Well-being and the Great Recession in Spain

ABSTRACT This letter assesses the impact of the Great Recession on well-being in Spanish provinces using two alternative composite indicators of objective well-being that include somewhat different dimensions. Whereas the crisis notably eroded economic well-being, its impact on overall well-being – which in addition to economic dimensions also includes non-economic ones – was imperceptible. This result points to the need to carefully define and assess well-being in empirical analyses.


I. Introduction and motivation
Assessing well-being is a challenging task. Whereas traditional measures have been based on simple economic indicators, mostly GDP per capita, society as a whole calls for a more comprehensive way to gauge well-being. In addition to economic issues, fresh measures should include other non-economic dimensions of well-being such as health, education or the environment, to name just a few. In response to this social demand, the Human Development Index was created by the United Nations in the 1990s, and included income per capita, education and life expectancy (see UN 2016). Later on, the Commission on the Measurement of Economic Performance and Social Progresslaunched in 2009 and headed up by the Nobel laureate Joseph Stiglitzsuggested several non-economic dimensions that, beyond economic ones, can affect well-being (Stiglitz, Sen, and Fitoussi 2009). Furthermore, the OECD provides data at both national (Better Life Index dataset), and regional (OECD Regional Well-being Database) levels on these dimensions (see Durand 2015).
Recently, several papers have employed these datasets to building objective composite indicators of well-being, mostly at the country level (e.g. Lorenz, Brauer, and Lorenz 2017;Peiró-Palomino and Picazo-Tadeo 2018). Nonetheless, the rankings of countries resulting from these indicators are not significantly different from those derived from conventional measures of wellbeing, including GDP per capita. In addition, other recent research papers also conclude that there is a close relationship between subjective well-being and income (Stevenson and Wolfers 2013).
According to the arguments outlined above, it might be sensible to assume that indicators based on economic dimensions such as GDP per capita are good proxies of overall well-being. However, in this letter we show that this might not be the case when it comes to assessing the impact of an economic crisis on well-being. In doing so, we empirically evaluate the impact of the Great Recession that began in 2008 (IMF, 2009;Camacho, Gadea, and Pérez-Quirós 2018) on wellbeing in Spanish provinces, using two composite indicators of objective well-being computed with Data Envelopment Analysis (DEA) and Multi-Criteria-Decision-Making techniques (MCDM). One of themreferred to as economic wellbeingjust includes GDP per capita and the unemployment rate as essential economic dimensions of well-being, whereas the othernamed overall well-beingalso includes 5 non-economic dimensions. We find that: i) there are notable disparities in well-being across Spanish provinces; ii) when objective well-being is assessed only with economic dimensions a sharp decline is observed as a result of the Great Recession; conversely, when non-economic dimensions are also accounted for, well-being remains fairly stable. Accordingly, our conclusion is that the choice of well-being indicators should be carefully justified in empirical analyses.

II. Data and methodology
We employ information about 7 well-being dimensions at the level of the 50 Spanish provinces, 1 built with data from different sources for the period 2000-14 (Table 1). In constructing this dataset, we have attempted to assemble a set of indicators as close as possible to those proposed by Stiglitz, Sen, and Fitoussi (2009) and provided by the OECD at both national and regional levels, but using the much more limited information available for the Spanish provinces. Using these figures, we have computed averages for each indicator in the growth period 2000-07 and the crisis period 2008-14. To ensure comparability across dimensions, and given that the indicators have different measurement units, the data have been standardised on a 0-1 scale using the min-max method, with higher values representing better performance; minimum and maximum values are chosen from the whole 2000-14 period to ensure comparability over time. Finally, normalised data on dimensions have been used to build a couple of composite indicators of objective well-being: the first, economic well-being, includes only the economic dimensions of income and jobs, while the other, overall well-being, considers all 7 dimensions.
Regarding the methodology, we have used DEA and MCDM techniques as in Peiró-Palomino and Picazo-Tadeo (2018). First, following Lovell, Pastor, and Turner (1995) we have computed a composite well-being indicator for each province p' with DEA as: w dp 0 dimension dp 0 Subject to: P D d¼1 w dp 0 dimension dp 1 p ¼ 1; . . . ; 50 where dimension dp is the observed value for dimension d in province p, and w dp is the idiosyncratic weight assigned to dimension d in the composite indicator of province p. Moreover, composite indicators from (1) are, by construction, bounded between zero and one, the latter representing the highest well-being; i.e. the lower the score, the worse the well-being.
Whereas DEA provides a successful approach to the building of a composite well-being indicator, it might be less effective when it comes to ranking provinces. In this respect, comparisons might be meaningless as provinces' well-being indicators are computed with different sets of weightings (Kao and Hung 2005); besides, program (1) could assign a score of onemeaning highest well-beingto some provinces just because of a lack of discriminating power (see technical details in Dyson et al. 2001). In order to ensure comparability and also increase the discriminating power, in a second stage we have employed MCDM techniques to compute a composite well-being indicator with common weights across provinces for dimensions, as proposed by Despotis (2002). Formally: where w d is the common weight assigned to dimension d; ε is a non-Archimedean small number; h is a non-negative parameter to be estimated; m p represents the deviation between the composite indicator for province p calculated with DEA, and that computed with MCDM; finally, t is a parameter ranging from 0 to 1, which we have set to 1 (see details in Peiró-Palomino and Picazo-Tadeo 2018).

III. Results and discussion
Table 2 displays averages for economic and overall well-being in both growth (2000-07) and crisis (2008-14) periods for provinces and regions; also, averages for regions are weighted by population. 2 In this respect, as well-being affects people, we have considered population-weighted averages to be much more illustrative than simple ones. Figure 1 illustrates the geographic distribution of well-being in Spain, with darker colours representing better performance. Overall well-being is unevenly distributed across space with no clear patterns, although the lowest scores are found in the Mediterranean coast and Southern provinces. Furthermore, a positive (although moderate) association is observed between overall well-being and the level of development of provinces and regionsmeasured by real GDP per capita-, particularly in the recession period. 3 Besides, lower economic well-being is found in Southern and Western provinces while Northern and Eastern provinces perform notably better. Lastly, intraregional heterogeneity is high in most cases, especially for overall well-being. Generally speaking, our results show geographic patterns of well-being that are in line with those Note: Well-being is categorised through the quintile distribution of our composite indicators for the entire period, in order to evaluate disparities across regions and also over time. from other studies of well-being (or quality of life) in Spain carried out using different methodological approaches, aggregation levels and time periods. In this regard, Ventura (2011, 2018) focused on quality of life at the municipal level, although they only considered a limited sample of municipalities and overlooked temporal variation. Murias, Martinez, and de Miguel (2006) and Zarzosa Espina and Somarriba Arechavala (2013) (2018) also assessed well-being at the region level for the period 2006-15, but they did not provide a composite indicator.
Regarding the impact of the Great Recession, a severe deterioration is observed in all provinces between 2000-07 and 2008-14 when well-being is assessed considering only economic dimensions; e.g. population-weighted average economic wellbeing decreases from 0.809 to 0.439. Conversely, well-being remains much more stable when it is assessed with our overall well-being indicator, with a weighted average that even increases slightly from 0.747 to 0.792. 4 These findings can be clearly seen in Figure 1, which also suggests that the geographical North-East versus South-West division observed in the growth years persisted during the crisis. Furthermore, the population-weighted distributions of overall well-being among Spanish provinces are not statistically different between 2000-07 and 2008-14 ( Figure 2); conversely, those of economic well-being are statistically different; i.e. a notable shift to the left has occurred as result of the crisis.

IV. Conclusions
In this letter, we report two main conclusions. First, the Great Recession has profoundly affected the economic dimensions of well-being in Spain, whereas overall well-beingwhich also includes other non-economic dimensionshas remained fairly stable. Second, leaving aside the desire of academics, international organisations and society as a whole to broaden the notion of well-being, this concept needs to be carefully defined and assessed in empirical studies, as different measures may lead to quite different interpretations.