occurred.

To test whether a fixed effect regression is preferred over a random effect regression a Hausman test was performed on the eight single year time frames, namely 2008-2015. The highest P-value found was 0.0035 for 2012, which is still under the threshold of 5% to reject the null-hypothesis stating a random effect regression is preferred. Based on this test this paper proceeded the analysis by using fixed effects. By using a fixed effect regression this paper made sure that the model controls for variables that are constant across time but vary over countries.

Results

Table 1 shows the results of regressing life evaluation scores on all six explanatory variables for the time period of 2009-2011 as well as the results of a regression on the explanatory variables, the dummy variable and the interaction terms for the same period. This period was chosen as a time frame containing the recession, which was indicated by literature. When regressing life evaluation scores on the six explanatory variables three variables were found to be of statistical significance. These three were life expectancy, perception of corruption and the natural logarithm of the GDP, with the first mentioned significant at an alpha of 0.05 while the two other variables were found statistically significant at α = 0.01. When both the dummy variables and the interaction terms were added only two of the dependent variables were found to be a significant influence on life evaluation scores, both of them at α = 0.01, despite all six explanatory

coefficients being positive. These two significant variables in the unrestricted model consists of the logarithm of the GDP and the perception of corruption. This indicates that the effect of the variable life expectancy, which lost significance, is in part explained by the added variables of the unrestricted model. Neither the dummy variable nor any of the interaction terms were found significant. This indicates the H0 claiming there is no structural breakage in this time period cannot be rejected. However since there is an existence of potential lag more time frames had to be analyzed before conclusions can be made regarding the entire sample.

Table 2 shows the most important indicators of structural breakage for different time frames. The dummy variable recession appeared to be significant in neither of the 28 time frames. No significant F-test results on the interaction terms were found for 27 time frames. The time frame 2012-2013 showed a significant outcome of the F-test (α = 0.025) on the interaction terms, though this would require a relatively long lag. Since there was no available data on generosity for the years 2008 and 2009 all time frames containing either of these years have been regressed on five dependent variables. The period 2009-2011 has been regressed on both six and five variables.

The between R² reports how much of the variance between separate panel countries the model accounted for. The within R² is the amount of variance within the countries that are accounted for by the model. As can be seen in Table 2 the between R² is larger in every period. This indicates that the model better explains variance between countries than it does within a country.

The negative coefficient of the dummy variable containing the largest absolute value was found in 2008. Although the coefficient is found to be statistically insignificant, this is in line with literature. Results of the unrestricted model show that that none of the dummy variable coefficients in either time frame are significant. Noteworthy is that, contrary to what literature suggests, coefficients aren’t consistently negative. Only 14 coefficients out of 28 show a negative value. However the large amount of overlap in the time frames accounts for most of this observation. The average values of the coefficients of the dummy variable in the eight single year time frames was found to be -0.971, with only two of them being positive.

Table 1

Regression of life evaluation scores on six explanatory variables

(1) (2) VARIABLES Life evaluations Life evaluations Life expectation 0.0242* 0.0489 (0.0143) (0.034) Generosity 0.00401 0.0025 (0.00417) (0.005) Freedom 0.149 0.0262 (0.128) (0.189) Ln GDP 0.167*** 0.6282*** (0.0442) (0.201) Corruption 0.176*** 0.2131*** (0.0411) (0.080) Social Support 0.000578 0.0028 Recession Recession#Life expectation Recession#Generosity Recession#Freedom Recession#LnGDP Recession#Corruption (0.00626) (0.0158) 0.8108 (0.921) -0.0049 (0.011) 0.0009 (0.003) -0.0634 (0.103) 0.0073 (0.024) 0.0003 (0.03)

Recession#Social support -0.0012 (0.0023) Constant -2.413* -15.3107** (1.395) (5.882) Observations 426 426 Number of country1 84 84

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 2

Indicators of structural breakage in different time frames

R² Dummy variable recession P-values F-test

Year Within Between Overall Coefficient P-value All variables Interaction terms 2010 0.0633 0.2539 0.2783 -0.9952 0.3550 0.0650 0.8301 2011 0.0830 0.4296 0.4177 1.0667 0.2580 0.0243 0.9139 2012 0.0694 0.3113 0.3252 -0.1083 0.8880 0.0911 0.6665 2013 0.0851 0.3552 0.3598 -0.3746 0.7260 0.0046 0.0796 2014 0.0586 0.2905 0.3069 -0.1700 0.8750 0.1019 0.9057 2015 0.0818 0.2217 0.2504 1.7006 0.0700 0.0495 0.1033 2010-2011 0.0946 0.4846 0.4607 0.8108 0.3810 0.0020 0.8526 2010-2012 0.0621 0.4045 0.3949 0.3009 0.7700 0.1341 0.9970 2010-2013 0.0801 0.1333 0.1788 -0.9961 0.4590 0.0418 0.3992 2010-2014 0.0642 0.2920 0.3082 0.0430 0.9630 0.1156 0.7660 2011-2012 0.0613 0.3364 0.3433 0.5558 0.4690 0.1098 0.9593 2011-2013 0.0642 0.2920 0.3082 0.0430 0.9630 0.1156 0.7660 2011-2014 0.0638 0.3150 0.3259 0.0163 0.9800 0.1024 0.8381 2012-2013 0.0900 0.3473 0.3554 -0.7604 0.3140 0.0039 0.0222 2012-2014 0.0839 0.3887 0.3835 -0.7707 0.2920 0.0434 0.2071 2013-2014 0.0700 0.3829 0.3788 -0.5271 0.5040 0.0594 0.4594 2009-2011 0.0946 0.4846 0.4607 0.8108 0.3810 0.0020 0.8526 2008 0.0907 0.2307 0.2416 -1.7729 0.1560 0.0053 0.3562 2009 0.0947 0.2574 0.2631 -0.3173 0.8110 0.0000 0.3211 2009-2011 0.1083 0.4229 0.3960 0.6410 0.4790 0.0002 0.5122 2008-2009 0.0864 0.2281 0.2394 -1.2259 0.2230 0.0002 0.7427 2008-2010 0.0884 0.1834 0.2039 -1.2603 0.1950 0.0004 0.6080 2008-2011 0.1083 0.4229 0.3960 0.6410 0.4790 0.0002 0.5122 2008-2012 0.0876 0.3460 0.3303 0.6147 0.5330 0.0031 0.9569 2008-2013 0.0873 0.1711 0.1912 -0.2439 0.8130 0.0049 0.6399 2009-2010 0.0894 0.2398 0.2483 -0.4561 0.6670 0.0002 0.4461 2009-2012 0.0885 0.2802 0.2800 1.0474 0.2080 0.0025 0.8634

2009-2013 0.0883 0.2524 0.2577 0.7569 0.3600 0.0014 0.7078 Note. Bottom 11 time frames are regressed on five variables with generosity removed

Discussion

By performing statistical tests on both life evaluation scores and the six variables that are assumed to affect these scores, this paper answered whether or not structural breakage, or regime switching, occurred during the most recent recession. For this to occur statistically significant changes in the coefficients have to be observed. This paper has found a single time frame in which only the interaction terms with the dummy variable recession turned out to be significantly different from zero. In the remaining 27 time frames neither the interaction terms nor the dummy variable was found significant. This cautiously answers the research question by not rejecting the null-hypothesis of no structural breakage occurring during a recession.

A possible explanation for the absence of positive results of structural breakage could be found in the way surveys are taken. Literature on this topics suggests there is still room for improvement, which makes it reasonable to assume current survey methods are not fully capturing the entire impact a recession has on individuals in its current form. Not only could this error of measurement influence the independent variable, it could also affect three of the explanatory variables that also rely on survey results, namely perception of corruption, freedom and generosity. Another explanation possibly lies in the replacement of some of the variables used by the WHR by different proxies, which arguably show weaker relationships than the variables used by a developed model like the WHR. Another reasonable explanation for these insignificant result potentially lies in the size of the decline in both the explanatory variables as well as the independent variable as a consequence of the recession, perhaps these were still too small for statistically significant structural breakage to occur. The insignificant values found could also be the result of two opposing effects occurring simultaneous. An example of this would be an overreaction to a recession, as mentioned in the literary review, combined with the hedonic treadmill phenomena. This results in a short-lived steeper decline than the model predicts, which the only yearly taken surveys potentially fail to detect. If the hedonic treadmill effect occurs simultaneous, the aggregated life evaluations, on average, increase gradually back to old levels with the extra decline never captured. To assess whether or not this occurred is outside the scope of this paper. Finally, it could be the case that that the relationship simply performs equally well explaining life evaluation scores when the explanatory

variables increase as when they decline. If this were in fact true it would increase the applicability of the model, however more testing is required before this can be concluded.

An unexpected finding was the significance of the variable corruption. Literature suggests an overall low correlation. This could indicate an increasing importance of the perception of corruption during a recession, however further testing is needed. Another surprising result found was the seeming random behavior of the dummy variable coefficients with half of the values found being positive. Although this is in part explained by the overlap in time frames. After accounting for this overlap the effect of the dummy variables recession seems to be generally negative, which is in line with literature’s prediction.

Conclusion

The most recent recession seemed to have had no effect on the coefficients used in the current -most widely used- model explaining life evaluation scores. This study has looked at a vast set of time frames, testing each of them on the occurrence of structural breakage, or regime switching. In neither of the 28 time frames a significant effect of the dummy variable recession on life evaluations was found. In the time frame 2012-2013 a significant p-value of the F-test on the interaction terms was observed.

In the unrestricted model two variables were found to be of statistical significance, these consisted of the logarithm of the GDP and the perception of corruption. The latter was surprising, since it was predicted to be of relatively low significance by literature. The dummy variable recession in the eight single year time frames used was negative overall, which was in line with literature.

Overall no structural breakage has been found between the life evaluation scores and its six assumed to be dependent variables. A number of possible explanations are given, these include potential measurement error in survey results, the replacement of some of the variables used by the WHR and the perhaps too limited size of the effect of the recession on both the explanatory variables as well as the independent variable. Another explanation offered by this study is the occurrence of two opposing simultaneous effects occurring. An example of this would be an overreaction of a recession combined with the hedonic treadmill. Since survey results are taken yearly it’s reasonable to assume short-lived effects aren’t always captured. The final explanation given by this paper would assume a bi-directional relationship between the variables, as a consequence the model would explain a decrease in life evaluation scores reasonably identical to an increase.

Bibliography

Applasamy, V., Gamboa, R. A., Al-Atabi, M., & Namasivayam, S. (2014). Measuring Happiness in Academic Environment: A Case Study of the School of Engineering at Taylor's University (Malaysia). Procedia-Social

and Behavioral Sciences, 123, 106-112.

Boffo, M., Brown, A., & Spencer, D. A. (2017). From happiness to social provisioning: addressing well-being in times of crisis. New Political Economy, 22(4), 450-462.

Carlsen, L. (2018). Happiness as a sustainability factor. The world happiness index: a posetic-based data analysis. Sustainability Science, 13(2), 549-571.

Clark, A. E., Frijters, P., & Shields, M. A. (2008). Relative income, happiness, and utility: An explanation for the Easterlin paradox and other puzzles. Journal of Economic literature, 46(1), 95-144.

Diener, E., Ng, W., Harter, J., & Arora, R. (2010). Wealth and happiness across the world: material prosperity predicts life evaluation, whereas psychosocial prosperity predicts positive feeling. Journal of

personality and social psychology, 99(1), 52.

Diener, E., & Seligman, M. E. (2004). Beyond money: Toward an economy of well-being. Psychological

science in the public interest, 5(1), 1-31.

Gudmundsdottir, D. G. (2013). The impact of economic crisis on happiness. Social indicators research,

110(3), 1083-1101.

Hagerty, M. R., & Veenhoven, R. (2003). Wealth and happiness revisited–growing national income does go with greater happiness. Social indicators research, 64(1), 1-27.

Jany-Catrice, F., & Méda, D. (2013). Well-being and the Wealth of Nations: How are They to Be Defined?. Review of Political Economy, 25(3), 444-460.

Kahneman, D., & Deaton, A. (2010). High income improves evaluation of life but not emotional well-being.

Michaelson, J., Abdallah, S., Steuer, N., Thompson, S., Marks, N., Aked, J., ... & Potts, R. (2009). National accounts of well-being: Bringing real wealth onto the balance sheet.

Norberg, J. (2010). GDP and its enemies: the questionable search for a happiness index.

Sanz, M. T., Caselles, A., Micó, J. C., & Soler, D. (2018). A stochastic dynamical social model involving a human happiness index. Journal of Computational and Applied Mathematics, 340, 231-246.

Veenhoven, R. (2000). Well‐being in the welfare state: Level not higher, distribution not more equitable.

Journal of Comparative Policy Analysis: Research and Practice, 2(1), 91-125.

Veenhoven, R. (2002). Average happiness in 68 nations in the 1990s: How much people enjoy their life-as-a-whole. Retrieved April, 12, 2004.