The Relationship Between Social Leisure and Life ... - Springer Link

Soc Indic Res (2012) 108:453–490 DOI 10.1007/s11205-011-9887-5

The Relationship Between Social Leisure and Life Satisfaction: Causality and Policy Implications Leonardo Becchetti • Elena Giachin Ricca • Alessandra Pelloni

Accepted: 5 June 2011 / Published online: 7 July 2011 Springer Science+Business Media B.V. 2011

Abstract Social leisure is generally found to be positively correlated with life satisfaction in the empirical literature. We ask if this association captures a genuine causal effect by using panel data from the GSOEP. Our identification strategy exploits the change in social leisure brought about by retirement, since the latter is an event after which the time investable in (the outside job) relational life increases. We instrument social leisure with various measures of the age cohort specific probability of retirement. With such approach we document that social leisure has a positive and significant effect on life satisfaction. Our findings shed some light on the age-happiness pattern. Policy implications are also discussed. Keywords

Life satisfaction Social leisure Retirement

1 Introduction Economists have traditionally shied away from empirical assessments of subjective indicators. However, there has been a change of heart in the last decade during which the number of papers investigating the determinants of self declared life satisfaction published

L. Becchetti Department of Economics, University of Rome ‘‘Tor Vergata’’, Via Columbia 2, 00133 Rome, Italy e-mail: [email protected] E. Giachin Ricca University of Rome ‘‘Tor Vergata’’, Via Columbia 2, 00133 Rome, Italy e-mail: [email protected] A. Pelloni (&) Department of Economics, University of Rome ‘‘Tor Vergata’’, Via Columbia 2, 00133 Rome, Italy e-mail: [email protected]

123

454

L. Becchetti et al.

in economic journals has been steadily growing (see Clark et al. 2008; Frey 2008).1 A key motivation for this cultural shift has been the increasing awareness of the impact of non– traded goods on social welfare. The Life Satisfaction Approach (Frey et al. 2004) can in fact be seen as an alternative to the traditional methods of measurement of such impact based on contingent valuation or revealed preferences. Happiness data offer a way to calculate the elasticity of the indirect utility function of agents with respect to the availability of any good which cannot be bought directly on the market. An important class of non traded goods is represented by non instrumental social relationships or ‘relational goods’, as they are sometimes defined in the literature. Examples are interactions with friends, participation in the life of clubs, religious bodies, political parties, unions, civic and cultural organizations etc. In standard economic models individuals maximize the utility they derive from consumption of market goods and non-work time, while the choice between solitary and ‘relational’ leisure is left in the background. However, due to coordination or cognitive failures the consumption of relational goods may be inefficiently low at the individual level. The possibility of coordination failures arises simply because, by definition, we cannot consume ‘‘relational goods’’ in isolation: the choices on how much to consume of them are contingent on the choices of others, both at the individual and at the societal levels. Moreover, the possibility of cognitive failures arises because relational goods have high entry costs (it needs time to build a circle of friends) and can be displaced by artificial substitutes (for instance watching television) of increasing sophistication and decreasing prices which require less initial efforts even if are ultimately less rewarding. The social welfare loss arising from investing too little in relationships may be sizable. Indeed, many studies in psychology support the conclusion that social ties are essential to well-being.2 So far, happiness data have been used to evaluate social relationships by Helliwell and Putnam (2004), Bartolini et al. (2009), Aslam and Corrado (2007), Becchetti et al. (2008), Bruni and Stanca (2008), Meier and Stutzer (2008) and Powdthavee (2008) among others. All these works confirm the findings by psychologists that relational goods are positively associated with SWB. This positive association cannot lead us to conclude that a higher level of their consumption would increase individual (and social) welfare. It could very well be the case that an active social life is not a determinant but an effect of high subjective well being (henceforth SWB). At the end of the story, the direction of causality is still an open question in this literature. A further, related problem is that a hidden common factor could influence both regressand and regressors. For instance, the positive association between income and happiness (Diener et al. 2010) may be, at least partially, determined by unobserved individual traits (optimism, a well balanced personality, etc.) which positively affect both SWB and professional success. To quote Angrist and Pischke (2009, p. 9): ‘‘…the most interesting research in social science is about questions of cause and effect […]. A causal relationship is useful for 1

This change of heart has begun to have an impact on policy making. For instance, the final report by the Stiglitz Commission created by the French President Sarkozy stresses the need for our measurement systems to movetheir focus from economic production to people’s well-being. A similar advice the Office of National Statistics in Britain has been invited to follow by the Prime Minister D. Cameron. On the other side of the Atlantic, three leading happiness scholars, Betsey Stevenson, Alan Krueger and Cass Sunstein all have senior government positions in the Obama administration

2

We refer the interested reader to the comprehensive overview in Diener and Seligman (2004).

123

The Relationship Between Social Leisure and Life Satisfaction

455

making predictions about the consequences of changing circumstances or policies; it tells us what would happen in alternative (or ‘‘counterfactual’’) worlds. In this work, we try to address the causality nexus going from social leisure to SWB. A first important opportunity of dealing with the endogeneity of regressors is offered by the German Socio Economic Panel (GSOEP) which contains both cross–sectional and longitudinal information (from 1984 to 2007) on many variables, including self declared satisfaction with life as a whole and over some domains (work and leisure among them) and various measures of relational life, for a large sample of individuals.3 Panel data are very important in happiness studies because they allow investigators to focus on an assessment of changes in levels of happiness for the same individuals, avoiding the problem of unobserved heterogeneity both cultural and natural. In studying personal relationships it is quite obvious that a cheerful disposition, whether due to genes or to upbringing, will make one’s social life easier and more rewarding.4 But even using panel data techniques the problem remains that intra-individual variation in SWB may affect happiness determinants. This time varying dimension of the endogeneity problem is particularly severe when we consider the relational goods–subjective well being nexus. Just by introspection, it seems quite obvious that not only temperament, but also transient feelings affect our propensity to meet people. To deal with this form of endogeneity we have to find an instrument for our potentially endogenous regressor, i.e. a variable which is not correlated with the disturbance term in the regression but that is correlated with the endogenous regressor.5 Our instrumentation strategy hinges on retirement. Retiring gives us the opportunity (indeed, forces us) to drop habits and to experiment with new activities. In particular, the fall in hours worked is potentially investable in relationships. In other words, there is a chance that retirement helps overcoming the cognitive failures that may have induced workers to consume too little relational goods. Moreover, we tend to socialise with people belonging to our age group, who are therefore likely to retire at the same time as we do. This is another good reason why we expect social activities to increase at this stage of life. This collective, simultaneous increase in the endowment of time may make the double (or multiple) coincidence of wants needed for socializing within easier reach. So, there is an obvious reason to expect that coordination failures depressing the consumption of relational goods will also be less binding for the retired. Our data confirm that retirement does indeed induce people to socialise more. Even if retirement possesses important properties for the solution of the problem, it cannot instrument as such the relational goods indicator because it belongs to the equation as a regressor, i.e.it can have a direct positive (or negative) effect on SWB. This is because in standard economic theory leisure is a good, so retiring means a reduced disutility from work. However, working can certainly be a source of self-realization and identity, as well as of social status. More generally, working allows establishing connections with others. So, stop working could instead bring about a reduction in well being. Finally, retiring may be a choice influenced by one’s own well-being, a second reason not to use it as an 3

The GSOEP is a longitudinal household survey sponsored by the Deutsche Forschungsgemeinschaft and organized by the German Institute for Economic Research (Berlin) and the Center for Demography and Economics of Aging (Syracuse University). We are grateful to these institutes and to the project director Dr. G. Wagner for making this dataset available.

4

Becchetti et al. (2008) and Powdthavee (2008) show that the link between happiness and social life survives the elimination of this fixed component by using respectively German and British panel data.

5

Meier and Stutzer (2008), who concentrate on volunteering, tackle the causality problem by using the collapse of the East Germany volunteering infrastructure.

123

456

L. Becchetti et al.

instrument. For instance, in Germany the mandatory age for retirement is 65 but the law creates a wide window of opportunities for retirement decisions around this age. Our quest for a variable correlated with the time spent in social life lead us to choose as an instrument the sample proportion of the retired by year and region (East vs. West). The rationale for this choice is that the increased availability of time of the retired makes socializing more likely for the retired themselves but also for the people they come in contact with. In fact, the contribution of the retired to the organizational activity that makes social events possible can increase. They may also increase their involvement in home production, thus liberating (potentially social) leisure time for other members of their household, family, etc. For these reasons, we expect our instrument to be a strong one. This prediction is confirmed by the data. The consumption of relational goods is strongly correlated with the ratio of retired in the sample. Our ratio of retired is also clearly exogenous at the individual level and unlikely to have a direct effect on individual well being. Summing up, we create value added in the happiness literature by improved identification of the causal effect of social leisure on life satisfaction, an important step towards the construction of a measurement system focused on the quality of life. Our results emphasize that relational consequences of economic policies need to be carefully taken into account when pursuing the goal of maximising social welfare. The paper is divided into five sections (including introduction and conclusions). The second illustrates the theoretical and methodological background for our analysis. The third and the fourth present and comment our descriptive and econometric findings.

2 Overview of the Literature The concept of ‘Relational goods’ was introduced in economics by Gui (1987) and Uhlaner (1989) to define a set of intangibles ranging from companionship, sympathy and intimacy, to feeling part of a community with the same values or tastes etc. Bardsley and Sugden (2006) borrow from Adam Smith’s ‘‘Theory of Moral Sentiments’’ the term ‘fellowfeeling’, to describe the mental states produced during such non instrumental social interactions. The production process of these goods is the meeting–‘encounter’ in Gui (2005)’s definition–with family and friends or with a wider net of partners.6 Many kinds of social events (association gatherings, cultural or sport events, etc.) can be defined as ‘encounters’. Participating in a political debate, volunteering, applauding at a theatre are examples of relational goods produced on this larger scale. A defining feature of relational goods is that their value crucially depends on the sincerity and genuineness of the people involved. This implies that they can be generated as a by product of some instrumental activity but not exchanged through the market or indeed produced by the state. So they don’t have a price and their value has instead to be estimated. Nor can the estimation be done just by looking at their opportunity cost in terms of labour income given up by choosing leisure. Indeed, leisure includes many heterogeneous activities which can be relational, pseudo-relational (watching TV or reading novels) or non relational. Interestingly, life satisfaction has been found to be negatively correlated with TV viewing, directly in Frey et al. (2007a, b) and indirectly, by reducing the time 6

The concept is central M. Buber’s philosophy. For the Jewish scholar ‘Encounter’ (Begegnung) is an event or situation in which a relation (Beziehung) occurs. He also calls the encounter the sphere of the between. For Buber, ‘All real life is encounter’ ie we only live in relation to others.

123


457

spent in relational activities, in Bruni and Stanca (2008). Frey et al. (2007a, b) find this evidence difficult to reconcile with the theory of revealed preferences, by which any observed choice is utility maximizing, and interpret the finding as suggesting that people do not always act rationally, but often just follow habits and impulses. Indeed Frey and Stutzer (2006a, b) argue that individuals are prone to mispredict utility, through underestimation of adaptation, distorted memories of past experiences and materialistic beliefs fostered by institutions (e.g. marketing) and that these cognitive limits lead them to overconsume goods satisfying extrinsic needs (material goods beside basic necessities) and under-consume goods satisfying intrinsic needs, relational goods among them. Empirical evidence on these distorted choices is offered by these authors by studying commuting. Similar ideas were first developed by Scitosvky who, in his Joyless Economy (1976), advised that we’d be better off spending our time on things that we will not adapt to (meeting friends being a prominent example)rather than getting ‘‘joyless’’ goods, the comfort of which is just temporary. Goods are ‘‘joyful’’ only if they present us with challenge and if their use produces a sense of accomplishment and fulfilment. However the enjoyment of such goods intrinsically requires an effort on the part of the consumer, not just on the part of the producer, an effort which is higher when one starts to consume them. So basically there are high entry costs that may discourage the use of such goods.7 A further problem is that the market can push consumers to non optimal choices by offering inferior substitutes for ‘‘joyful’’ goods, at a lower cost due to technological progress: an obvious case of Baumol’s disease. Genuine relational goods can be more and more crowded out by virtual relational goods.8 A different explanation, by no means alternative to the ‘‘behavioral’’ one described above, hinges on the fact that relational goods, by definition, are not an option freely available at the individual level. An individual’s time use choices may be contingent on the time use choices of others, because the utility derived from leisure time (relational goods) often benefits from (requires) the presence of companionable others. Antoci et al. (2007), Azariadis et al. (2008), Bruni and Stanca (2008), Corneo (2005) and Jenkins and Osberg (2003) develop models starting from this premise that one cannot have a social life unilaterally. Various types of external effects concerning relational goods can be distinguished: there are externalities in the formation of an agent’s social network as the probability of a successful match with a partner increases with the time the agent and the potential partners devote to searching, while a second type of externality concerns the efforts by the agent and the potential partners in cultivating their skills as partners. Finally, there are externalities at the aggregate level since it is easier and more rewarding to participate in an association in a social context characterized by a rich network of associative opportunities. In this respect, Merz and Osberg (2006) find that the proportion of leisure time devoted to social leisure is higher in Lander with more public holidays.

7

For a formalization of Scitovsky arguments see Pugno (2010).

8

Corneo (2005) shows that in the OECD countries the hours spent in front of the TV grow together with the hours of work. An explanation for this phenomenon may be that overworked people will tend to resort to this kind of entertainment just because it requires less energy than more fulfilling leisure activities, relationships among them. In the information age a new impersonality risks to affect everyone: think of Facebook, a Web site that through bodiless sharing promises to combine the aura of intimacy with the safety of distance. The oxymoron is splendidly revealed by the cinematic portrait (The social network) of his inventor Mark Zuckerberg who creates a five-hundred-million-circle of ‘‘friends’’, but is so work-obsessed and withdrawn that he can’t stay close to anyone.

123

458

L. Becchetti et al.

Due to these multi-level net of externalities, equilibria with low socializing may coexist with equilibria with high socializing for groups of individuals as well as for nations as a whole.9 The consumption of relational goods could affect labour supply decisions: when other persons increase their hours of paid work, the probability of a feasible and desirable leisure match falls, thereby decreasing the personal utility of non-work time. The consequences of such strategic complementarities in the enjoyment of leisure are considered in Alesina et al. (2005) and Burda et al. (2008) in analyzing the difference in hours worked between Europe and the US, which has emerged in the 1970’s and has been increasing since then. This difference might not be due to a difference in the tax system, as maintained by Prescott (2004), or in tastes as suggested by Blanchard (2006), instead history (e.g. the first oil shock) and institutions (labour–market regulations) might have simply led otherwise identical Americans and Europeans to coordinate on different equilibria.10 In the ‘‘US’’ equilibrium, individuals work a lot, consume a lot, and have little time for communal activities. In the ‘‘European’’ equilibrium, consumers work less and consume less, but enjoy more common leisure. Finally, the theme of relational goods is at least implicitly present in the vast literature on social capital, which studies the impact of social ties on the productivity of traditional private goods. Higher social participation may bring about social capital accumulation as a by-product. For instance, trust (or empathy) may be reinforced and generalised through social interactions.11 This rhapsodic overview of the recent economic literature on relational goods is far from being complete. However we hope it is enough to convince the reader that the empirical study of the hypothesis that less common leisure leads to lower lifetime utility, on which we report in the following sections, has vast implications for the study of contemporary societies. Subjective assessments of well-being have been used to estimate the shadow value of a wide range of environmental and social conditions, such as air quality and pollution (Welsch 2002, 2006), airport noise (Van Praag and Baarsma 2005), terrorism (Frey et al. 2007a, b), the fear of crime (Moore and Shepherd 2006), marriage (Blanchflower and Oswald 2004; Frey and Stutzer 2006a, b; Johnson and Wu 2002) and unemployment (Clark and Oswald 1994; Di Tella et al. 2001, 2003, Gallie and Russell 1998). The use of happiness data has been proposed as a general method for eliciting market valuation of a non-market good, with the potential to avoid the pitfalls of traditional revealed preference and stated preference methods. In the case of relational goods this would imply for instance trying to infer the value of social relationships from the higher prices of houses in city areas offering better opportunities for socializing, or how long people are willing to travel to reach such areas. This could perhaps be done but only

9

Antoci et al. (2007) show how bounded individual rationality and externalities combine in producing ‘‘social poverty’’ traps.

10 According to these authors one of the strongest pieces of evidence in favor of complementarities across leisure is that an overwhelming share of the population both in Europe and the US takes its two days of leisure during Saturday and Sunday. There would be huge benefits from staggering work so that different people take different days off during the week: this could reduce commuting time and would allow capital to be spread over more workers. The fact that this is not done suggest that the costs in terms of forgone welfare due to less coordinated leisure would be sizable as well. However the relevant complementarities could be across work, rather than leisure. 11 We notice however that the econometric techniques we use are unable to capture these more universal benefits of relational goods.

123


459

relatively to some specific kinds of socializing, because only some kinds of socializing take place in public places, e.g. meeting with friends and family often happen at home. In the case of stated preference methods (e.g. contingent valuation method), individuals are directly asked to value the public good in question. This can be a cognitively demanding task and may induce strategic answers: individuals have an incentive to understate their demand for a good when it positively affects contribution requirements. The life satisfaction approach used in this paper correlates the degree of public goods or public bads with individuals’ reported subjective well-being and evaluates them directly in terms of life satisfaction. Individuals are not asked to value the public good directly, but to evaluate their general subjective well–being, life satisfaction or happiness. This is presumably a cognitively less demanding task and there is no reason to expect strategic behaviour.

3 Descriptive Empirical Findings The empirical analysis is based on the data retrieved by the German Socio Economic Panel (GSOEP) over the period 1984–2007.12 This longitudinal dataset has been widely used for investigating the pattern of subjective life satisfaction and its domains as it offers rich information on the issue (we refer to Van Praag and Ferrer-i-Carbonell 2008). The panel started in 1984 by interviewing residents in the Federal Republic of Germany (West Germany). From 1990 it has included people of the former East Germany. As the investigation is subject to the availability of variables over time, we end up with a sample of 156,232 observations.13 Overall 73.7% of the total observations come from West Germans, 48.5% are males. The age of interviewees ranges from 17 to 76, while people older than 60 represents 20.4%. The main dependent variable is subjective life satisfaction as recorded in the GSOEP personal questionnaire: ‘‘Please answer by using the following scale in which 0 means totally unhappy, and 10 means totally happy. How satisfied are you with your life, all things considered?’’ In Fig. 1 we plot the sample’s average life satisfaction by age: males start with a higher level of life satisfaction, but women declare to be happier during their 30s. Female life satisfaction remains slightly above the one reported by males until the age of 63. From that age on men seem to be slightly happier than women. If we analyze the distribution of average leisure time satisfaction over age, pooling all the observations in the panel (Fig. 2), we notice that the U shaped trend for men is quite similar to that for women.14 From the age of 52–63 women have a higher leisure satisfaction but men do better in their mid 60s. The variable of interest is the time devoted to social leisure. We measure it by creating a Relational Time Index (hence RTI). The index is built by aggregating the information gathered in five questions asking to people how much time they devote to (1) ‘‘attend 12

The data used in this paper was extracted using the Add-On package PanelWhiz for Stata. PanelWhiz (http://www.PanelWhiz.eu) was written by Dr. John P. Haisken-DeNew ([email protected]). See Haisken-DeNew and Hahn (2006) for details. The PanelWhiz generated DO file to retrieve the data used here is available from us upon request.

13 For the purpose of our empirical analysis we did not consider the subsamples of immigrant households and those belonging to of the high income households, as they are likely to increase the heterogeneity of the effect of social leisure over SWB. 14 Question in 2007 SOEP personal questionnaire ‘‘How satisfied are you with your free time, 0 means totally unhappy, and 10 means totally happy’’.

123

460

L. Becchetti et al.

Fig. 1 Average life satisfaction by age and gender, GSOEP 1984–2007

Fig. 2 Average leisure satisfaction by age and gender, GSOEP 1984–2007

social gatherings’’15; (2) ‘‘attend cultural events’’; (3) ‘‘participate in sports’’; (4) ‘‘perform volunteer work’’; (5) ‘‘attend church or religious events’’. We believe that all these activities produce relational goods of the kind described in the previous section. We are also aware that the degree of their productivity in creating or strengthening ties among participants may vary, so we choose to combine all of them. However, our proxy of relational goods is imperfect and does not cover the entire concept which is wider than what can be captured in our survey. What remains outside the scope of our analysis is for instance the amount of relational goods enjoyed in the job place. The time-use questions are codified differently over the SOEP waves. In order to make use of all the available years, we have to reclassify homogeneously the frequencies of attending or participating to a certain event. By preserving the ordinal scale of each relational activity assigned in the SOEP questionnaire, each activity ranges from 0 to 3, where 0 = Never, 1 = Less Frequently, 2 = Every Month, 3 = Every Week. The RTI variable is then simply the mean of the points given to the 5 selected time-use questions where all the components have the same weight16. In this sense RTI assumes values from 15 In the 2007 SOEP personal questionnaire the variable named as ‘‘social gatherings’’ refers to the following question: Which of the following activities do you take part in during your free time? Please check off how often you do each activity: at least once a week, at least once a month, less often, never. Among the possible activity there was: ‘‘Meeting with friends, relatives or neighbours’’. In this case, the variable ‘‘social gathering’’ should include the most common relational goods’ generating activities. For a detailed description of the variables see the ‘‘Appendix’’, Table 8. 16 In computing RTI we assume that each of the five selected relational activities has the same weight or importance in determining the value of the index, so we allow for perfect substitution between the relational activities. We also apply factor analysis and principal component analysis. We start by considering the

123


461

0 to 3. It is likely that the distance from ‘‘every month’’ to ‘‘every week’’ corresponds to a more than proportional increase in sociability than the distance between ‘‘less frequently’’ and ‘‘every month’’. If this is the case, our un-weighted average flattens high intensity responses and may be conceived as a sort of log transform of the true unobserved variable.17 In Table 1 we present the correlation between the different activities. All the selected social leisure activities are positively associated with each others. This is especially true for attending social gatherings, i.e. meeting with friends, relatives or neighbours, and participating to sport activities and for participating to religious activities and volunteering. But none of the activities are really the same thing, so the five questions represent different ways of spending leisure time interacting with others. In Fig. 3 we track the dynamics of our index of social leisure over the life cycle. As before, individual observations are pooled. The decline of the index is strong for both genders until individuals reach the age of 35: in their early 20s most people devote time to social leisure at least once a month, 10 years later this is reduced sharply to once a year. We notice that the time index for women rises after the minimum at 55. For men, the increase is weaker. As anticipated in the introduction, our instrumentation strategy hinges on retirement. The idea came by observing that in our sample people retire (voluntary or involuntary) mainly in their early 60s. In that age bracket we notice a contemporaneous increase in time spent in social life as well as an increase in life and in leisure satisfaction. In the pooled GSOEP waves 1984–2007, the share of retired individuals by age jumps up from 60 to 65 (from 30 to 90 percent, see Fig. 4). Inspecting again the age-happiness pattern we find that the increase in life and leisure satisfaction is well visible in the first part of the 60s. If we exclude what happens after the mid-seventies, the average life satisfaction as a function of age exhibits the U-shape trend found in many happiness studies (and summarised in Frijters and Beatton 2008). The age pattern in life satisfaction is paralleled by a similar, and more pronounced, U-shape in leisure satisfaction. In particular, there is a spike in leisure satisfaction among people of 59 and 63 years old. A similar conclusion comes looking at the average annual working hours which drops both for women and men after the mid of their 50s (see Fig. 5).

Footnote 16 continued correlation among the five selected questions (see Table 1) with the aim of aggregating them in a new relational composite indicator (Rel_CI). By applying the procedure used by Nicoletti et al. (2000) we selected a subset of principal components that accounts for the largest amount of variance among the individual sub-indicators. The other technical steps deal with the construction of the weights from the matrix of factor loadings after rotation. The final composite indicator (Rel_CI) we obtain is: Rel_CI = (0.17 * social gathering) ? (0.09* volunteering activities) ? (0.31*cultural activities) ? (0.17 * sport) ? (0.25 * religious activity). Findings obtained with such indicator are shown in Table 9 of the ‘‘Appendix’’. Even if the distribution of the index is different, our findings confirm the significance of the RTI index. 17 We underline that the use of an averaged value rather than of each single component is mainly motivated by data availability: none of the five selected questions are asked in the same year. Subject to the availability of at least one of the variables entering the RTI, we run our empirical analysis on the following waves: 1984–1986, 1998, 1992, 1994–1999, 2001, 2003, 2005 and 2007. In order to check the robustness of our findings to alternative ways of calculating the index we compute RTI2 which gives the following weights to different intensities of relational activities: 0 = never, 0.08 = once a year, 1 = once a month, 4 = once a week. The scale takes as point of reference the level of intensity equal to once a month, so that we assign 1 to once a month, 4 to once a week (as in a month there may be four weeks), 1/12 to less frequently because we interpret it as at least once a year (as in a year there are 12 months), and zero for never. Results are substantially unchanged and available in Table 10 of the ‘‘Appendix’’.

123

462

L. Becchetti et al.

Table 1 Correlation matrix of the five time-use questions used for the RTI (SOEP 1984–2007) Social gatherings

Volunteering activities

Cultural activities

Sport

Social gatherings

1.0000

Volunteering activities

0.0958

Cultural activities

0.1493

0.1238

1.0000

Sport

0.2203

0.1489

0.1536

1.0000

Religious activities

0.0503

0.2027

0.0805

0.0165

Religious activities

1.0000

1.0000

92,295 individual observations

These findings are strongly consistent with those found by Engfer (2009) analysing time budget data collected by the Federal Statistical Agency of Germany in 2001/2. He concludes that prime–age adults suffer from time pressure and that their leisure satisfaction rises significantly around the retirement age. Indeed, after the retirement transition the time previously devoted to work is reallocated to socializing with family and friends, as well as to household maintenance and other generic leisure activities. Also, after retirement, the share of people feeling incapable of spending enough time with their spouse or friends drops consistently.

4 Econometric Analysis Based on these descriptive findings we go on to test the relational goods-happiness nexus through the following steps: (1) we start with a base specification (2) we add our relational index to this base specification; (3) we perform an IV estimate in which the relational index is instrumented; (4) we do robustness checks with various subsamples and estimation models. 4.1 The Baseline Model Our base specification includes the explanatory variables typically found in happiness regressions: marital and employment status, changes in employment and marital status, years of education, health status, number of children in the household, log of equivalised real household income, an East/West dummy, house ownership. We also include time dummies and age categories.18 Opinions on the inclusion of year dummies in these types of estimates are mixed. On the one side, it is observed that these dummies capture aggregate shocks to macroeconomic performance, political events etc. whose influence can be important so that excluding them would cause serious omitted variable bias. On the other side, when fixed effects are included and age and age squared are entered as regressors, including year dummies would create perfect collinearity: this is why, following Clark (2007) we use age categories instead. In fact, this choice is crucial for estimating the 18 Differently from two previous studies which investigate the age-happiness relationship on the same data (Frijters and Beatton 2008; Van Landeghem, 2008), we do not restrict the analysis to West Germans, as in Frijters and Beatton (2008), and do not work only on the balanced panel, as in Van Landeghem (2008). Our main results are however supported also in these two specific subsamples. Results are omitted for reasons of space and available upon request.

123


463

Fig. 3 Average RTI index by age and gender, GSOEP 1984–2007

Fig. 4 Mean of retired people by age and gender, GSOEP 1984–2007 (pooling observations)

Fig. 5 Average annual working hours (in the market) by age and gender, GSOEP 1984–2007 (pooling observations)

relationship between age and SWB: if year dummies are not included, as in Frijters and Beatton (2008), the U-shaped relationship found when using age categories disappears and SWB is monotonically decreasing in age. This is in our opinion due to the fact that the panel, even the unbalanced one, ages, so that a disproportionate number of observations on the young come from the first years. These were happy years for Germany, presumably because of the reunification, so that excluding year dummies from the regression biases the coefficient on age.19 19

For a different opinion, focused on entry and survivorship bias, see Frijters and Beatton (2008).

123

464

L. Becchetti et al.

In the first four columns of Table 2 we present the following specifications: (1) the base equation; (2) the base equation plus the retirement variable; (3) the base equation plus the RTI variable; (4) the base equation plus the retirement and RTI variables. Our findings confirm the ‘‘almost stylised facts’’ of the happiness literature, from the positive and significant effect of household income, marriage status to the negative and significant effect of separation, unemployment and bad health status (Table 2, column 1). A distinctive element of ours with respect to most papers in the literature is our use of equivalised household income computed following the OECD equivalence scale,20 together with the number of children variable. This makes the children variable positive and significant. In this way we are able to disentangle two effects caused by having a child: a negative one represented by the reduction of per capita income within the household and a positive one represented by the psychological value of being a parent. Both the retirement and the relational good variables are positive and significant when separately considered and when jointly introduced in the estimates (Table 2, columns 2–4). The rationale for the retirement effect is twofold. On the one side, consistently with the standard assumption in economics that leisure is a good, people will enjoy retirement as the disutility from work ceases. However another complementary explanation is that with retirement an increase in the quantity and quality of social life is possible. Indeed, as we have seen in the previous section on the descriptive statistics, and as demonstrated by Engfer (2009) on time budget data for people reaching pension age, people become more satisfied with their everyday agenda when they retire. We notice that the coefficients of the age cohorts from 59 to 61 are still positive and significant even when we include the RTI and retirement variables. We thought of two plausible arguments reconciling such findings with the hypothesis that what explains the surge in SWB at the age of retirement is indeed retirement: (1) hours worked are reduced in this age category even for those still working, as emerges from Fig. 3, so the disutility of work is reduced for them, even if to a lower degree than for the retired; (2) if the consumption of relational goods increases even for those still working, one could argue that, when many of the peers of the non retired are retired, it is easier for all in this cohort to avoid the relational poverty trap. People are better off when their reference group starts to retire, whether or not they themselves retire.21 4.2 Tackling the Endogeneity Problem: The IV Estimates In our base estimates the coefficient of the relational time variable is significant and strongly positive, even when permanent personality traits are netted out by fixed effect estimation. This result will be found to be robust to different subsamples and estimation methods (see Sect. 4.3). 20 Equivalised income is household income which is adjusted by using an equivalence scale to take into account the size and composition of the household. Here we use the ‘‘OECD equivalence scale’’ which assigns a value of 1 to the first household member, of 0.7 to each additional adult and of 0.5 to each child. This scale (also called ‘‘Oxford scale’’) was mentioned by OECD (1982) for possible use in ‘‘countries which have not established their own equivalence scale’’. For this reason, this scale is sometimes labeled ‘‘(old) OECD scale’’. Adoption of different equivalent scales or of the simple household income in the estimate does not affect the substance of our findings. Evidence is omitted and available upon request. 21 Of course it is not difficult to think of other (concurrent) explanations: it is possible that those who retire later hold particularly psychologically rewarding jobs, or, due to a relatively strong work ethics and competitive attitude take pride from working at a later age. However we have not attempted to disentangle these various possible effects.

123


465

Table 2 The effect of Relational Goods on Life Satisfaction: GSOEP, 1984–2007 (fixed effect regression) Variables

Base

Base retired

RTI Retired

Base RTI

Base retired_RTI

IV_2

IV_3

0.216*** (0.012)

0.215*** (0.012)

2.141*** (0.529)

1.645*** (0.443)

0.233*** (0.028)

0.200*** (0.028)

0.208*** (0.027)

0.237*** (0.024)

Log household income

0.193*** (0.013)

0.192*** (0.013)

0.207*** (0.015)

0.206*** (0.015)

0.180*** (0.016)

0.187*** (0.015)

House owner

0.078*** (0.016)

0.079*** (0.016)

0.078*** (0.018)

0.079*** (0.018)

0.076*** (0.018)

0.077*** (0.017)

No. year of education

0.004 (0.006)

0.002 (0.006)

0.005 (0.006)

0.003 (0.006)

0.021*** (0.008)

0.017** (0.007)

Unemployed

-0.199*** (0.020)

-0.178*** (0.020)

-0.213*** (0.023)

-0.193*** (0.023)

-0.202*** (0.024)

-0.200*** (0.023)

Job loss

-0.094*** (0.023)

-0.093*** (0.023)

-0.112*** (0.029)

-0.108*** (0.029)

-0.101*** (0.032)

-0.103*** (0.030)

Full-time job

0.314*** (0.016)

0.359*** (0.017)

0.325*** (0.019)

0.370*** (0.019)

0.397*** (0.020)

0.390*** (0.019)

Part-time job

0.151*** (0.018)

0.182*** (0.018)

0.155*** (0.022)

0.186*** (0.022)

0.143*** (0.025)

0.154*** (0.024)

Vocational training

0.238*** (0.028)

0.261*** (0.028)

0.252*** (0.033)

0.276*** (0.033)

0.290*** (0.036)

0.287*** (0.034)

Marg. part-time job

0.068*** (0.021)

0.083*** (0.021)

0.056** (0.027)

0.071*** (0.027)

-0.016 (0.037)

0.006 (0.034)

Military service

-0.015 (0.053)

0.010 (0.053)

-0.024 (0.070)

0.003 (0.070)

-0.110 (0.082)

-0.080 (0.077)

Married

0.148*** (0.028)

0.154*** (0.028)

0.181*** (0.032)

0.188*** (0.032)

0.405*** (0.068)

0.349*** (0.059)

Get married

0.245*** (0.024)

0.242*** (0.024)

0.220*** (0.031)

0.217*** (0.031)

0.198*** (0.035)

0.203*** (0.033)

Separated

-0.095* (0.055)

-0.090* (0.055)

-0.042 (0.063)

-0.038 (0.063)

0.186** (0.091)

0.128 (0.082)

Get separated

-0.319*** (0.058)

-0.320*** (0.058)

-0.306*** (0.076)

-0.307*** (0.076)

-0.331*** (0.083)

-0.324*** (0.079)

Divorced

0.094** (0.045)

0.097** (0.045)

0.118** (0.050)

0.121** (0.050)

0.361*** (0.082)

0.300*** (0.072)

Get divorced

-0.074* (0.045)

-0.074* (0.045)

-0.073 (0.057)

-0.073 (0.056)

-0.117* (0.063)

-0.106* (0.061)

Widowed

-0.222*** (0.057)

-0.250*** (0.057)

-0.194*** (0.062)

-0.223*** (0.062)

-0.236*** (0.057)

-0.233*** (0.055)

Child in HH

0.040*** (0.009)

0.041*** (0.009)

0.037*** (0.010)

0.038*** (0.010)

0.071*** (0.013)

0.063*** (0.012)

Hospital stay

-0.181*** (0.010)

-0.179*** (0.010)

-0.177*** (0.013)

-0.175*** (0.013)

-0.121*** (0.020)

-0.134*** (0.018)

Occupational disable

-0.261*** (0.023)

-0.278*** (0.023)

-0.238*** (0.025)

-0.256*** (0.025)

-0.214*** (0.027)

-0.225*** (0.025)

West Germany

0.286*** (0.064)

0.286*** (0.064)

0.246*** (0.072)

0.247*** (0.072)

0.179** (0.070)

0.197*** (0.067)

Age 17_19

-0.220 (0.210)

-0.203 (0.210)

-0.010 (0.253)

0.007 (0.252)

0.214 (0.279)

0.161 (0.266)

123

466

L. Becchetti et al.

Table 2 continued Variables

Base

Base retired

Base RTI

Base retired_RTI

IV_2

IV_3

Age 20_22

-0.355* (0.202)

-0.339* (0.201)

-0.132 (0.243)

-0.117 (0.242)

0.117 (0.270)

0.057 (0.257)

Age 23_25

-0.335* (0.193)

-0.314 (0.193)

-0.148 (0.232)

-0.127 (0.232)

0.089 (0.257)

0.034 (0.244)

Age 26_28

-0.329* (0.184)

-0.300 (0.184)

-0.127 (0.221)

-0.098 (0.221)

0.113 (0.244)

0.059 (0.233)

Age 29_31

-0.283 (0.175)

-0.247 (0.175)

-0.086 (0.210)

-0.050 (0.210)

0.168 (0.233)

0.112 (0.222)

Age 32_34

-0.280* (0.167)

-0.237 (0.166)

-0.087 (0.200)

-0.046 (0.199)

0.124 (0.219)

0.081 (0.208)

Age 35_37

-0.260* (0.158)

-0.211 (0.157)

-0.101 (0.189)

-0.054 (0.188)

0.053 (0.204)

0.026 (0.195)

Age 38_40

-0.246* (0.149)

-0.190 (0.149)

-0.082 (0.178)

-0.028 (0.178)

0.015 (0.191)

0.004 (0.183)

Age 41_43

-0.218 (0.140)

-0.155 (0.140)

-0.086 (0.167)

-0.026 (0.167)

-0.024 (0.179)

-0.024 (0.172)

Age 44_46

-0.205 (0.132)

-0.135 (0.131)

-0.084 (0.156)

-0.017 (0.156)

-0.022 (0.167)

-0.021 (0.160)

Age 47_49

-0.193 (0.123)

-0.116 (0.123)

-0.093 (0.146)

-0.019 (0.146)

-0.028 (0.156)

-0.025 (0.149)

Age 50_52

-0.172 (0.115)

-0.090 (0.115)

-0.075 (0.136)

0.004 (0.136)

-0.026 (0.144)

-0.018 (0.138)

Age 53_55

-0.144 (0.107)

-0.057 (0.107)

-0.078 (0.125)

0.006 (0.126)

-0.057 (0.134)

-0.041 (0.128)

Age 56_58

-0.019 (0.098)

0.067 (0.098)

0.014 (0.115)

0.097 (0.115)

0.004 (0.124)

0.028 (0.119)

Age 59_61

0.162* (0.090)

0.225** (0.090)

0.213** (0.105)

0.272*** (0.105)

0.121 (0.118)

0.160 (0.112)

Age 62_64

0.334*** (0.082)

0.344*** (0.082)

0.363*** (0.095)

0.371*** (0.095)

0.153 (0.116)

0.209* (0.108)

Age 65_67

0.434*** (0.075)

0.407*** (0.075)

0.455*** (0.086)

0.428*** (0.086)

0.144 (0.118)

0.217** (0.107)

Age 68_70

0.395*** (0.067)

0.363*** (0.067)

0.401*** (0.077)

0.369*** (0.077)

0.064 (0.115)

0.142 (0.103)

Age 71_73

0.352*** (0.060)

0.326*** (0.060)

0.355*** (0.067)

0.329*** (0.067)

0.045 (0.103)

0.118 (0.092)

Age 74_76

0.282*** (0.052)

0.261*** (0.052)

0.302*** (0.060)

0.283*** (0.060)

0.037 (0.089)

0.101 (0.080)

Age 77_79

0.162*** (0.041)

0.149*** (0.041)

0.206*** (0.050)

0.193*** (0.050)

0.048 (0.064)

0.085 (0.059)

Time dummies

Yes

Yes

Yes

Yes

Yes

Yes

1992

0.788*** (0.054)

0.826*** (0.054)

0.743*** (0.063)

0.779*** (0.063)

0.685*** (0.072)

0.709*** (0.068)

Constant

5.222*** (0.141)

5.107*** (0.141)

4.806*** (0.163)

4.696*** (0.164)

Observations

271,274

271,274

179,456

179,456

172,537

172,536

Number of ID

36,250

36,250

35,818

35,818

28,899

28,899

R-squared

0.042

0.043

0.044

0.045

-0.202

-0.091

123


467


Base

Base retired

Base RTI

Base retired_RTI

IV_2

IV_3

F-first-excluded

32.48

28.30

Hansen overidentification test (p value)

0.4586

0.0721

Davidson-MacKinnon test of exogeneity (p value)

5.9e-10

8.7e-08

Robust standard errors in parentheses, stars for significance levels: *** p \ 0.01, ** p \ 0.05, * p \ 0.1. Omitted dummies: age category [79, working zero hours, being single. IV estimates: RTI instrumented by sample and national retirement patterns. Instruments for IV_2 regression: QuotaAge (share of retired individuals of the same age cohort in the sample at time t) and QuotaAge_1 (share of retired individuals of the age cohort younger than one year in the sample at time t). Instruments for IV_3 regression: QuotaAge, QuotaAge_1 and SexRetAvgAge (national average retirement entrance age of the opposite sex, respecting the year and the regional location of the individual at time t)

However the direction of causality could run from time-demeaned subjective well being to time-demeaned intensity of social contacts. Our attempt here is to deal with this two-way causation type of endogeneity. In what follows we explain our strategy for estimating a within group instrumental variable (IV) regression: the results are presented in Table 2, column 5–6. As anticipated in the introduction, our identification strategy hinges on retirement. However we do not use individual retirement as an instrument because our analysis in the previous section strongly suggests that an exclusion restriction for this variable is not possible. Instead, we adopt the proportion of retirees in the population sample by age cohort, for each year. Given the large number of observations available we are quite confident that the sample statistic does conveniently approximate that of the entire population22 and retains as well the characteristics of not being influenced by the observed individual retirement decision. Therefore, we make use of two instrumental variables: the probability that people as old as the individual may retire (QuotaAge) and the probability that people 1 year younger than the individual may retire (QuotaAge_1). Results for this two–stages least squared estimation are shown in Table 2, column 5. It is plausible that these two probabilities do not affect directly subjective well being since they are not choice variables for the individual and therefore cannot be related to the time-varying psychological factors captured in the error of our structural equation. Hence, on logical grounds, the retirement age patterns of the sample seem to be reasonably valid instruments since they it is likely that they are only indirectly associated with the individual life satisfaction. The validity of instruments used is commonly tested through the Hansen test. Because the p value is greater than 0.05, we do not reject the null hypothesis of lack of correlation between the instrumental variables and the error term of the structural regression. In other words, we can conclude that our instruments are valid. At the same time, these two instruments are enough correlated to the individual level of RTI to be relevant in predicting the endogenous regressor. The F statistic for joint significance of instruments in the first stage regression of the endogenous regressor is much higher than 10, the critical value suggested by Stock et al. (2002) as indicating a weak instrument. 22

Bo¨rsch-Supan and Schnabel (1999) in their overview on the German Social Security system, as well as Berkel and Bo¨rsch-Supan (2003) in their estimations of the long term impact of reforms on retirement decisions in Germany made use of the retirement statistics drawn from the GSOEP to describe the national figures. Proportion of retirees and mean retired people by gender over the years, by West and East Germany are shown in Table 11 of the ‘‘Appendix’’.

123

468

L. Becchetti et al.

An important statistic to compute in the instrumented specification is the Davidson and MacKinnon (1993) test of exogeneity for a fixed-effect regression estimated via instrumental variables, where the null hypothesis states that an ordinary least squares (OLS) estimator of the same equation would yield consistent estimates. Our F statistic strongly rejects the null of exogeneity and confirms the need for instruments (Table 2, column 5). The instrumented RTI coefficient remains positive and significant even though the fixed effect IV coefficient is quite higher than the fixed effect OLS coefficient (almost ten times). This might be due to the limited cross-sectional variability of our instrument. Consider however that our principal aim at this stage of research is not to quantify the shadow value of social leisure but rather to verify its existence and to sign it. So far, our results are not inconsistent with the possibility that a causal effect of social leisure on SWB exists and is positive. The individual retirement observation nonetheless contributes to the sample average even though its contribution is negligible. To overcome this problem we added a third variable: the average retirement entrance age in Germany.23 Actually the data is detailed by year, sex and regional location (West and East Germany). In order to satisfy the requirement of validity of our instrument, we designed a variable which assigns to each individual the average retirement entrance age of the opposite sex, respecting the year and the regional location (SexRetAvAge). The relevance of this last instrument per se, confirmed by our tests, is not immediate and needs to be explained. A higher average retirement age implies more distance from retirement in the worker’s time horizon and a lower probability of earlier retirement. Since relational goods are jointly produced and consumed a lower probability of earlier retirement of other individuals in the same age cohort who are potential relational goods partners matters. This implies that, in the end, this third instrument is correlated with the previous two as it is relevant in determining the time one individual can devolve to relational activities. Again, from a technical point of view, the variable cannot be accused of being related to the error term in the life satisfaction specification but can be shown to be correlated with the endogenous variable of interest. Besides the econometric requirements, the national average age of retirement is more correlated with the individual retirement choice than the average age of retirement of the opposite sex. The introduction of this additional variable among the instruments does not modify substantially our diagnostics but reduces the magnitude of the coefficient and the standard error of our endogenous variable (Table 2, column 6). Note that the significance of the 55–57 up to 59–61 age categories disappears in these second set of IV estimates and that the only significant age categories remained are the 62–64 and 65–67 ones. This finding does not contradict our hypothesis that the upward bump in happiness-age function found in the early 60s may be determined (at least for a relevant part) by the retirement/increase in social leisure combination. 4.3 Robustness in Subsample Splits Table 3 shows how our findings replicate in different subsamples (women, men, East and West Germans, occupationally disabled24 and not, registered as unemployed and not). 23 National statistics are supplied by the German association of public pension providers (VerbandDeutscherRentenversicherungstra¨ger–VdR) and cover the period 1989–2005. The data elaboration we use here is by Sackmann (2007). 24 Any person whose capacity for social and/or occupational integration is severely restricted by an impairment or reduction of their physical and/or mental capacity is eligible for the aid awarded by the social assistance.

123

0.023

Individuals

R2

0.047

Individuals

R2

0.045

R2

1.523 (0.887)

0.040 (0.042)

89,285

14,812

RTI

Retired

Observations

Individuals

IV_2

92,812

18,337

Individuals

0.225*** (0.017)

RTI

Observations

0.074* (0.037)

Retired

Base retired RTI

92810

18337

Observations

0.225*** (0.017)

RTI

Base RTI

140227

18548

Observations

0.100*** (0.032)

Retired

Base retired

Women

14,087

83,252

0.475*** (0.042)

2.632*** (0.653)

0.094

17,481

86,646

0.209*** (0.016)

0.491*** (0.040)

0.094

17481

86646

0.210*** (0.016)

0.073

17702

131047

0.471*** (0.035)

Men

6,193

36,293

0.223 (0.102)

4.96** (2.206)

0.045

7,546

37,646

0.203*** (0.027)

0.365*** (0.057)

0.043

7546

37646

0.206*** (0.027)

0.045

7611

54231

0.376*** (0.047)

East

22,882

135,987

0.180*** (0.031)

1.840*** (0.529)

0.046

28,707

141,810

0.213*** (0.013)

0.206*** (0.031)

0.046

28707

141810

0.0214*** (0.013)

0.044

29115

217043

0.208*** (0.027)

West

3,651

17,438

0.243* (0.091)

2.458*** (0.915)

0.065

5,651

19,438

0.368*** (0.042)

0.371*** (0.066)

0.062

5651

19438

0.377*** (0.041)

0.055

6076

30269

0.403*** (0.058)


26,721

152,584

0.195*** (0.031)

1.844*** (0.62)

0.052

34,155

160,018

0.179*** (0.012)

0.219*** (0.030)

0.057

34155

160018

0.179*** (0.012)

0.037

34720

241005

0.210*** (0.025)

Not occupational disable

3,926

1,3161

0.271*** (0.091)

1.264 (1.827)

0.079

7,859

17,094

0.197*** (0.054)

0.282*** (0.085)

0.075

7859

17094

0.198*** (0.054)

0.075

8778

25184

0.261*** (0.061)

Unemployed

Table 3 Robustness in subsample splits: the effect of retirement and Relational Goods on Life Satisfaction, GSOEP 1984–2007 (fixed effect regression)

27,337

155,155

0.110*** (0.031)

1.837*** (0.512)

0.047

34,544

162,364

0.216*** (0.012)

0.147*** (0.030)

0.052

34544

162362

0.217*** (0.012)

0.023

35142

246090

0.147*** (0.026)

Not unemployed

The Relationship Between Social Leisure and Life Satisfaction 469

123

123

0.862

11.04

0.507

F-first-excluded

Hansen test p value

0.180

27.90

14087

83251

0.479*** (0.040)

2.044*** (0.468)

0.378

23.38

Men

0.470

3.98

6193

36293

0.247 (0.087)

4.158** (1.685)

0.348

4.21

East

0.002

35.39

22882

135987

0.199*** (0.028)

0.641* (0.360)

0.616

29.42

West

0.908

13.16

3651

17438

0.148* (0.088)

2.535*** (0.905)

0.561

13.16


0.282

23.90

26721

152583

0.202*** (0.029)

1.335*** (0.467)

0.507

21.59


0.907

2.74

3926

13161

0.273*** (0.090)

1.089 (1.732)

0.740

3.67

Unemployed

0.214

27.58

27337

155154

0.118*** (0.030)

1.466*** (0.433)

0.590

31.55

Not unemployed

Robust standard errors in parentheses, stars for significance levels: *** p \ 0.01, ** p \ 0.05, * p \ 0.1. Omitted dummies: age category [79, working zero hours, being single. IV estimates: RTI instrumented by sample and national retirement patterns. Instruments for IV_2 regression: QuotaAge (share of retired individuals of the same age cohort in the sample at time t) and QuotaAge_1 (share of retired individuals of the age cohort younger than one year in the sample at time t). Instruments for IV_3 regression: QuotaAge, QuotaAge_1 and SexRetAvgAge (national average retirement entrance age of the opposite sex, respecting the year and the regional location of the individual at time t)

89285

14812

Individuals

0.026 (0.041)

Retired

Observations

2.077*** (0.759)

RTI

IV_3

10.52

Hansen test p value

Women

F-first-excluded

Table 3 continued

470 L. Becchetti et al.


471

The retirement effect on life satisfaction is much stronger for males than for females, while the enjoyment of relational life is similar for the two sexes. This may be interpreted in the sense that job-induced relational poverty during their working years is much stronger for males, who work longer hours and have full time jobs more often than women. Being retired attracts a significant coefficient for both unemployed and disabled workers.25 In particular, among those who were registered as unemployed, the retirement effect is much higher than for those who were not: it seems likely that this is due to the end of a condition in which the individual was carrying a social stigma. Indeed, unemployment is found to be very detrimental for SWB in many studies, starting from Clark and Oswald (1994). The RTI variable is always significant in the observed subsamples even when we introduce the retirement variable. When instrumenting the endogenous regressor using the two set of instruments explained above, we find results similar for the two estimations over the groups (even though weaker for the West Germans and Women with strong significance only in one of the two cases). 4.4 Robustness Across Estimation Methods In this section we would like to address some of the limitations that our estimation strategy may have. We start by checking whether the effect of relational goods on happines remains significant in relevant subsamples if we modify the choices on how to include age, time and individual fixed effects. As described above (see Sect. 4.1), the benchmark model is estimated with a fixed effect regression including time dummies and age categories. Analysing here the possible alternative specifications with their drawbacks and advantages allows us to better justify our estimation choices. The first choice to make was how to introduce age in the regression: nearly all recent papers enter terms in age and age squared. Frijters and Beatton (2008) show that in most of these studies the effect of the linear term in age is always negative, whilst that of agesquared is positive, indicating a U-shape. Although this seems to be a typical finding in happiness regressions, we prefer not to impose a rigid functional form on age. Following Clark (2007) and Van Landeghem (2008), we use dummies representing age-bounded categories. Each age category comprises 3 years: 17–19, 20–22 … 77–79, and the omitted category is the age group containing individuals in their eighties. Another issue is whether to estimate a pooled cross-sectional or a fixed effects regression. In Table 4 we present the pooled regression results where we compare the two possible ways to enter age. The coefficient of the relational time index is strongly significant and positive over all subsamples and it maintains almost the same value regardless of the way age enters the regression, even when we introduce the retirement variable. On the contrary, the retirement variable has a positive effect on life satisfaction when age is entered in a quadratic form and the opposite when we use age categories (negative impact). 25 Besides old age pensions the German welfare system provides disability benefits to workers of all ages not able to carry on a regular employment. If this inability is complete they receive full old age benefits, the so called disability pension (‘‘Erwerbsunfa¨higkeitsrente’’, EU). A person that can work only half of the time or less compared to a healthy person received two-thirds of old age benefits (‘‘Berufsunfa¨higkeitsrente’’, BU). In the 1970s and early 1980s, the German jurisdiction has interpreted the rules on disability very broadly, in particular the applicability of the first rule. Disability is the most important pathway to retirement for civil servants: 47% of those who retired in the year 1999 used disability retirement. Hence we may consider the disabled group as a hybrid set (of not fully- irregularly employed partially subsidized workers) which stands between full employment and straight unemployment. See Borsch-Supan and Wilke (2004).

123

123

Yes

Yes

271,274

0.112

Age categories

Time dummies

Observations

R-squared

271,274

0.107

Time dummies

Observations

R-squared

Yes

Yes

179,456

0.125

Age categories

Time dummies

Observations

R-squared

0.431*** (0.008)

Yes

Yes

RTI

Age AgeSquared

Time dummies

Pooled 2

0.416*** (0.008)

RTI

Pooled 1

Yes

Yes

Age AgeSquared

0.213*** (0.015)

Retired

Pooled 2

0.012 (0.017)

Retired

Pooled 1

All sample

Yes

Yes

0.464*** (0.011)

0.119

92,810

Yes

Yes

0.452*** (0.011)

0.099

140,227

Yes

Yes

0.118*** (0.020)

0.104

140,227

Yes

Yes

-0.067*** (0.023)

Women

Yes

Yes

0.406*** (0.011)

0.141

86,646

Yes

Yes

0.388*** (0.011)

0.124

131,047

Yes

Yes

0.512*** (0.025)

0.130

131,047

Yes

Yes

0.323*** (0.028)

Men

Yes

Yes

0.460*** (0.019)

0.126

37,646

Yes

Yes

0.436*** (0.019)

0.114

54,231

Yes

Yes

0.567*** (0.035)

0.122

54,231

Yes

Yes

0.244*** (0.041)

East

Yes

Yes

0.422*** (0.009)

0.100

141,810

Yes

Yes

0.407*** (0.009)

0.082

217,043

Yes

Yes

0.183*** (0.017)

0.087

217,043

Yes

Yes

0.012 (0.019)

West

Yes

Yes

0.706*** (0.027)

0.133

19,438

Yes

Yes

0.675*** (0.027)

0.094

30,269

Yes

Yes

0.183*** (0.040)

0.104

30,269

Yes

Yes

0.096** (0.041)


Yes

Yes

0.392*** (0.008)

0.106

160,018

Yes

Yes

0.380*** (0.008)

0.090

241,005

Yes

Yes

0.258*** (0.017)

0.094

241,005

Yes

Yes

0.053*** (0.020)


Table 4 Robustness check in alternative models: pooled regression with age categories (1) or quadratic age specification (2)

Yes

Yes

0.430*** (0.031)

0.131

17,094

Yes

Yes

0.421*** (0.031)

0.122

25,184

Yes

Yes

0.454*** (0.056)

0.127

25,184

Yes

Yes

0.401*** (0.057)

Unemployed

Yes

Yes

0.425*** (0.008)

0.100

162,362

Yes

Yes

0.413*** (0.008)

0.082

246,090

Yes

Yes

0.134*** (0.016)

0.087

246,090

Yes

Yes

-0.068*** (0.018)

Not unemployed


0.416*** (0.008)

Yes

Yes

179,456

0.125

RTI

Age categories

Time dummies

Observations

R-squared

179,456

0.121

Observations

R-squared

0.115

92,810

Yes

Yes

0.464*** (0.011)

0.124*** (0.025)

0.119

92,810

Yes

Yes

0.451*** (0.011)

-0.038 (0.028)

0.115

92,810

Women

0.138

86,646

Yes

Yes

0.406*** (0.011)

0.511*** (0.031)

0.142

86,646

Yes

Yes

0.389*** (0.011)

0.334*** (0.034)

0.134

86,646

Men

0.120

37,646

Yes

Yes

0.456*** (0.019)

0.553*** (0.042)

0.126

37,646

Yes

Yes

0.437*** (0.019)

0.274*** (0.048)

0.116

37,646

East

0.097

141,810

Yes

Yes

0.421*** (0.009)

0.171*** (0.021)

0.100

141,810

Yes

Yes

0.407*** (0.009)

0.018 (0.024)

0.096

141,810

West

0.127

19,438

Yes

Yes

0.708*** (0.027)

0.183*** (0.050)

0.133

19,438

Yes

Yes

0.677*** (0.027)

0.110** (0.051)

0.126

19,438


0.103

160,018

Yes

Yes

0.390*** (0.008)

0.249*** (0.020)

0.106

160,018

Yes

Yes

0.380*** (0.008)

0.059** (0.024)

0.102

160,018


0.129

17,094

Yes

Yes

0.428*** (0.031)

0.421*** (0.068)

0.133

17,094

Yes

Yes

0.420*** (0.031)

0.366*** (0.069)

0.127

17,094

Unemployed

0.097

162,362

Yes

Yes

0.425*** (0.008)

0.132*** (0.019)

0.100

162,362

Yes

Yes

0.412*** (0.008)

-0.047** (0.022)

0.097

162,362

Not unemployed

Robust standard errors in parentheses, stars for significance levels: *** p \ 0.01, ** p \ 0.05, * p \ 0.1. Omitted dummies: age category [79, working zero hours, being single

Same controls as in the benchmark model with time dummies

Yes

Yes

Time dummies

0.431*** (0.008)

RTI

Age AgeSquared

0.210*** (0.018)

Retired

Pooled 2

0.031 (0.021)

Retired

Pooled 1

179,456

0.120

R-squared

All sample

Observations

Table 4 continued


123

123

36,250

R-squared

Number of ID

271,274

0.036

36,250

Observations

R-squared

Number of ID

35,818

R-squared

Number of ID

RTI

0.213*** (0.011)

0.040

Observations

Fixed effect 2

Yes

179,456

Age Categories

0.204*** (0.011)

RTI

Fixed effect 1

Yes

Age AgeSquared

RTI

Retired

0.358*** (0.022)

0.037

Observations

Fixed effect 2

Yes

271,274

Age Categories

0.221*** (0.024)

Retired

Fixed effect 1

All sample

0.222*** (0.015)

18,337

0.033

92,810

Yes

0.215*** (0.015)

18,548

0.031

140,227

Yes

0.204*** (0.031)

18,548

0.031

140,227

Yes

0.080** (0.033)

Women

0.206*** (0.015)

17,481

0.053

86,646

Yes

0.195*** (0.015)

17,702

0.047

131,047

Yes

0.614*** (0.033)

17,702

0.050

131,047

Yes

0.461*** (0.035)

Men

0.226*** (0.025)

7,546

0.038

37,646

Yes

0.204*** (0.024)

7,611

0.033

54,231

Yes

0.550*** (0.046)

7,611

0.037

54,231

Yes

0.341*** (0.048)

East

0.208*** (0.012)

28,707

0.042

141,810

Yes

0.201*** (0.012)

29,115

0.037

217,043

Yes

0.317*** (0.025)

29,115

0.038

217,043

Yes

0.194*** (0.027)

West

0.349*** (0.039)

5,651

0.043

19,438

Yes

0.336*** (0.038)

6,076

0.038

30,269

Yes

0.464*** (0.057)

6,076

0.042

30,269

Yes

0.396*** (0.058)


0.180*** (0.011)

34,155

0.034

160,018

Yes

0.173*** (0.011)

34,720

0.030

241,005

Yes

0.335*** (0.024)

34,720

0.031

241,005

Yes

0.190*** (0.026)


Table 5 Robustness check in alternative models: fixed effect regression with age categories (1) or quadratic age specification (2)

0.192*** (0.048)

7,859

0.062

17,094

Yes

0.191*** (0.048)

8,778

0.057

25,184

Yes

0.264*** (0.061)

8,778

0.058

25,184

Yes

0.232*** (0.061)

Unemployed

0.213*** (0.011)

34,544

0.035

162,362

Yes

0.206*** (0.011)

35,142

0.031

246,090

Yes

0.270*** (0.025)

35,142

0.032

246,090

Yes

0.130*** (0.026)

Not unemployed


179,456

0.037

35,818

R-squared

Number of ID

0.041

35,818

R-squared

Number of ID

0.209*** (0.011)

Yes

179,456

0.039

35,818

RTI

Age AgeSquared

Observations

R-squared

Number of ID

18,337

0.033

92,810

Yes

0.219*** (0.015)

0.183*** (0.035)

18,337

0.034

92,810

Yes

0.214*** (0.015)

0.055 (0.037)

18,337

0.032

92,81

Yes

Women

17,481

0.052

86,646

Yes

0.202*** (0.015)

0.634*** (0.038)

17,481

0.056

86,646

Yes

0.196*** (0.015)

0.485*** (0.041)

17,481

0.046

86,646

Yes

Men

7,546

0.035

37,646

Yes

0.216*** (0.025)

0.544*** (0.056)

7,546

0.039

37,646

Yes

0.202*** (0.024)

0.331*** (0.058)

7,546

0.030

37,646

Yes

East

28,707

0.041

141,810

Yes

0.205*** (0.012)

0.314*** (0.029)

28,707

0.042

141,810

Yes

0.201*** (0.012)

0.194*** (0.031)

28,707

0.039

141,810

Yes

West

5,651

0.043

19,438

Yes

0.338*** (0.039)

0.424*** (0.065)

5,651

0.046

19,438

Yes

0.329*** (0.038)

0.371*** (0.067)

5,651

0.039

19,438

Yes


34,155

0.033

160,018

Yes

0.177*** (0.011)

0.349*** (0.028)

34,155

0.034

160,018

Yes

0.173*** (0.011)

0.200*** (0.030)

34,155

0.031

160,018

Yes


7,859

0.061

17,094

Yes

0.191*** (0.048)

0.292*** (0.085)

7,859

0.063

17,094

Yes

0.190*** (0.048)

0.264*** (0.085)

7,859

0.059

17,094

Yes

Unemployed

34,544

0.034

162,362

Yes

0.210*** (0.011)

0.272*** (0.028)

34,544

0.035

162,362

Yes

0.206*** (0.011)

0.132*** (0.030)

34,544

0.033

162,362

Yes

Not unemployed

Robust standard errors in parentheses, stars for significance levels: *** p \ 0.01, ** p \ 0.05, * p \ 0.1. Omitted dummies: age category [79, working zero hours, being single

Same controls as in the benchmark model, no time dummies

0.355*** (0.026)

Retired

Fixed effect 2

Yes

179,456

Observations

0.204*** (0.011)

RTI

Age Categories

0.218*** (0.027)

Retired

Fixed effect 1

Yes

Observations

All sample

Age AgeSquared

Table 5 continued


123

476

L. Becchetti et al.

Table 6 The effect of Relational Goods on Life Satisfaction: GSOEP, 1984–2007 (ordered probit regression with Mundlak correction) Variables

Base

Retired

Base_retired

Base_RTI

0.010 (0.012)

RTI

Base_RTI_retired 0.022 (0.015)

0.268*** (0.008)

0.268*** (0.008)


0.116*** (0.012)

0.116*** (0.012)

0.115*** (0.015)

0.115*** (0.015)

House owner

0.162*** (0.007)

0.162*** (0.007)

0.125*** (0.009)

0.126*** (0.009)


0.004 (0.008)

0.004 (0.008)

-0.001 (0.010)

-0.001 (0.010)

Unemployed

-0.280*** (0.015)

-0.277*** (0.016)

-0.257*** (0.019)

-0.251*** (0.019)

Job loss

-0.006 (0.028)

-0.006 (0.028)

-0.001 (0.033)

0.000 (0.033)

Full-time job

0.047*** (0.010)

0.052*** (0.011)

0.068*** (0.013)

0.077*** (0.014)

Part-time job

0.093*** (0.015)

0.096*** (0.016)

0.097*** (0.019)

0.105*** (0.019)

Vocational training

-0.537** (0.216)

-0.534** (0.216)

-0.503** (0.239)

-0.497** (0.239)

Marg. part-time job

0.079*** (0.022)

0.080*** (0.022)

0.068** (0.028)

0.072*** (0.028)

Military service

-0.304 (0.508)

-0.301 (0.508)

-0.369 (0.583)

-0.365 (0.583)

Married

0.133*** (0.020)

0.134*** (0.020)

0.139*** (0.025)

0.141*** (0.025)

Get married

0.078 (0.051)

0.078 (0.051)

0.105* (0.062)

0.104* (0.062)

Separated

-0.207*** (0.040)

-0.206*** (0.040)

-0.229*** (0.050)

-0.228*** (0.050)

Get separated

-0.107 (0.066)

-0.107 (0.066)

0.025 (0.084)

0.025 (0.084)

Divorced

-0.057** (0.023)

-0.057** (0.023)

-0.050* (0.029)

-0.049* (0.029)

Get divorced

-0.024 (0.061)

-0.023 (0.061)

-0.008 (0.077)

-0.008 (0.077)

Widowed

-0.011 (0.023)

-0.012 (0.023)

-0.009 (0.029)

-0.011 (0.029)

Child in HH

0.074*** (0.011)

0.073*** (0.011)

0.080*** (0.013)

0.079*** (0.013)

Hospital stay

-0.277*** (0.010)

-0.277*** (0.010)

-0.267*** (0.013)

-0.267*** (0.013)


-0.371*** (0.009)

-0.372*** (0.009)

-0.344*** (0.011)

-0.348*** (0.012)

West Germany

0.360*** (0.009)

0.361*** (0.009)

0.284*** (0.012)

0.287*** (0.012)

Age 50_52

-0.284*** (0.016)

-0.278*** (0.017)

-0.280*** (0.020)

-0.268*** (0.022)

123


477


Base

Base_retired

Base_RTI

Base_RTI_retired

Age 53_55

-0.297*** (0.016)

-0.292*** (0.017)

-0.303*** (0.020)

-0.292*** (0.021)

Age 56_58

-0.220*** (0.016)

-0.215*** (0.017)

-0.230*** (0.019)

-0.219*** (0.021)

Age 59_61

-0.125*** (0.015)

-0.121*** (0.016)

-0.126*** (0.019)

-0.118*** (0.020)

Age 62_64

-0.058*** (0.015)

-0.056*** (0.015)

-0.062*** (0.018)

-0.058*** (0.018)

Age 65_67

-0.006 (0.015)

-0.006 (0.015)

-0.009 (0.018)

-0.008 (0.018)

Time dummies

Yes

Yes

Yes

Yes

Dummy for 1992

0.273*** (0.025)

0.273*** (0.025)

0.288*** (0.025)

0.288*** (0.025)

Mundlak correction terms

Yes

Yes

Yes

Yes

Observations

85,693

85,693

56,014

56,014

Log likelihood

-159670.83

-159670.47

-104164.89

-104163.77

Z-statistics are in parenthesis, stars for significance levels: ** \5%, *** \1%. Mundlak correction terms are the averages over time of the socio demographic variables. Omitted dummies: age category [67, working zero hours, being single. Turn around: individuals aged from 50 to 70

In Table 5 we show the results with fixed effect estimation. We did not include time dummies because of the perfect multicollinearity that relates them to age in its linear form. The RTI variable continues to show a strongly significant effect on life satisfaction. For all other regressors, here omitted for reasons of space, we confirm the standard results in the literature. However, both the model with age in linear form and with age categories suffer from a possible omitted variable bias due to the exclusion of time dummies. Thus, we ended up choosing a benchmark model estimated with fixed effect regression where we include time dummies and age categories. An important limit of this model can be the approximation of the categorical life satisfaction variable by a continuous one. Although this is common in the happiness literature, we verify whether our results on relational life are confirmed when we take into account the discrete ordered categorical nature of our dependent variable by means of an Ordered Probit model. In order to make use of the panel structure of the dataset and to correct for individual fixed effects, we follow the Mundlak (1978) approach. We incorporate as correction factors the mean across time of the socio demographic regressors entered in the base equation. These averages are used to capture the correlation between time varying explanatory variables and individual effects due to persistent personality traits. Moreover, we make use of time dummies so to account for fixed time effects. Results are presented in Table 6. This estimation approach confirms that the impact of RTI on SWB is strongly significant and positive. An even deeper concern is whether our results remain robust when we depart from the normality assumption of the variables. So far we do correct for heteroskedasticity as we adopt robust standard errors, but the homoskedasticity assumption is only one of the necessary conditions for using the linear model. We therefore perform a robustness check by controlling whether our findings are still confirmed when we use the bootstrap method, that is, the well-known resampling method used to approximate standard errors, confidence

123

478

L. Becchetti et al.

Table 7 The effect of Relational Goods on Life Satisfaction: GSOEP, 1984–2007 Variables

q25

q50

q75

RTI

0.564*** (0.016) 0.091*** (0.035)

0.343*** (0.010) 0.004 (0.032)

0.132*** (0.049) -0.038 (0.027)


0.560*** (0.015)

0.389*** (0.012)

0.171*** (0.063)

House owner

0.197*** (0.018)

0.156*** (0.012)

0.048*** (0.018)


0.022*** (0.003)

0.012*** (0.002)

0.005** (0.002)

Unemployed

-0.662*** (0.031)

-0.659*** (0.028)

-0.462** (0.194)

Job loss

-0.060 (0.049)

0.119*** (0.045)

0.285 (0.260)

Full-time job

0.264*** (0.021)

0.177*** (0.017)

0.051** (0.020)

Part-time job

0.245*** (0.029)

0.195*** (0.019)

0.072*** (0.028)

Vocational training

0.212*** (0.046)

0.146*** (0.034)

0.049 (0.030)

Marg. part-time job

0.061 (0.039)

0.058** (0.027)

0.009 (0.015)

Military service

-0.001 (0.080) 0.286*** (0.024)

-0.067 (0.060) 0.271*** (0.018)

-0.102** (0.049) 0.115*** (0.044)

Get married

0.248*** (0.045)

0.157*** (0.031)

0.252** (0.115)

Separated

-0.444*** (0.074)

-0.307*** (0.057)

-0.055 (0.038)

Get separated

-0.426*** (0.121)

-0.206** (0.102)

-0.035 (0.043)

Divorced

-0.136*** (0.036)

-0.117*** (0.033)

-0.004 (0.014)

Get divorced

-0.121 (0.110)

0.110 (0.093)

0.030 (0.034)

Widowed

-0.062 (0.041)

0.058** (0.028)

0.053** (0.023)

Child in HH

0.074*** (0.007)

0.043*** (0.006)

0.021** (0.008)

Hospital stay

-0.446*** (0.026)

-0.285*** (0.018)

-0.101** (0.040)


-0.838*** (0.026)

-0.686*** (0.020)

-0.212*** (0.080)

West Germany

0.462*** (0.017) 0.487*** (0.080)

0.489*** (0.013) 0.290*** (0.063)

0.178*** (0.068) 0.150* (0.085)

Retired

Married

Age 17_19

123


479


q25

q50

q75

Age 20_22

0.409*** (0.064)

0.173*** (0.058)

0.031 (0.035)

Age 23_25

0.337*** (0.066)

0.063 (0.058)

-0.093* (0.048)

Age 26_28

0.241*** (0.064)

-0.026 (0.050)

-0.150** (0.063)

Age 29_31

0.139** (0.064)

-0.144*** (0.052)

-0.194** (0.078)

Age 32_34

0.004 (0.058)

-0.236*** (0.049)

-0.233** (0.091)

Age 35_37

-0.091 (0.063)

-0.301*** (0.051)

-0.269** (0.106)

Age 38_40

-0.210*** (0.062)

-0.371*** (0.060)

-0.288** (0.112)

Age 41_43

-0.308*** (0.058)

-0.469*** (0.051)

-0.323*** (0.125)

Age 44_46

-0.329*** (0.059)

-0.458*** (0.048)

-0.318*** (0.123)

Age 47_49

-0.344*** (0.055) -0.318*** (0.060)

-0.484*** (0.050) -0.465*** (0.056)

-0.318*** (0.123) -0.317*** (0.122)

Age 53_55

-0.308*** (0.060)

-0.482*** (0.047)

-0.331*** (0.127)

Age 56_58

-0.157** (0.067)

-0.383*** (0.053)

-0.258** (0.101)

Age 59_61

0.063 (0.050)

-0.194*** (0.049)

-0.116** (0.051)

Age 62_64

0.133*** (0.051)

-0.067 (0.042)

-0.050* (0.027)

Age 65_67

0.229*** (0.048)

0.036 (0.039)

-0.026 (0.018)

Age 68_70

0.212*** (0.060)

0.035 (0.046)

-0.033 (0.022)

Age 71_73

0.173*** (0.065)

0.065 (0.051)

-0.002 (0.019)

Age 74_76

0.234*** (0.066)

0.099** (0.048)

0.031 (0.022)

Age 77_79

0.188*** (0.058)

0.074 (0.064)

-0.019 (0.025)

Time dummies

Yes

Yes

Yes

1992

0.210*** (0.038)

0.197*** (0.022)

0.101** (0.041)

Constant

1.754*** (0.098)

4.353*** (0.071)

6.918*** (0.403)

Observations

179,456

179,456

179,456

Age 50_52

Quintile regression, estimation on different quintiles for RTI Robust standard errors in parentheses, stars for significance levels: *** p \ 0.01, ** p \ 0.05, * p \ 0.1. Omitted dummies: age category [79, working zero hours, being single

123

480

L. Becchetti et al.

intervals, and p values for test statistics, based on the sample data, when the theoretical distribution of the test statistic is unknown. Obtained findings document that our results survive when using this approach.26 Moreover, to tackle the problem of mean-conditionality, we re-estimate our model with a quantile regression which considers different points of the distribution (first quintile, second quintile–median, third quintile) and find that the coefficient of the RTI index remains always strongly significant and positive. Results are shown in Table 7 and document as a general result that most variables (the RTI variable included) which are significant in explaining the conditional mean of life satisfaction have a stronger impact in the lower quintiles of the distribution. This implies that positive/ negative life events impact more for low than for high life satisfaction levels.

5 Conclusions Common sense tells us that relational life plays an important role in life satisfaction: however the former can be both cause and effect of the latter. Moreover there are many factors that influence the propensity to go out and the level of life satisfaction that we are unable to capture. So estimating the effect of these two variables is not trivial. So far, with the Meier and Stutzer (2008) exception, the empirical contributions trying to calculate the elasticity of SWB with respect to social life, have not managed to solve this endogeneity conundrum. In this paper we devise a new approach to tackle this issue. We consider that the retirement event allows individuals to re-master their own agenda and to invest more time in social and relational activities. However, retirement is possibly a direct determinant of SWB and partially endogenous itself. So, we observe the age pattern behind retirement decisions and use it to create valid instruments. Our findings document that relational goods have a significant effect on life satisfaction which is quite robust under different specifications. In standard economic models relational arguments do not appear in agents’ utility functions and are therefore not taken into account when evaluating policies. Our findings suggest that this omission is important and has to be corrected in formulating ‘quality of life’ indicators. Given the heterogeneity of individual experiences reactions opinions about retirement and job satisfaction (and the impact of retirement on sociability and SWB demonstrated by our findings), fostering voluntary retirement schemes or even exchanges of rights to anticipate/postpone retirement, within a framework of compatibility with aggregate budget constraints, may be an example of new policies inspired by the SWB literature. More generally, the advice stemming from our paper is that measures aimed at stimulating social life and at preventing negative side-effects on it of policies are of crucial importance. In fact a wide range of policies can directly or indirectly encourage social life, from labour market regulations to urban planning and public funding of cultural, civic and sport activities. Other measures go from offering options for part time work to those who desire to pursue more community activities outside the workplace to making the workplace itself more productive of social connectivity (for instance fiscally encouraging the provision by employers of space and time for discussion groups, service clubs etc.). Another field for which our findings could prove inspirational is urban planning: a sprawling pattern of metropolitan settlement imposes heavy personal and economic costs: pollution, 26

Results are available by the authors upon requests.

123


481

congestion, and lost time. More pedestrian-friendly areas, with ample availability of public space could certainly offer more opportunities for socializing with friends and neighbours. Finally cultural, sport, religious activities should be encouraged not only for their intrinsic values but also as vehicles for social togetherness. Acknowledgments The ideas for this paper have been discussed in several seminars on the economics of life satisfaction with colleagues such as Stefano Bartolini, Luigino Bruni, Andrew Clark, Raphael Di Tella, Bruno Frey, Benedetto Gui. Luca Stanca Alois Stutzer, Robert Waldmann. We are grateful to them for comments and suggestions received.

Appendix See Tables 8, 9, 10, 11.

Table 8 Variable description Variable Life satisfaction

Mean Overall

7.006564

Between Within RTI

Retired

Overall

0.9801552

N = 353572 n = 44842

1.290032

-2.147282

14.74569

T-bar = 7.88484

0.5882822

0

3

N = 222390

3

n = 40977

3.611973

T-bar = 5.42719

Overall

0.1981415

Overall

4.860681

0.3985999

0

1

N = 374167

0.3554901

0

1

n = 47850

0.1991405

-.7601919

1.156475

T-bar = 7.81958

0.5656252

-3.050426

9.488941

N = 365652

Between

0.5495363

-2.016057

8.671215

n = 47278

Within

0.3314167

-2.45616

8.819505

T-bar = 7.73408

Overall

0.4634374

Overall

11.47696

0.4986621

0

1

N = 374167

0.4645321

0

1

n = 47850

0.2269824

-.4948959

1.421771

T-bar = 7.81958

2.582071

7

18

N = 346801

Between

2.567363

7

18

n = 43240

Within

0.7182003

2.042177

20.04839

T-bar = 8.02037

0.2871887

0

1

N = 350843

0.219799

0

1

n = 44625 T-bar = 7.86203

Overall

0.0907044

Overall

0.0331097

Between Within

Part-time job

10 10

0

Within

Full-time job

0 0

-.8073448

Between Job loss

1.839602 1.5025

0.3512454

Within

Unemployed

Observations

0.5148685

Between No. of year of education

Max

Within

Within

House owner

Min

Between

Between Log household equivalent income

SD

0.2165679

-.8467956

1.049038

0.1789232

0

1

N = 301181

0.1063345

0

1

n = 37679

0.1645487

-.633557

0.9896314

T-bar = 7.99334

0.4958469

0

1

N = 354369

Between

0.4325172

0

1

n = 44837

Within

0.2850782

-.5226491

1.394018

T-bar = 7.90349

0.2794131

0

1

N = 354369

Overall

Overall

0.4356843

0.0853574

Between

0.2195584

0

1

n = 44837

Within

0.1913713

-.872976

1.043691

T-bar = 7.90349

123

482

L. Becchetti et al.

Table 8 continued Variable Vocational training

Mean Overall

0.0312951

Between Within Marginal part-time job

Military service

Overall

0.0303017

Overall

0.0038604

Overall

0.6269813

0.062012

0

1

N = 354369

0.0396699

0

1

n = 44837

0.0563482

-.6628063

0.9621937

T-bar = 7.90349

0.4836077

0

1

N = 358722

0.4670519

0

1

n = 45137

Within

0.2126763

-.331352

1.585315

T-bar = 7.9474

Overall

0.0154463

Overall

0.0154158

Overall

0.0062778

Overall

0.0631213

0.1233196

0

1

N = 308230

0.0684895

0

1

n = 38416 T-bar = 8.02348

0.1154112

-.4845537

0.971968

0.1232

0

1

N = 358722

0.0993308

0

1

n = 45137 T-bar = 7.9474

0.0977143

-.9012508

0.9737492

0.0789835

0

1

N = 308230

0.0450167

0

1

n = 38416

0.0734811

-.4937222

0.9627995

T-bar = 8.02348

0.2431813

0

1

N = 358722

Between

0.2139428

0

1

n = 45137

Within

0.1220635

-.895212

1.021455

T-bar = 7.9474

Overall

0.0061707

Overall

0.062639

Within Overall

0.6395246

Between Within Overall

0.1182251

Between Within Overall

0.093469

Between Within

123

T-bar = 7.90349 N = 354369

Between

Between

West Germany

0.9896284 1

n = 44837

Within


-.8437049 0

T-bar = 7.90349

Between

Hospital stay

0.1374826 0.1714166

1

Within

Child in household

N = 354369 n = 44837

0.9886351

Between

Widowed

1 1

0

Within

Get Divorced

0 0

-.8696983

Between

Divorced

0.1741142 0.1556841

0.1398518

Within

Get separated

Observations

0.1250801

Between Separated

Max

Within

Within

Get married

Min

Between

Between Married

SD

Overall

0.7857374

0.0783113

0

1

N = 308230

0.0476624

0

1

n = 38416

0.0727878

-.4938293

0.9626925

T-bar = 8.02348

0.2423129

0

1

N = 358722

0.2346854

0

1

n = 45137 T-bar = 7.9474

0.0953242

-.8956943

1.020972

0.9684324

0

10

N = 374167

0.8707439

0

8.285714

n = 47850 T-bar = 7.81958

0.5258427

-6.471587

6.562602

0.322875

0

1

N = 328492

0.206967

0

1

n = 44517 T-bar = 7.37902

0.2833399

-.8151082

1.072771

0.2910889

0

1

N = 374167

0.2347721

0

1

n = 47850

0.1769248

-.8477075

1.051802

T-bar = 7.81958

0.4103104

0

1

N = 374167

Between

0.4036852

0

1

n = 47850

Within

0.0627448

-.1725959

1.730182

T-bar = 7.81958


483

Table 9 Relational time composite index (Rel_CI): linear fixed effect estimation for life satisfaction and IV fixed effect estimation with over identified models. GSOEP 1984–2007, whole sample Variables

Base

Base_retired

Rel_CI Retired

Base_rel_CI

Base_ retired_rel_CI

IV_2

IV_3

0.319*** (0.017)

0.317*** (0.017)

2.878** (1.244)

0.870 (0.685)

0.240*** (0.032)

0.174*** (0.048)

0.226*** (0.036)

0.237*** (0.024)


0.193*** (0.013)

0.192*** (0.013)

0.185*** (0.019)

0.184*** (0.019)

0.111*** (0.041)

0.169*** (0.027)

House owner

0.078*** (0.016)

0.079*** (0.016)

0.109*** (0.022)

0.110*** (0.022)

0.123*** (0.025)

0.113*** (0.021)


0.004 (0.006)

0.002 (0.006)

0.004 (0.007)

0.002 (0.007)

0.020* (0.012)

0.006 (0.008)

Unemployed

-0.199*** (0.020)

-0.178*** (0.020)

-0.201*** (0.028)

-0.177*** (0.028)

-0.169*** (0.031)

-0.176*** (0.028)

Job loss

-0.094*** (0.023)

-0.093*** (0.023)

-0.133*** (0.037)

-0.130*** (0.037)

-0.137*** (0.042)

-0.131*** (0.037)

Full-time job

0.314*** (0.016)

0.359*** (0.017)

0.326*** (0.023)

0.374*** (0.024)

0.392*** (0.027)

0.378*** (0.023)

Part-time job

0.151*** (0.018)

0.182*** (0.018)

0.149*** (0.027)

0.183*** (0.027)

0.129*** (0.040)

0.171*** (0.030)

Vocational training

0.238*** (0.028)

0.261*** (0.028)

0.283*** (0.045)

0.309*** (0.045)

0.314*** (0.050)

0.311*** (0.043)

Marg. part-time job

0.068*** (0.021)

0.083*** (0.021)

0.034 (0.034)

0.050 (0.034)

-0.031 (0.055)

0.032 (0.039)

Military service

-0.015 (0.053)

0.010 (0.053)

0.029 (0.090)

0.057 (0.090)

-0.072 (0.123)

0.029 (0.095)

Married

0.148*** (0.028)

0.154*** (0.028)

0.160*** (0.038)

0.166*** (0.038)

0.470*** (0.154)

0.232*** (0.089)

Get married

0.245*** (0.024)

0.242*** (0.024)

0.247*** (0.040)

0.244*** (0.040)

0.237*** (0.047)

0.243*** (0.040)

Separated

-0.095* (0.055)

-0.090* (0.055)

-0.127 (0.078)

-0.124 (0.078)

0.109 (0.145)

-0.074 (0.101)

Get separated

-0.319*** (0.058)

-0.320*** (0.058)

-0.201** (0.095)

-0.202** (0.095)

-0.124 (0.114)

-0.185* (0.097)

Divorced

0.094** (0.045)

0.097** (0.045)

0.115* (0.059)

0.119** (0.059)

0.388*** (0.145)

0.177* (0.091)

Get divorced

-0.074* (0.045)

-0.074* (0.045)

-0.086 (0.073)

-0.085 (0.073)

-0.141 (0.086)

-0.097 (0.075)

Widowed

-0.222*** (0.057)

-0.250*** (0.057)

-0.199*** (0.074)

-0.213*** (0.074)

-0.177** (0.080)

-0.205*** (0.069)

Child in HH

0.040*** (0.009)

0.041*** (0.009)

0.045*** (0.012)

0.047*** (0.012)

0.063*** (0.015)

0.050*** (0.012)

Hospital stay

-0.181*** (0.010)

-0.179*** (0.010)

-0.170*** (0.017)

-0.167*** (0.017)

-0.102*** (0.037)

-0.153*** (0.024)


-0.261*** (0.023)

-0.278*** (0.023)

-0.255*** (0.033)

-0.275*** (0.033)

-0.183*** (0.056)

-0.255*** (0.039)

West Germany

0.286*** (0.064)

0.286*** (0.064)

0.235*** (0.082)

0.236*** (0.082)

0.232*** (0.085)

0.235*** (0.076)

123

484

L. Becchetti et al.


Base

Base_retired

Base_rel_CI


IV_2

IV_3

Age 17_19

-0.220 (0.210)

-0.203 (0.210)

0.465 (0.331)

0.476 (0.331)

1.015** (0.452)

0.592* (0.357)

Age 20_22

-0.355* (0.202)

-0.339* (0.201)

0.364 (0.317)

0.371 (0.317)

0.874** (0.429)

0.480 (0.341)

Age 23_25

-0.335* (0.193)

-0.314 (0.193)

0.340 (0.303)

0.354 (0.303)

0.811** (0.403)

0.453 (0.323)

Age 26_28

-0.329* (0.184)

-0.300 (0.184)

0.346 (0.289)

0.367 (0.288)

0.789** (0.380)

0.458 (0.306)

Age 29_31

-0.283 (0.175)

-0.247 (0.175)

0.358 (0.274)

0.386 (0.274)

0.803** (0.365)

0.476 (0.292)

Age 32_34

-0.280* (0.167)

-0.237 (0.166)

0.325 (0.260)

0.359 (0.260)

0.704** (0.334)

0.433 (0.272)

Age 35_37

-0.260* (0.158)

-0.211 (0.157)

0.297 (0.245)

0.338 (0.245)

0.585* (0.299)

0.391 (0.250)

Age 38_40

-0.246* (0.149)

-0.190 (0.149)

0.317 (0.231)

0.363 (0.231)

0.514* (0.269)

0.396* (0.231)

Age 41_43

-0.218 (0.140)

-0.155 (0.140)

0.280 (0.217)

0.334 (0.216)

0.393 (0.246)

0.347 (0.214)

Age 44_46

-0.205 (0.132)

-0.135 (0.131)

0.253 (0.202)

0.313 (0.202)

0.313 (0.229)

0.313 (0.200)

Age 47_49

-0.193 (0.123)

-0.116 (0.123)

0.205 (0.188)

0.272 (0.188)

0.225 (0.215)

0.262 (0.186)

Age 50_52

-0.172 (0.115)

-0.090 (0.115)

0.201 (0.174)

0.274 (0.174)

0.198 (0.201)

0.258 (0.173)

Age 53_55

-0.144 (0.107)

-0.057 (0.107)

0.142 (0.160)

0.222 (0.160)

0.102 (0.191)

0.196 (0.161)

Age 56_58

-0.019 (0.098)

0.067 (0.098)

0.201 (0.146)

0.282* (0.146)

0.108 (0.188)

0.245 (0.152)

Age 59_61

0.162* (0.090)

0.225** (0.090)

0.377*** (0.132)

0.434*** (0.132)

0.172 (0.199)

0.377** (0.148)

Age 62_64

0.334*** (0.082)

0.344*** (0.082)

0.481*** (0.119)

0.486*** (0.119)

0.136 (0.219)

0.410*** (0.150)

Age 65_67

0.434*** (0.075)

0.407*** (0.075)

0.532*** (0.106)

0.501*** (0.106)

0.080 (0.238)

0.410*** (0.153)

Age 68_70

0.395*** (0.067)

0.363*** (0.067)

0.456*** (0.093)

0.422*** (0.094)

-0.018 (0.239)

0.327** (0.149)

Age71_73

0.352*** (0.060)

0.326*** (0.060)

0.390*** (0.082)

0.363*** (0.083)

-0.047 (0.220)

0.275** (0.136)

Age 74_76

0.282*** (0.052)

0.261*** (0.052)

0.246*** (0.071)

0.225*** (0.072)

-0.178 (0.211)

0.139 (0.128)

Age 77_79

0.162*** (0.041)

0.149*** (0.041)

0.202*** (0.060)

0.189*** (0.060)

-0.080 (0.146)

0.131 (0.092)

Time dummies

Yes

Yes

Yes

Yes

Yes

Yes

1992

0.788*** (0.054)

0.826*** (0.054)

0.633*** (0.081)

0.673*** (0.081)

0.483*** (0.128)

0.632*** (0.094)

Constant

5.222*** (0.141)

5.107*** (0.141)

4.646*** (0.208)

4.536*** (0.208)

Observations

271,274

271,274

108,238

108,238

100,464

100,464

123


485


Base

Base_retired

Base_rel_CI


IV_2

IV_3

Number of ID

36,250

36,250

31,219

31,219

23,445

23,445

R-squared

0.042

0.043

0.050

0.051

-0.298

0.035

F-first-excluded

8.043

14.11

Hansen test P value

0.025

0.002

Robust standard errors in parentheses, stars for significance levels: *** p \ 0.01, ** p \ 0.05, * p \ 0.1. Omitted dummies: age category [79, working zero hours, being single. IV estimates: RTI instrumented by sample and national retirement patterns. Instruments for IV_2 regression: QuotaAge (share of retired individuals of the same age cohort in the sample at time t) and QuotaAge_1 (share of retired individuals of the age cohort younger than 1 year in the sample at time t). Instruments for IV_3 regression: QuotaAge, QuotaAge_1 and SexRetAvgAge (national average retirement entrance age of the opposite sex, respecting the year and the regional location of the individual at time t)

Table 10 Life satisfaction and RTI2 (index computed aggregating the 5 relational time variables normalized as follows: 0 = never, 0.08 = once a year, 1 = once a month, 4 = once a week): linear fixed effect estimation for life satisfaction and IV fixed effect estimation with over identified models (GSOEP 1984–2007, whole sample) Variables

Base

Base_retired

RTI2 Retired

Base_RTI

Base_retired_RTI

IV_2

IV_3

0.134*** (0.008)

0.133*** (0.008)

1.490*** (0.370)

1.078*** (0.299)

0.232*** (0.028)

0.181*** (0.030)

0.197*** (0.028)

0.237*** (0.024)


0.193*** (0.013)

0.192*** (0.013)

0.208*** (0.015)

0.207*** (0.015)

0.190*** (0.016)

0.195*** (0.015)

House owner

0.078*** (0.016)

0.079*** (0.016)

0.080*** (0.018)

0.081*** (0.018)

0.097*** (0.018)

0.092*** (0.017)


0.004 (0.006)

0.002 (0.006)

0.005 (0.006)

0.003 (0.006)

0.019** (0.008)

0.014** (0.007)

Unemployed

-0.199*** (0.020)

-0.178*** (0.020)

-0.215*** (0.023)

-0.195*** (0.023)

-0.221*** (0.025)

-0.213*** (0.024)

Job loss

-0.094*** (0.023)

-0.093*** (0.023)

-0.110*** (0.029)

-0.107*** (0.029)

-0.083** (0.032)

-0.090*** (0.031)

Full-time job

0.314*** (0.016)

0.359*** (0.017)

0.327*** (0.019)

0.372*** (0.019)

0.421*** (0.023)

0.406*** (0.021)

Part-time job

0.151*** (0.018)

0.182*** (0.018)

0.159*** (0.022)

0.190*** (0.022)

0.183*** (0.023)

0.186*** (0.021)

Vocational training

0.238*** (0.028)

0.261*** (0.028)

0.253*** (0.033)

0.278*** (0.033)

0.311*** (0.037)

0.301*** (0.034)

Marg. part-time job

0.068*** (0.021)

0.083*** (0.021)

0.060** (0.027)

0.075*** (0.027)

0.016 (0.033)

0.034 (0.030)

Military service

-0.015 (0.053)

0.010 (0.053)

-0.019 (0.070)

0.007 (0.071)

-0.078 (0.080)

-0.052 (0.074)

Married

0.148*** (0.028)

0.154*** (0.028)

0.178*** (0.032)

0.184*** (0.032)

0.395*** (0.066)

0.331*** (0.056)

Get married

0.245*** (0.024)

0.242*** (0.024)

0.220*** (0.031)

0.217*** (0.031)

0.198*** (0.035)

0.204*** (0.033)

123

486

L. Becchetti et al.


Base

Base_retired

Base_RTI

Base_retired_RTI

IV_2

IV_3

Separated

-0.095* (0.055)

-0.090* (0.055)

-0.049 (0.063)

-0.046 (0.063)

0.138* (0.083)

0.082 (0.075)

Get separated

-0.319*** (0.058)

-0.320*** (0.058)

-0.303*** (0.076)

-0.304*** (0.076)

-0.303*** (0.081)

-0.303*** (0.078)

Divorced

0.094** (0.045)

0.097** (0.045)

0.113** (0.050)

0.116** (0.050)

0.335*** (0.077)

0.269*** (0.067)

Get divorced

-0.074* (0.045)

-0.074* (0.045)

-0.073 (0.057)

-0.073 (0.057)

-0.130** (0.064)

-0.113* (0.061)

Widowed

-0.222*** (0.057)

-0.250*** (0.057)

-0.193*** (0.062)

-0.222*** (0.062)

-0.221*** (0.057)

-0.221*** (0.054)

Child in HH

0.040*** (0.009)

0.041*** (0.009)

0.037*** (0.010)

0.038*** (0.010)

0.075*** (0.014)

0.064*** (0.012)

Hospital stay

-0.181*** (0.010)

-0.179*** (0.010)

-0.180*** (0.013)

-0.178*** (0.013)

-0.146*** (0.016)

-0.156*** (0.015)


-0.261*** (0.023)

-0.278*** (0.023)

-0.241*** (0.026)

-0.259*** (0.026)

-0.249*** (0.025)

-0.252*** (0.024)

West Germany

0.286*** (0.064)

0.286*** (0.064)

0.248*** (0.072)

0.248*** (0.072)

0.189*** (0.070)

0.208*** (0.066)

Age 17_19

-0.220 (0.210)

-0.203 (0.210)

-0.023 (0.253)

-0.007 (0.253)

0.092 (0.275)

0.062 (0.262)

Age 20_22

-0.355* (0.202)

-0.339* (0.201)

-0.142 (0.243)

-0.127 (0.243)

0.039 (0.266)

-0.011 (0.253)

Age 23_25

-0.335* (0.193)

-0.314 (0.193)

-0.155 (0.232)

-0.135 (0.232)

0.036 (0.254)

-0.016 (0.241)

Age 26_28

-0.329* (0.184)

-0.300 (0.184)

-0.131 (0.222)

-0.103 (0.221)

0.088 (0.243)

0.030 (0.230)

Age 29_31

-0.283 (0.175)

-0.247 (0.175)

-0.089 (0.211)

-0.054 (0.210)

0.161 (0.232)

0.096 (0.220)

Age 32_34

-0.280* (0.167)

-0.237 (0.166)

-0.086 (0.200)

-0.045 (0.200)

0.156 (0.220)

0.095 (0.208)

Age 35_37

-0.260* (0.158)

-0.211 (0.157)

-0.096 (0.189)

-0.048 (0.189)

0.125 (0.207)

0.073 (0.196)

Age 38_40

-0.246* (0.149)

-0.190 (0.149)

-0.073 (0.178)

-0.020 (0.178)

0.116 (0.193)

0.075 (0.184)

Age 41_43

-0.218 (0.140)

-0.155 (0.140)

-0.075 (0.168)

-0.015 (0.167)

0.104 (0.181)

0.068 (0.172)

Age 44_46

-0.205 (0.132)

-0.135 (0.131)

-0.071 (0.157)

-0.004 (0.157)

0.118 (0.169)

0.081 (0.161)

Age 47_49

-0.193 (0.123)

-0.116 (0.123)

-0.080 (0.146)

-0.006 (0.146)

0.117 (0.157)

0.080 (0.149)

Age 50_52

-0.172 (0.115)

-0.090 (0.115)

-0.061 (0.136)

0.017 (0.136)

0.116 (0.145)

0.086 (0.138)

Age 53_55

-0.144 (0.107)

-0.057 (0.107)

-0.064 (0.126)

0.019 (0.126)

0.089 (0.132)

0.068 (0.126)

Age 56_58

-0.019 (0.098)

0.067 (0.098)

0.029 (0.115)

0.111 (0.116)

0.156 (0.121)

0.142 (0.115)

Age 59_61

0.162* (0.090)

0.225** (0.090)

0.230** (0.105)

0.289*** (0.105)

0.287*** (0.109)

0.287*** (0.104)

123


487


Base

Base_retired

Base_RTI

Base_retired_RTI

IV_2

IV_3

Age 62_64

0.334*** (0.082)

0.344*** (0.082)

0.381*** (0.096)

0.390*** (0.095)

0.324*** (0.100)

0.344*** (0.095)

Age 65_67

0.434*** (0.075)

0.407*** (0.075)

0.475*** (0.086)

0.447*** (0.086)

0.323*** (0.094)

0.360*** (0.088)

Age 68_70

0.395*** (0.067)

0.363*** (0.067)

0.420*** (0.077)

0.388*** (0.077)

0.231*** (0.089)

0.279*** (0.082)

Age 71_73

0.352*** (0.060)

0.326*** (0.060)

0.369*** (0.068)

0.344*** (0.068)

0.170** (0.082)

0.223*** (0.075)

Age 74_76

0.282*** (0.052)

0.261*** (0.052)

0.315*** (0.060)

0.296*** (0.060)

0.148** (0.071)

0.193*** (0.065)

Age 77_79

0.162*** (0.041)

0.149*** (0.041)

0.215*** (0.050)

0.202*** (0.050)

0.125** (0.054)

0.148*** (0.051)

Time dummies

Yes

Yes

Yes

Yes

Yes

Yes

1992

0.788*** (0.054)

0.826*** (0.054)

0.744*** (0.063)

0.780*** (0.063)

0.682*** (0.072)

0.712*** (0.068)

Constant

5.222*** (0.141)

5.107*** (0.141)

4.905*** (0.163)

4.796*** (0.164)

Observations

271,274

271,274

179,456

179,456

172,537

172,536

Number of ID

36,250

36,250

35,818

35,818

28,899

28,899

R-squared

0.042

0.043

0.044

0.044

-0.198

-0.073

F-first-excluded

33.98

31.69

Hansen test P value

0.392

0.042

Robust standard errors in parentheses, stars for significance levels: *** p \ 0.01, ** p \ 0.05, * p \ 0.1. Omitted dummies: age category [79, working zero hours, being single. IV estimates: RTI instrumented by sample and national retirement patterns. Instruments for IV_2 regression: QuotaAge (share of retired individuals of the same age cohort in the sample at time t) and QuotaAge_1 (share of retired individuals of the age cohort younger than 1 year in the sample at time t). Instruments for IV_3 regression: QuotaAge, QuotaAge_1 and SexRetAvgAge (national average retirement entrance age of the opposite sex, respecting the year and the regional location of the individual at time t)

Table 11 Proportion of retirees (Quota) and mean retired people by gender over the years, by West and East Germany Year Sample

Quota West

Retired men West

Retired women West

Quota East

Retired men East

Retired women East

1984

0.16

0.15

0.17

n.a.

n.a.

n.a.

1985

0.16

0.14

0.17

n.a.

n.a.

n.a.

1986

0.15

0.14

0.17

n.a.

n.a.

n.a.

1987

0.16

0.14

0.17

n.a.

n.a.

n.a.

1988

0.16

0.15

0.17

n.a.

n.a.

n.a.

1989

0.16

0.14

0.17

n.a.

n.a.

n.a.

1990

0.16

0.14

0.18

n.a.

n.a.

n.a.

1991

0.17

0.15

0.19

n.a.

n.a.

n.a.

1992

0.17

0.15

0.19

0.18

0.15

0.21

1993

0.17

0.16

0.19

0.20

0.18

0.23

123

488

L. Becchetti et al.

Table 11 continued Year Sample

Quota West

Retired men West

Retired women West

Quota East

Retired men East

Retired women East

1994

0.18

0.16

0.19

0.21

0.18

0.23

1995

0.18

0.17

0.19

0.20

0.18

0.23

1996

0.18

0.17

0.20

0.21

0.18

0.23

1997

0.19

0.18

0.20

0.20

0.17

0.23

1998

0.19

0.19

0.20

0.21

0.18

0.23

1999

0.17

0.18

0.17

0.20

0.18

0.22

2000

0.18

0.19

0.17

0.22

0.20

0.24

2001

0.18

0.19

0.17

0.22

0.20

0.24

2002

0.18

0.20

0.17

0.23

0.21

0.25

2003

0.19

0.20

0.18

0.24

0.22

0.25

2004

0.19

0.20

0.18

0.23

0.22

0.25

2005

0.19

0.20

0.18

0.24

0.23

0.26

2006

0.20

0.21

0.19

0.25

0.24

0.27

2007

0.20

0.20

0.19

0.25

0.24

0.26

Total

0.18

0.18

0.18

0.20

0.18

0.22

The East Germany sample was added to the SOEP in 1990, but the retirement variable is available since 1992. The proportion of retirees (Quota) is exogenous to the individual information by construction, but approximately is equal to the average Retired variable into the population. The reference population excludes people with more than 76 years old

References Alesina, A. F., Glaeser, E. L., & Sacerdote, B. (2005). Work and Leisure in the U.S. and Europe: Why so different? CEPR Discussion Paper No. 5140. Angrist, J. D., & Pischke, J. S. (2009). Mostly harmless econometrics: An empiricist’s companion. Princeton: University Press. Antoci, A., Sacco, P. L., & Vanin, P. (2007). Social capital accumulation and the evolution of social participation. Journal of Socio-Economics, 36(1), 128–143. Aslam, A., & Corrado, L. (2007). No Man is an Island, the inter-personal determinants of regional WellBeing in Europe. Cambridge Working Papers in Economics No. 0717. Azariadis, C., Chen, B. L., Lu, C. H., & Wang, Y. C. (2008) A two-sector model of endogenous growth with leisure extenalities. Mimeo, Washington University in St. Louis. Bardsley, N., & Sugden, R. (2006). Human nature and sociality in economics. Handbookon the Economics of Giving, Reciprocity and Altruism, 1, 731–768. Bartolini, S., Bilancini, E., & Pugno, M. (2009). American decline in social capital and happiness. Is there a Linkage? Mimeo University of Siena. Becchetti, L., Pelloni, A., & Rossetti, F. (2008). Happiness and sociability. Kyklos, 61(3), 343–363. Berkel, B., & Bo¨rsch-Supan, A. (2003). Pension reform in Germany: The impact on retirement decision. MEA discussion paper series 03036, Mannheim Research Institute for the Economics of Aging (MEA), University of Mannheim. Blanchard, O. (2006). European unemployment: The evolution of facts and ideas. Economic Policy, 21(45), 5–59. Blanchflower, D. G., & Oswald, A. J. (2004). Well-being over time in Britain and the USA. Journal of Public Economics, 88(7–8), 1359–1386. Bo¨rsch-Supan, A., & Wilke, C. B. (2004). The German public pension system: How it was, How it will be. NBER Working Paper No. 10525. Bo¨rsch-Supan, A., & Schnabel, R. (1999). Social security and retirement in Germany. In J. Gruber, & D. A. Wise (Eds.). Social Security and Retirement around the World, Series: National Bureau of Economic Research Conference Report.

123


489

Bruni, L. (2009). Happiness. In J. Peil, & I. van Staveren (Eds.), Handbook of economics and ethics. Edward Elgar, Cheltenham, UK. Bruni, L., & Stanca, L. (2008). Watching alone: Relational goods, television and happiness. Journal of Economic Behavior & Organization, 65(3–4), 506–528. Buber, M. (1973). Meetings (M. Friedman, Trans.). Open Court Publishing, La Salle, Illinois. Burda, M., Hamermesh, D., & Weil, P. (2008). The distribution of total work in the Us and EU. In Are Europeans Lazy or Americans Crazy? Oxford University Press, forthcoming. Clark, A. J. (2007). Born to be mild? Cohort Effects Don’t (Fully) Explain Why Well-Being Is U-Shaped in Age. IZA Discussion Paper No. 3170. Clark, A. J., Frijters, P., & Shields, M. (2008). Relative income, happiness and utility: An explanation for the Easterlin Paradox and other puzzles. Journal of Economic Literature, 46(1), 95–144. Clark, A. E., & Oswald, A. J. (1994). Unhappiness and unemployment. Economic Journal, 104(424), 648–659. Corneo, G. (2005). Work and television. European Journal of Political Economy, 21(1), 99–113. Davidson, R., & MacKinnon, J. G. (1993). Estimation and inference in econometrics. New York: Oxford University Press. Di Tella, R., Mac Culloch, R., & Oswald, A. (2001). Preferences over inflation and unemployment: Evidence from surveys of happiness. American Economic Review, 91(1), 335–341. Di Tella, R., Mac Culloch, R., & Oswald, A. (2003). The macroeconomics of happiness. Review of Economics and Statistics, 85(4), 809–827. Di Tella, R., & MacCulloch, R. (2005). Partisan social happiness. Review of Economic Studies, 72(2), 367–393. 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), 52–61. Diener, E., & Seligman, M. E. P. (2004). Beyond money. Psychological Science in the Public Interest, 5(1), 1–31. Engfer, U. (2009). Restructuring of activities and rising satisfaction with time use after retirement: Findings from the German time budget survey. In Proceedings from the 2009 ISQOLS Conference, Florence 2009. Frey, B. S. (2008). Happiness: A revolution in economics. Cambridge: MIT Press. Frey, B. S., Benesch, C., & Stutzer, A. (2007a). Does watching TV make us happy? Journal of Economic Psychology, 28(3), 283–313. Frey, B. S., Luechinger, S., & Stutzer, A. (2004). Valuing public goods: The life satisfaction approach. CESifo Working Paper Series No. 1158. Frey, B. S., Luechinger, S., & Stutzer, A. (2007b). Calculation tragedy: Assessing the costs of terrorism. Journal of Economic Surveys, 21(1), 1–24. Frey, B. S., & Stutzer, A. (2006a). Does marriage make people happy, or do happy people get married? Journal of Socio Economics, 35(2), 326–347. Frey, B. S., & Stutzer, A. (2006b). What happiness research can tell us about self-control problems and utility misprediction. IEW–Working Papers iewwp267, Institute for Empirical Research in Economics–IEW. Frijters, P., & Beatton, T. (2008).The mystery of the U-shaped relationship between happiness and age. National Centre for Econometric Research Working Paper Series No.26. Gallie, D., & Russell, H. (1998). Unemployment and life satisfaction: A cross-cultural comparison. Archives Europe´ennes de Sociologie XXXIX, 248–280. Gui, B. (1987) E´le´ments pour une definition d’e´conomie communautaire. Notes et Documents de l’Institut International Jacques Maritain 19/20. Gui, B. (1996). On ‘‘relational goods’’: Strategic implications of investment in relationships. International Journal of Social Economic, 23, 260–278. Gui, B. (2005). From transactions to encounters. The joint generation of relational goods and conventional values. In B. Gui & R. Sugden (Eds.), Economics and social interaction: Accounting for Interpersonal Relations. Cambridge: Cambridge University Press. Haisken-DeNew, J. P, & Hahn, M. (2006). PanelWhiz: A Menu-Driven Stata/SE interface for accessing panel data. Mimeo, www.panelwhiz.eu. Helliwell, J. F., & Putnam, R. D. (2004). The social context of well-being. Philosophical Transactions, 359(1449), 1435–1446. Jenkins, S. P. & Osberg, L. (2003). Nobody to Play with? The implications of leisure coordination. Discussion Papers of DIW Berlin No. 368.

123

490

L. Becchetti et al.

Johnson, D. R., & Wu, J. (2002). An empirical test of crisis, social selection, and role explanations of the relationship between marital disruption and psychological distress: A pooled time-series analysis of four-wave panel data. Journal of Marriage and the Family, 64, 211–224. Krueger, A. (2008). Comment on Stevenson and Wolfers, economic growth and subjective well-being: Reassessing the easterlin paradox. Brookings Papers on Economic Activity (pp. 95–100). Spring. Meier, S., & Stutzer, A. (2008). Is volunteering rewarding in itself? Economica, 75(297), 39–59. Merz, J., & Osberg, L. (2006). Keeping in touch: A benefit of public holidays. Working Papers No.37, ECINEQ, Society for the Study of Economic Inequality. Moore, S., & Shepherd, J. P. (2006). The cost of fear: Shadow pricing the intangible costs of crime. Applied Economics, 38, 293–300. Mundlak, Y. (1978). On the pooling of time series and cross section data. Econometrica, 46(1), 69–85. Nicoletti, G., Scarpetta, S., & Boylaud, O. (2000). Summary indicators of product market regulation with an extension to employment protection legislation, OECD Economics Department Working Papers No. 226. Powdthavee, N. (2008). Putting a price tag on friends, relatives, and neighbours: Using surveys of life satisfaction to value social relationships. Journal of Socio Economics, 37(4), 1459–1480. Prescott, E. C. (2004). Why do Americans work so much more than Europeans? Quarterly Review, Federal Reserve Bank of Minneapolis, Issue Jul, 2–13. Pugno, M. (2010). Scitovsky’s economics of happiness: A suggested model. Mimeo: University of Cassino. Sackmann, R. (2007). Social protection in Germany: Is there a growing generational inequality? Horizons Strate´giques, No 5. Stock, J. H., Wright, J. H., & Yogo, M. (2002). A survey of weak instruments and weak identification in generalized method of moments. Journal of Business and Economic Statistics, 20(4), 518–528. Uhlaner, C. J. (1989). Relational goods and participation: Incorporating sociability into a theory of rational action. Public Choice Journal, 62(3), 253–285. Van Landeghem, B. G. M. (2008). Human well-being over the life cycle: Longitudinal evidence from a 20-Year Panel. LICOS Centre for Institutions and Economic Performance, Discussion Paper 213/2008. Van Praag, B. M. S., & Baarsma, B. E. (2005). Using happiness surveys to value intangibles: The case of airport noise. Economic Journal, 115(500), 224–246. Van Praag B. M. S., & Ferrer-i-Carbonell, A. (2008). Happiness quantified. A satisfaction calculus approach. Oxford: Oxford University Press. Welsch, H. (2002). Preferences over prosperity and pollution: Environmental valuation based on happiness surveys. Kyklos, 55(4), 473–494. Welsch, H. (2006). Environment and happiness: Valuation of air pollution using life satisfaction data. Ecological Economics, 58(4), 801–813.

123

The Relationship Between Social Leisure and Life ... - Springer Link

The Relationship Between Social Leisure and Life ... - Springer Link

Suggest Documents

The Relationship Between Depression and Leisure

THE RELATIONSHIP BETWEEN SKEWNESS AND ... - Springer Link

Relationship between plant life forms and ecological ... - Springer Link

The relationship between dysglycaemia and ... - Springer Link

The Relationship Between Social Support, Loneliness ... - Springer Link

Gender modifies the relationship between social ... - Springer Link

Multidimensionality of the relationship between social ... - Springer Link

The relationship between serious leisure ...

Relationship between deposition and recrystallization ... - Springer Link

Relationship between migraine and malocclusion - Springer Link

Relationship between obesity and unrecognized ... - Springer Link

THE DIALECTIC RELATIONSHIP BETWEEN ... - Springer Link

The Relationship Between Attitudes Towards ... - Springer Link

Exploring the Relationship Between Treatment ... - Springer Link

The relationship between proposed maintaining ... - Springer Link

P028: Examining the relationship between ... - Springer Link

Relationship between Social Support, Life ... - Revista Javeriana

Relationship between microstructure homogeneity and ... - Springer Link

Relationship Between Sunitinib Pharmacokinetics and ... - Springer Link

Resilience Moderates the Relationship Between ... - Springer Link

The Relationship between Childhood Family ... - Springer Link

Relationship between composite structures and ... - Springer Link

Relationship between RANKL and neuroendocrine ... - Springer Link

Relationship Between Emotional Intelligence and ... - Springer Link