No mental retirement: estimating voluntary adult education activities of older workers

ABSTRACT

Adult education may keep older workers productive if they voluntarily engage in learning. Examining a generous partial retirement reform in Germany that encouraged early retirement among male workers, we estimate voluntary adult education activities for early retirees. Using county-level administrative data on all public adult education centers (Volkshochschulen; VHS), we employ a difference-in-differences approach exploiting the predetermined county-specific age structure of men for identification. The results show a strong increase in participation in adult education, specifically in cognitively demanding courses. This supports an intrinsic willingness of older workers to acquire skills and abilities independent of financial incentives.

KEYWORDS:

• Partial retirement
• early retirement
• older workers
• adult education

JEL CLASSIFICATION:

• E24
• I21
• J14
• J24
• J26

Acknowledgments

We are grateful to the editor and two anonymous reviewers for helpful comments and suggestions. We also would like to thank the German Institute for Adult Education (Deutsches Institut für Erwachsenenbildung, DIE) for providing the VHS statistics, as well as to Olaf Hübler, Thomas Zwick, and conference participants at the annual meetings of the CEA 2019 (Banff, Alberta), ESPE 2019 (Bath), VfS 2019 (Leipzig), and EEA 2020 (Virtual, Rotterdam) for their very helpful comments and discussions.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 See, e.g. Belbase, Sanzenbacher, and Gillis (2016), Hudomiet et al. (2018), Kotlikoff and Gokhale (1992), Lorenz, Zwick, and Bruns (2022), Spirduso, Francis, and MacRae (1995), and Skirbekk (2004). The number of days absent from work due to illness rise monotonously with age (BKK Dachverband 2019, p. 69), thus increasing costs for the employer. Older workers may also completely drop out of the workforce for health reasons (Buslei et al. 2019). Moreover, older workers may worry about future job demands and may therefore stop working earlier (Hudomiet et al. 2021). At the same time, however, many studies find limited evidence for notable productivity declines, e.g. Börsch-Supan and Weiss (2016), Colonia-Willner (1998), Göbel and Zwick (2013) and Van Ours (2009). In fact, De Grip et al. (2015) find that retirees face lower declines in their cognitive flexibility than those who remain employed, although other studies (see, e.g. Mosca and Wright 2018) confirm that there are measurable cognitive declines in retirement.

2 Studies on discrimination against older workers show that many firms prefer to hire younger workers if they can (see, e.g. Johnson and Neumark 1997; Lahey 2008; Neumark 2018; Neumark, Burn, and Button 2019). Moreover, some countries have enacted retirement policies that actively encourage early retirement to make jobs ‘available’ for younger job-seekers (Casey 1996; Eichhorst et al. 2014); even though there is little empirical support for the assumed substitutability of older and younger workers (Böheim and Nice 2019; Eichhorst et al. 2014; Wise 2010).

3 In this study, we refer to partial retirement as the possibility of drawing a pension early while continuing to work part-time. Kantarci and Van Soest (2008, p. 114) refer to this as gradual retirement. They differentiate between two types of gradual retirement: partial retirement, which they define as ‘changing to a less demanding job with usually fewer hours and lower earnings’ and phased retirement, which implies ‘reducing work hours in the same job’. The reform that we study is consistent with the latter characterization of retirement.

4 Similar kinds of generalized DiDs have been used by, e.g. Berlinski, Galiani, and Gertler (2009); Havnes and Mogstad (2011), and Sandner and Thomsen (2018) in the context child care reforms.

5 In a related experimental study, Kovalchik et al. (2005) also document that the decision behavior of elderly people (average age 82) is not very different from the decision behavior of younger individuals (average age 20).

6 Most of those workers not covered are either civil servants or self-employed workers. See Börsch-Supan and Wilke (2004) and Börsch-Supan, Rausch, and Goll (2021) for excellent overviews of the German pension system and its history.

7 There were several reforms of the retirement system prior to 1996 (see footnote 6). The first reform was implemented in 1972 in reaction to mass unemployment. By enabling employees to effectively retire at the age of 59 instead of 65 without any deductions (Gatter and Hartmann 1995, p. 413), the reform lead to the (to date) most radical shortening in retirement age in the developed world (Börsch-Supan 2015) and a massive increase in the number of early retirees (Börsch-Supan and Schnabel 1998).

8 A qualifying period includes years of mandatory insurance contributions from working as well as unemployment spells and (to some degree) times of parental leave. Approximately 85% of all workers insured in the public pension system fulfilled these eligibility criteria (Lorenz, Zwick, and Bruns 2022).

9 Even before the PR act was passed, a number of collective bargaining agreements, e.g. in the insurance and banking sector or the chemical industry, already included early retirement options similar to PR (Schmähl 2003). However, they were hardly used because their conditions were less attractive (Lorenz, Zwick, and Bruns 2022).

10 For smaller companies with a maximum of 50 employees training an apprentice instead ensured the employer's exemption from the supplemental payments. Wanger (2009) reports that subsidized PR contracts constituted approximately 35% of all PR contracts and that the subsidies of the FEA were substantial. The cumulative expenses from 1996 to 2007 amounted to 7.2 billion euros.

11 The high net salary was facilitated by progressive income taxation. Berg et al. (2020, pp. 1228-9) show that the negotiated supplements even exceeded these numbers in many industries.

12 From 1996 to 1999, individuals had to work at least 18 hours per week after reducing their working time by half in order to be eligible. In 2000, the program was extended to persons who worked part-time before entering PR.

13 Berg et al. (2015, Appendix Table A2) show that a number of collective bargaining agreements limited the number of PR contracts to 5% of the workforce.

14 Wanger (2010, Figure 5) shows that in 2007 more than half of all workers in PR came from the public sector or the manufacturing sector. As a share of all workers aged 55 to 64 years within a given industry, energy and water companies had the highest share (44%), followed by the public administration and the banking and insurance sector (36% each). Wanger (2010, table 4) shows that there was no typical PR job, i.e. no single job dominated in PR contracts, although white-collar jobs were over-represented.

15 Appendix Figure B-1 presents information on the distribution of the durations of PR contracts that were subsidized by the FEA between 2002 and 2007 based on Wanger (2009, p. 5).

16 Note that we use the population share and not the share in employees because of endogeneity issues. Brussig, Knuth, and Wojtkowski (2009, p. 5) and Wanger (2009, pp. 3-4) use the share of PR workers in employees and arrive at numbers that are more than twice as large as ours.

17 Note that this option was limited because employees in subsidized contracts were not allowed to earn more than the marginal earnings threshold (Geringfügigkeitsgrenze) defined by Chapter 4 of Social Code Book XII (see §5 AltTZG). This corresponded to 400 Euro until 2013 and 450 Euro afterward. Employees in non-subsidized contracts had to follow the rules lined out in the contract, often the sectoral agreement. Second jobs were usually possible but had to be negotiated with the employer.

18 Consistent with this prediction, Burkhardt and Schupp (2019) document, based on the German Socio-Economic Panel (SOEP), that the ‘generation of 1968’, which they define as born between 1941 and 1954 and which largely overlaps with the generation analyzed in this paper, has so far been the most active retired generation around and after retirement. Similar observations have been made for the United States, where Ameriks et al. (2020) show that many retirees express a strong willingness to work and that this willingness increases significantly if working hours can be chosen flexibly. This argument is strenghtened by the observation that those sectors with high PR rates (see footnote 14) also have above-average participation rates in non-formal continued education (see, e.g. Dummert 2018, table 23a).

19 Looking from the other direction, Stenberg, de Luna, and Westerlund (2012) detect no effect of adult education on the timing of retirement, whereas Picchio and van Ours (2013) find that older workers who receive training are more likely to remain employed.

20 Berg et al. (2020) exclude all years after 2004 because during this time, PR was reformed in the course of the Hartz III legislation. The Hartz reforms dramatically changed the retirement behavior in Germany. (However, our event studies show that this does not affect our results.) As a consequence, there is only a small overlap with the period analyzed in this paper.

21 Appendix A provides details on the long history and legal basis of VHS.

22 See https://www.die-bonn.de/weiterbildung/statistik/vhs-statistik/default.aspx?lang=en&. The latest developments are summarized in yearly reports; see, e.g. Huntemann et al. (2020).

23 The VHS statistics list age group and gender separately before 2008.

24 While these numbers refer to nonformal activities, i.e. organized that does not necessarily lead to a certificate, we do not cover informal learning activities (e.g. reading newspapers and magazines, having discussions), which are not formally organized in any way.

25 This number includes free cities that do not belong to a county. For simplicity, we refer to both as counties.

26 The estimated median net income of a German employee was equal to €20,053 per year or €1,671 per month in 2014 (Krause, Franz, and Fratzscher 2017).

27 The number of guest students aged between 55 and 64 is not available for all states as a time series. In states that published numbers, these are negligible. This probably reflects the higher effort and commitment that comes with enrolling.

28 Missing information is mainly due to VHS that do not report data for every course and every year. However, the missing values are evenly distributed across years and states without any clear pattern.

29 The average county in our sample registers approximately 1,610 VHS courses with a standard deviation (SD) of 1,382 courses. On average, there are 210 work-related courses per VHS, with a SD of 195. As for the county population, we observe that the VHS offer 8.4 total courses and 1.1 work-related courses per 1,000 inhabitants on average.

30 This time allocation is about twice as large as for almost all other course areas (health: 16 hours/course; arts & culture: 18 hours/course; politics & society: 16 hours/course) and about the same amount as for language courses (36 hours/course).

31 During the time of our analysis, there were a few redivisions of county borders. We take this into account by using conversion keys for county population and area that were provided by the Federal Institute for Building, Urban Affairs and Spatial Research (BBSR). As a result, every county observation is available within its 2017 borders for the whole observation period.

32 Ideally, we would have used the share of 55 to 64-year-old men because PR could not be started before the age of 55 years. However, the VHS statistics do not offer more detailed age groups or sample splits along the age-gender dimension for the sample period.

33 Similar approaches can be found in, e.g. Bauernschuster, Hener, and Rainer (2016); Berlinski, Galiani, and Gertler (2009); Havnes and Mogstad (2011), and Sandner and Thomsen (2018) in the context child care reforms and Cantoni et al. (2017) in the context of evaluating school curricula reform regarding students' political attitudes in China.

34 The results do not change when we also control for gender-specific population groups (results not shown).

35 Appendix Figure B-5 shows considerable variation in the counties' population share of 55 to 64-year-old men, even after conditioning on year and county fixed effects.

36 Appendix Table B-4 shows the results of a triple difference-in-differences estimation, examining whether the treatment effect of the reform varies with the regional unemployment rate. This is not the case. This also implies that (the threat of) unemployment does not play a role for the individual decision to enroll in adult education.

37 To estimate a treatment-on-the-treated effect (ATT), we would have to know how many persons opted (had to opt) into PR and how their adult learning behavior developed relative to those persons in earlier or later years who could not retire early.

38 Since GDP per capita at the county level had to be imputed for counties in Lower Saxony and Saarland (see Table 2), we added a dummy indicating the imputation in the subsequent analysis. The results are unaffected. To ensure that the results were not driven by a few outlier counties, we also dropped county observations at or below the 1st and above the 99th percentile of the population share of 55 to 64-year-old men in each year. Appendix Table B-1 shows that the results were hardly affected.

39 Appendix Table B-2 show that the results are qualitatively the same when we use VHS participations of 50 to 64-year-old persons in work-related courses per 1,000 inhabitants who are 50 to 64 years old as the dependent variable.

40 The adult education survey (AES 2014) shows that individuals spend on average 36 hours on adult education activities within the last 12 months prior to the survey (see Bundesministerium für Bildung und Forschung 2015, p. 42, table 22).

41 In addition, we have also estimated a synthetic difference-in-differences model (Arkhangelsky et al. 2021). While the results from this analysis, which are available from the authors upon request, largely confirm our findings, they also reveal relatively large effect heterogeneity for some specific counties.

42 For the estimation, we use Stata's did_multiplegt command by de Chaisemartin and D'Haultfæuille.

43 The notion that VHS activities are independent of the PR reform is also strengthened by the fact that we did not find any correlation between the PR reform on the one hand and the number of VHS employees, the number of participants per course, or the expenses on advertisements on the other hand (not shown).