- Email: [email protected]
Late life transitions and social networks: The case of retirement
Economics Letters 125 (2014) 459–462
Contents lists available at ScienceDirect
Economics Letters journal homepage: www.elsevier.com/locate/ecolet
Late life transitions and social networks: The case of retirement✩ Jason M. Fletcher ∗ University of Wisconsin-Madison, La Follette School of Public Affairs, 1225 Observatory Drive, Madison, WI 53706, United States University of Wisconsin-Madison, Department of Sociology, 1180 Observatory Drive, Madison, WI 53706, United States University of Wisconsin-Madison, Center for Demography and Ecology, 1180 Observatory Drive, Madison, WI 53706, United States
highlights • • • •
Use new SHARE social networks data from 16 countries. Examine impacts of retirement on social network outcomes. Across-country pension eligibility age rules used as an instrument for retirement. Find no measurable impact of retirement on social networks.
article
info
Article history: Received 27 July 2014 Received in revised form 4 October 2014 Accepted 7 October 2014 Available online 16 October 2014
abstract I use SHARE social networks data from 16 countries to examine retirement impacts on social networks. I extend the literature by using across-country differences in pension eligibility ages as instruments to show limited impacts of retirement on social network outcomes. © 2014 Elsevier B.V. All rights reserved.
JEL classification: J26 J1 J14 Keywords: Social networks Retirement Instrumental variables Social isolation
1. Introduction and context Individuals’ social networks change considerably across the life course. While there is a large literature examining the composition
✩ I thank Sara Koliner for excellent research assistance. This paper uses data from SHARE wave 4 release 1.1.1, as of March 28th 2013. The SHARE data collection has been primarily funded by the European Commission through the 5th Framework Programme (project QLK6-CT-2001-00360 in the thematic programme Quality of Life), through the 6th Framework Programme (projects SHARE-I3, RII-CT-2006062193, COMPARE, CIT5- CT-2005-028857, and SHARELIFE, CIT4-CT-2006-028812) and through the 7th Framework Programme (SHARE-PREP, N◦ 211909, SHARELEAP, N◦ 227822 and SHARE M4, N◦ 261982). Additional funding from the US National Institute on Aging (U01 AG09740-13S2, P01 AG005842, P01 AG08291, P30 AG12815, R21 AG025169, Y1-AG-4553-01, IAG BSR06-11 and OGHA 04-064) and the German Ministry of Education and Research as well as from various national sources is gratefully acknowledged (see www.share-project.org for a full list of funding institutions). ∗ Correspondence to: University of Wisconsin-Madison, La Follette School of Public Affairs, 1225 Observatory Drive, Madison, WI 53706, United States. Tel.: +1 203 785 5760. E-mail address: [email protected].
http://dx.doi.org/10.1016/j.econlet.2014.10.004 0165-1765/© 2014 Elsevier B.V. All rights reserved.
and impacts of social network on child, adolescent and young adult outcomes, much less research has examined related questions for older age individuals. In part this difference is based on data availability; while school-based surveys such as Add Health are a natural way to collect social network information among youth, adults do not have a ‘‘natural’’ context in which to collect complete network data. Instead of attempting to collect complete social network information from adults, most surveys collect ego-networks, where each respondent describes the identities, characteristics, and frequency of contact with a set of network ties, which often include spouses, family, co-workers, and friends among others. From this literature, there is an emerging set of stylized facts about adult networks. First, contrary to some earlier evidence, some aspects of adult social networks are enhanced with age, including frequency of socializing with neighbors, volunteering, and religious participation (Cornwell et al., 2008), though there does seem to be evidence that age is negatively related to network size and closeness to members. In addition to the general age patterns related to social networks, the impacts of life events on social networks have been a
460
J.M. Fletcher / Economics Letters 125 (2014) 459–462 Table 1 Summary statistics. SHARE wave 4, ages 55–70 (N ∼ 31,000). Variable
Obs
Mean
SD
Min
Max
Social network satisfaction Size of social network Average closeness with social network No reported friendships in social network Number of friendships reported Retired Age Male Education category (country-specific) Married Missing education data
30 600 31 144 29 593 31 144 29 887 31 144 31 144 31 144 31 144 31 144 31 144
8.83 2.53 3.26 0.69 0.51 0.52 62.29 0.45 4.90 0.45 0.20
1.41 1.60 0.63
0 0 −0.5 0 0 0 55 0 1 0 0
10 7 4 1 7 1 70 1 24 1 1
0.90 4.48 3.85
Industry categories not reported.
major focus in research in this area. Some of the most convincing research has used longitudinal network data to leverage withinperson designs to compare networks before and after common life events (Kalmijn, 2012). A smaller literature has shown that there are also bi-directional relationships between social network ties and retirement (Lancee and Radl, 2012). The current paper extends methods from the economics of aging literature to estimate causal impacts of retirement by leverage across-country age-specific pension policy eligibility rules that have been shown to induce retirement (Coe and Zamarro, 2011; Rohwedder and Willis, 2010). I extend this method to the social network literature and present new evidence suggesting that retirement plays a limited role in shaping social network composition and characteristics in the short term. 2. Data This paper uses the Wave 4 collection of the Survey of Health, Ageing and Retirement in Europe (SHARE) data series to explore the difference in social networks of retired versus non-retired individuals. Social network characteristics have only be assessed in a new module beginning in Wave 4, so that only cross-sectional associations can be examined.1 The Social Networks module asks about relationships (i.e. family members, friends, neighbors, and acquaintances) during the previous 12 months and focuses on ‘‘discussion’’ networks: ‘‘who are the people with whom you most often discussed important things?’’ Each contact is then described based on relationship (wife/friend/etc.), frequency of contact, proximity, and closeness (‘‘how close do you feel to X?’’). The respondents are also asked about their overall network satisfaction. This type of network description is called an ‘‘ego network’’ because it only asks questions of a single individual and then spans out from that individual’s self-reported network. No network ‘‘alters’’ are included in the survey and there is no verification of whether these network links exist from the alter’s perspective.2 Summary statistics are presented in Table 1 for the sample of SHARE respondents who are between the ages of 55 and 70. This age restriction is made in order to focus on short term impacts of retirement for respondents in the age range that is relevant for initial retirement; indeed, 52% of individuals in these ages are retired. Overall, the individuals report being highly satisfied with their overall network (average 8.8 of 10) and feel very close to the network members (average 3.2 of 4). The average number of friendships in the network is 0.5 and nearly 70% of respondents do not report any friends within this ‘‘discussion network’’, but rather report spouses, children, and other family members with great frequency.
3. Empirical methods This paper begins with descriptive regressions that estimate associations between retirement and the characteristics of individuals’ social networks: networkic = β1 retiredic + β2 Xic + β3 Cc + µic
(1)
where individual i in country c has a social network characteristic network (described be closeness, size, satisfaction, lack of friend nomination) that is a function of whether the individual is currently retired, individual level characteristics X (including age, gender, education level,3 marital status, and industry in which the individual was employed or is currently employed), country level fixed effects C and an idiosyncratic error term. In order to then examine causal impacts of retirement, the empirical design of this paper extends work by Rohwedder and Willis (2010) (RW), who estimated the impacts of retirement on cognitive decline using the SHARE data. In order to estimate causal effects of the endogenous retirement choice in their models, the authors leverage across-country variation in pension eligibility ages that in theory should induce some individuals to retire ‘‘earlier’’ or ‘‘later’’ than the respondent’s preferred retirement age if there were no pension policies. Indeed, RW show that these pension policies increase the age-adjusted retirement likelihood by between 15 and 20 percentage points (see RW Table 1). Likewise, the current study uses an updated4 set of acrosscountry policies that take into account the additional countries that are now available in Wave 4 in the SHARE Data. This allows the following two-stage approach using instrumental variables. networkic = β1 retiredic + β2 Xic + β3 Cc + µic
(2)
retiredic = α1 (ageic > eligiblec ) + α2 Xic + α3 Cc + εic
(3)
where β1 is the coefficient of interest and, given a Local Average Treatment Effect (LATE) interpretation (Imbens and Angrist, 1994), will provide the effect of retirement on network characteristics for individuals induced to retired based on their country-specific pension eligibility policy. 4. Results Table 2 presents OLS results following Eq. (1) that estimate adjusted associations between retirement status and social network characteristics. As noted in the tables, industry indicators and country-level indicators are also controlled, following results from the literature that note large differences in the patterns
1 http://www.share-project.org/fileadmin/pdf_questionnaire_wave_4/_Share_ wave_4.8.8.pdf. 2 In contrast, studies like the Add Health survey entire schools in order to collect
3 http://www.share-project.org/_questionnaire/dev2.php?welleid=26. See for the definitions of the country-specific codes. 4 http://www.ssa.gov/policy/docs/progdesc/ssptw/2012-
complete (within-school) network data from all egos and alters.
2013/europe/ssptw12europe.pdf.
J.M. Fletcher / Economics Letters 125 (2014) 459–462
461
Table 2 OLS analysis of associations between retirement and social networks. Outcome Fixed effects
Network satisfaction Industry, Country
Retired
−0.006
Average closeness Industry, Country
No friend Industry, Country
0.064** (0.025)
−0.014
−0.017***
0.003 (0.002)
−0.007***
−0.002*
(0.003)
(0.001)
−0.158***
−0.484***
−0.024***
(0.023) Age Male
Network size Industry, Country
(0.009)
(0.006)
Number of friends Industry, Country 0.036*** (0.014)
0.003*** (0.001)
−0.004***
0.109*** (0.007)
−0.167*** 0.030*** (0.004)
(0.001)
(0.017)
(0.021)
(0.007)
Education
0.003 (0.003)
0.051*** (0.005)
0.004** (0.002)
−0.015***
Married
0.223*** (0.025)
0.081*** (0.026)
0.162*** (0.010)
0.150*** (0.007)
−0.286***
Missing education category
0.024 (0.023)
0.032*** (0.010)
−0.057***
(0.037)
0.044*** (0.012)
Constant
8.837*** (0.138)
3.169*** (0.170)
3.665*** (0.060)
0.437*** (0.045)
0.882*** (0.083)
Observations R-squared
30,482 0.032
31,017 0.070
29,481 0.143
31,017 0.097
29,773 0.090
−0.030
(0.002)
(0.013)
(0.017) (0.021)
Standard errors in parentheses, clustered at the country/birth year level. * p < 0.1. ** p < 0.05. *** p < 0.01.
Table 3 Two-stage least squares analysis of effects of retirement on social networks. Outcome Fixed effects
Network satisfaction Industry, Country
Network size Industry, Country
Average closeness Industry, Country
No friend Industry, Country
Number of friends Industry, Country
Retired
0.029 (0.113)
−0.111
0.022 (0.045)
−0.025
−0.020
(0.140) 31,017 0.071
29,481 0.141
Observations R-squared
30,482 0.032
(0.039)
(0.074)
31,017 0.101
29,773 0.093
Standard errors in parentheses, clustered at the country/birth year level. Same controls as Table 2 (not shown). ∗ p < 0.1. ∗∗ p < 0.05. ∗∗∗ p < 0.01.
of social networks and work force participation across European countries in the SHARE sample5 (Kohli et al., 2009). Following some previous work, the differences in network characteristics between retired and non-retired individuals are mixed. On one hand, the associations suggest that retired individuals have slightly larger networks and are less likely to report zero friends in their ‘‘discussion network’’; there is also some evidence of reduced levels of reported closeness with members in one’s social network. The more general finding is that, while there are some differences between retired and non-retired, they are small. Because retirement is itself a choice, it is likely premature to consider the results in Table 2 as representing causal impacts of retirement status on social networks. Indeed, the small ‘‘effects’’ in Table 2 could reflect that individuals who retire do so in circumstances and at a time in which their social networks would be minimally affected. Table 3 then separates causal and noncausal effects of retirement by leveraging across-country agedependent eligibility policies for pensions. The assumption is that these policies are not correlated with age profiles of social network characteristics within each country (note: country fixed effects are controlled in these analyses). Table 3 shows that the within-country instruments, which include indicators for whether each respondent is eligible for early retirement or eligible for regular retirement, (Rohwedder
5 Findings upon request show that the results are sensitive to omitting country level indicators.
and Willis, 2010) are strong, with an F -statistic above 50 (Staiger and Stock, 1997) (first stage results available upon request). As is typical with instrumental variables, the standard errors increase in comparison to the OLS estimates in Table 2, often by a factor of 4 or 5. This limits some ability to clearly accept or reject a null of zero effects or of small effects of retirement. In general, I find no statistical evidence of effects on social networks in these analyses, and the estimates can rule out moderate effect sizes. For example, the standard deviation for the average closeness measure is 0.63 and the estimate and standard error on the impact of retirement is 0.03 (0.045), suggesting that retirement has no more than modest impacts on measures of closeness. 5. Conclusions This paper presents new causal evidence of the impacts of retirement on social network characteristics. The literature on this topic suggests that important life events, such as retirement, can have large impacts on social networks, but also that social networks and retirement are bi-directionally related and endogenous. Using a sample of 55–70 year old respondents from 16 European countries combined with information on age-eligibility policies of pension receipt in each country, findings suggest very little evidence of moderate or large causal effects of retirement on a variety of social network characteristics. This new evidence suggests a limited role of retirement and social networks and that other aspects of aging may be responsible for the empirical regularities in most datasets showing that social networks generally shrink in old
462
J.M. Fletcher / Economics Letters 125 (2014) 459–462
age. The results also suggest that retirement, and specifically retirement induced by policies, may have a limited role in increasing social isolation and reduced social networks among those in old age. References Coe, Norma B., Zamarro, Gema, 2011. Retirement effects on health in Europe. J. Health Econ. 30 (1), 77–86. Cornwell, Benjamin, Laumann, Edward O., Schumm, L. Philip, 2008. The social connectedness of older adults: a national profile. Amer. Sociol. Rev. 73 (2), 185–203. http://dx.doi.org/10.1177/000312240807300201.
Imbens, Guido W., Angrist, Joshua D., 1994. Identification and estimation of local average treatment effects. Econometrica 62 (2), 467–475. Kalmijn, Matthijs, 2012. Longitudinal analyses of the effects of age, marriage, and parenthood on social contacts and support. Adv. Life Course Res. 17 (4), 177–190. Kohli, Martin, Hank, Karsten, Künemund, Harald, 2009. The social connectedness of older Europeans: patterns, dynamics and contexts. J. Eur. Soc. Policy 19 (4), 327–340. Lancee, Bram, Radl, Jonas, 2012. Social connectedness and the transition from work to retirement. J. Gerontol. Ser. B: Psychol. Sci. Soc. Sci. 67 (4), 481–490. Rohwedder, Susann, Willis, Robert J., 2010. Mental retirement. J. Econ. Perspect. 24 (1), 119–138. Staiger, Douglas, Stock, James H., 1997. Instrumental variables regression with weak instruments. Econometrica 557–586.
Contents lists available at ScienceDirect
Economics Letters journal homepage: www.elsevier.com/locate/ecolet
Late life transitions and social networks: The case of retirement✩ Jason M. Fletcher ∗ University of Wisconsin-Madison, La Follette School of Public Affairs, 1225 Observatory Drive, Madison, WI 53706, United States University of Wisconsin-Madison, Department of Sociology, 1180 Observatory Drive, Madison, WI 53706, United States University of Wisconsin-Madison, Center for Demography and Ecology, 1180 Observatory Drive, Madison, WI 53706, United States
highlights • • • •
Use new SHARE social networks data from 16 countries. Examine impacts of retirement on social network outcomes. Across-country pension eligibility age rules used as an instrument for retirement. Find no measurable impact of retirement on social networks.
article
info
Article history: Received 27 July 2014 Received in revised form 4 October 2014 Accepted 7 October 2014 Available online 16 October 2014
abstract I use SHARE social networks data from 16 countries to examine retirement impacts on social networks. I extend the literature by using across-country differences in pension eligibility ages as instruments to show limited impacts of retirement on social network outcomes. © 2014 Elsevier B.V. All rights reserved.
JEL classification: J26 J1 J14 Keywords: Social networks Retirement Instrumental variables Social isolation
1. Introduction and context Individuals’ social networks change considerably across the life course. While there is a large literature examining the composition
✩ I thank Sara Koliner for excellent research assistance. This paper uses data from SHARE wave 4 release 1.1.1, as of March 28th 2013. The SHARE data collection has been primarily funded by the European Commission through the 5th Framework Programme (project QLK6-CT-2001-00360 in the thematic programme Quality of Life), through the 6th Framework Programme (projects SHARE-I3, RII-CT-2006062193, COMPARE, CIT5- CT-2005-028857, and SHARELIFE, CIT4-CT-2006-028812) and through the 7th Framework Programme (SHARE-PREP, N◦ 211909, SHARELEAP, N◦ 227822 and SHARE M4, N◦ 261982). Additional funding from the US National Institute on Aging (U01 AG09740-13S2, P01 AG005842, P01 AG08291, P30 AG12815, R21 AG025169, Y1-AG-4553-01, IAG BSR06-11 and OGHA 04-064) and the German Ministry of Education and Research as well as from various national sources is gratefully acknowledged (see www.share-project.org for a full list of funding institutions). ∗ Correspondence to: University of Wisconsin-Madison, La Follette School of Public Affairs, 1225 Observatory Drive, Madison, WI 53706, United States. Tel.: +1 203 785 5760. E-mail address: [email protected].
http://dx.doi.org/10.1016/j.econlet.2014.10.004 0165-1765/© 2014 Elsevier B.V. All rights reserved.
and impacts of social network on child, adolescent and young adult outcomes, much less research has examined related questions for older age individuals. In part this difference is based on data availability; while school-based surveys such as Add Health are a natural way to collect social network information among youth, adults do not have a ‘‘natural’’ context in which to collect complete network data. Instead of attempting to collect complete social network information from adults, most surveys collect ego-networks, where each respondent describes the identities, characteristics, and frequency of contact with a set of network ties, which often include spouses, family, co-workers, and friends among others. From this literature, there is an emerging set of stylized facts about adult networks. First, contrary to some earlier evidence, some aspects of adult social networks are enhanced with age, including frequency of socializing with neighbors, volunteering, and religious participation (Cornwell et al., 2008), though there does seem to be evidence that age is negatively related to network size and closeness to members. In addition to the general age patterns related to social networks, the impacts of life events on social networks have been a
460
J.M. Fletcher / Economics Letters 125 (2014) 459–462 Table 1 Summary statistics. SHARE wave 4, ages 55–70 (N ∼ 31,000). Variable
Obs
Mean
SD
Min
Max
Social network satisfaction Size of social network Average closeness with social network No reported friendships in social network Number of friendships reported Retired Age Male Education category (country-specific) Married Missing education data
30 600 31 144 29 593 31 144 29 887 31 144 31 144 31 144 31 144 31 144 31 144
8.83 2.53 3.26 0.69 0.51 0.52 62.29 0.45 4.90 0.45 0.20
1.41 1.60 0.63
0 0 −0.5 0 0 0 55 0 1 0 0
10 7 4 1 7 1 70 1 24 1 1
0.90 4.48 3.85
Industry categories not reported.
major focus in research in this area. Some of the most convincing research has used longitudinal network data to leverage withinperson designs to compare networks before and after common life events (Kalmijn, 2012). A smaller literature has shown that there are also bi-directional relationships between social network ties and retirement (Lancee and Radl, 2012). The current paper extends methods from the economics of aging literature to estimate causal impacts of retirement by leverage across-country age-specific pension policy eligibility rules that have been shown to induce retirement (Coe and Zamarro, 2011; Rohwedder and Willis, 2010). I extend this method to the social network literature and present new evidence suggesting that retirement plays a limited role in shaping social network composition and characteristics in the short term. 2. Data This paper uses the Wave 4 collection of the Survey of Health, Ageing and Retirement in Europe (SHARE) data series to explore the difference in social networks of retired versus non-retired individuals. Social network characteristics have only be assessed in a new module beginning in Wave 4, so that only cross-sectional associations can be examined.1 The Social Networks module asks about relationships (i.e. family members, friends, neighbors, and acquaintances) during the previous 12 months and focuses on ‘‘discussion’’ networks: ‘‘who are the people with whom you most often discussed important things?’’ Each contact is then described based on relationship (wife/friend/etc.), frequency of contact, proximity, and closeness (‘‘how close do you feel to X?’’). The respondents are also asked about their overall network satisfaction. This type of network description is called an ‘‘ego network’’ because it only asks questions of a single individual and then spans out from that individual’s self-reported network. No network ‘‘alters’’ are included in the survey and there is no verification of whether these network links exist from the alter’s perspective.2 Summary statistics are presented in Table 1 for the sample of SHARE respondents who are between the ages of 55 and 70. This age restriction is made in order to focus on short term impacts of retirement for respondents in the age range that is relevant for initial retirement; indeed, 52% of individuals in these ages are retired. Overall, the individuals report being highly satisfied with their overall network (average 8.8 of 10) and feel very close to the network members (average 3.2 of 4). The average number of friendships in the network is 0.5 and nearly 70% of respondents do not report any friends within this ‘‘discussion network’’, but rather report spouses, children, and other family members with great frequency.
3. Empirical methods This paper begins with descriptive regressions that estimate associations between retirement and the characteristics of individuals’ social networks: networkic = β1 retiredic + β2 Xic + β3 Cc + µic
(1)
where individual i in country c has a social network characteristic network (described be closeness, size, satisfaction, lack of friend nomination) that is a function of whether the individual is currently retired, individual level characteristics X (including age, gender, education level,3 marital status, and industry in which the individual was employed or is currently employed), country level fixed effects C and an idiosyncratic error term. In order to then examine causal impacts of retirement, the empirical design of this paper extends work by Rohwedder and Willis (2010) (RW), who estimated the impacts of retirement on cognitive decline using the SHARE data. In order to estimate causal effects of the endogenous retirement choice in their models, the authors leverage across-country variation in pension eligibility ages that in theory should induce some individuals to retire ‘‘earlier’’ or ‘‘later’’ than the respondent’s preferred retirement age if there were no pension policies. Indeed, RW show that these pension policies increase the age-adjusted retirement likelihood by between 15 and 20 percentage points (see RW Table 1). Likewise, the current study uses an updated4 set of acrosscountry policies that take into account the additional countries that are now available in Wave 4 in the SHARE Data. This allows the following two-stage approach using instrumental variables. networkic = β1 retiredic + β2 Xic + β3 Cc + µic
(2)
retiredic = α1 (ageic > eligiblec ) + α2 Xic + α3 Cc + εic
(3)
where β1 is the coefficient of interest and, given a Local Average Treatment Effect (LATE) interpretation (Imbens and Angrist, 1994), will provide the effect of retirement on network characteristics for individuals induced to retired based on their country-specific pension eligibility policy. 4. Results Table 2 presents OLS results following Eq. (1) that estimate adjusted associations between retirement status and social network characteristics. As noted in the tables, industry indicators and country-level indicators are also controlled, following results from the literature that note large differences in the patterns
1 http://www.share-project.org/fileadmin/pdf_questionnaire_wave_4/_Share_ wave_4.8.8.pdf. 2 In contrast, studies like the Add Health survey entire schools in order to collect
3 http://www.share-project.org/_questionnaire/dev2.php?welleid=26. See for the definitions of the country-specific codes. 4 http://www.ssa.gov/policy/docs/progdesc/ssptw/2012-
complete (within-school) network data from all egos and alters.
2013/europe/ssptw12europe.pdf.
J.M. Fletcher / Economics Letters 125 (2014) 459–462
461
Table 2 OLS analysis of associations between retirement and social networks. Outcome Fixed effects
Network satisfaction Industry, Country
Retired
−0.006
Average closeness Industry, Country
No friend Industry, Country
0.064** (0.025)
−0.014
−0.017***
0.003 (0.002)
−0.007***
−0.002*
(0.003)
(0.001)
−0.158***
−0.484***
−0.024***
(0.023) Age Male
Network size Industry, Country
(0.009)
(0.006)
Number of friends Industry, Country 0.036*** (0.014)
0.003*** (0.001)
−0.004***
0.109*** (0.007)
−0.167*** 0.030*** (0.004)
(0.001)
(0.017)
(0.021)
(0.007)
Education
0.003 (0.003)
0.051*** (0.005)
0.004** (0.002)
−0.015***
Married
0.223*** (0.025)
0.081*** (0.026)
0.162*** (0.010)
0.150*** (0.007)
−0.286***
Missing education category
0.024 (0.023)
0.032*** (0.010)
−0.057***
(0.037)
0.044*** (0.012)
Constant
8.837*** (0.138)
3.169*** (0.170)
3.665*** (0.060)
0.437*** (0.045)
0.882*** (0.083)
Observations R-squared
30,482 0.032
31,017 0.070
29,481 0.143
31,017 0.097
29,773 0.090
−0.030
(0.002)
(0.013)
(0.017) (0.021)
Standard errors in parentheses, clustered at the country/birth year level. * p < 0.1. ** p < 0.05. *** p < 0.01.
Table 3 Two-stage least squares analysis of effects of retirement on social networks. Outcome Fixed effects
Network satisfaction Industry, Country
Network size Industry, Country
Average closeness Industry, Country
No friend Industry, Country
Number of friends Industry, Country
Retired
0.029 (0.113)
−0.111
0.022 (0.045)
−0.025
−0.020
(0.140) 31,017 0.071
29,481 0.141
Observations R-squared
30,482 0.032
(0.039)
(0.074)
31,017 0.101
29,773 0.093
Standard errors in parentheses, clustered at the country/birth year level. Same controls as Table 2 (not shown). ∗ p < 0.1. ∗∗ p < 0.05. ∗∗∗ p < 0.01.
of social networks and work force participation across European countries in the SHARE sample5 (Kohli et al., 2009). Following some previous work, the differences in network characteristics between retired and non-retired individuals are mixed. On one hand, the associations suggest that retired individuals have slightly larger networks and are less likely to report zero friends in their ‘‘discussion network’’; there is also some evidence of reduced levels of reported closeness with members in one’s social network. The more general finding is that, while there are some differences between retired and non-retired, they are small. Because retirement is itself a choice, it is likely premature to consider the results in Table 2 as representing causal impacts of retirement status on social networks. Indeed, the small ‘‘effects’’ in Table 2 could reflect that individuals who retire do so in circumstances and at a time in which their social networks would be minimally affected. Table 3 then separates causal and noncausal effects of retirement by leveraging across-country agedependent eligibility policies for pensions. The assumption is that these policies are not correlated with age profiles of social network characteristics within each country (note: country fixed effects are controlled in these analyses). Table 3 shows that the within-country instruments, which include indicators for whether each respondent is eligible for early retirement or eligible for regular retirement, (Rohwedder
5 Findings upon request show that the results are sensitive to omitting country level indicators.
and Willis, 2010) are strong, with an F -statistic above 50 (Staiger and Stock, 1997) (first stage results available upon request). As is typical with instrumental variables, the standard errors increase in comparison to the OLS estimates in Table 2, often by a factor of 4 or 5. This limits some ability to clearly accept or reject a null of zero effects or of small effects of retirement. In general, I find no statistical evidence of effects on social networks in these analyses, and the estimates can rule out moderate effect sizes. For example, the standard deviation for the average closeness measure is 0.63 and the estimate and standard error on the impact of retirement is 0.03 (0.045), suggesting that retirement has no more than modest impacts on measures of closeness. 5. Conclusions This paper presents new causal evidence of the impacts of retirement on social network characteristics. The literature on this topic suggests that important life events, such as retirement, can have large impacts on social networks, but also that social networks and retirement are bi-directionally related and endogenous. Using a sample of 55–70 year old respondents from 16 European countries combined with information on age-eligibility policies of pension receipt in each country, findings suggest very little evidence of moderate or large causal effects of retirement on a variety of social network characteristics. This new evidence suggests a limited role of retirement and social networks and that other aspects of aging may be responsible for the empirical regularities in most datasets showing that social networks generally shrink in old
462
J.M. Fletcher / Economics Letters 125 (2014) 459–462
age. The results also suggest that retirement, and specifically retirement induced by policies, may have a limited role in increasing social isolation and reduced social networks among those in old age. References Coe, Norma B., Zamarro, Gema, 2011. Retirement effects on health in Europe. J. Health Econ. 30 (1), 77–86. Cornwell, Benjamin, Laumann, Edward O., Schumm, L. Philip, 2008. The social connectedness of older adults: a national profile. Amer. Sociol. Rev. 73 (2), 185–203. http://dx.doi.org/10.1177/000312240807300201.
Imbens, Guido W., Angrist, Joshua D., 1994. Identification and estimation of local average treatment effects. Econometrica 62 (2), 467–475. Kalmijn, Matthijs, 2012. Longitudinal analyses of the effects of age, marriage, and parenthood on social contacts and support. Adv. Life Course Res. 17 (4), 177–190. Kohli, Martin, Hank, Karsten, Künemund, Harald, 2009. The social connectedness of older Europeans: patterns, dynamics and contexts. J. Eur. Soc. Policy 19 (4), 327–340. Lancee, Bram, Radl, Jonas, 2012. Social connectedness and the transition from work to retirement. J. Gerontol. Ser. B: Psychol. Sci. Soc. Sci. 67 (4), 481–490. Rohwedder, Susann, Willis, Robert J., 2010. Mental retirement. J. Econ. Perspect. 24 (1), 119–138. Staiger, Douglas, Stock, James H., 1997. Instrumental variables regression with weak instruments. Econometrica 557–586.