Cross-national differences in the timing and type of labour market exit

Presented at 2015 SLLS conference
Dublin, 19th October, 2015

Transcript

1. Cross-national differences in the timing and type of labour market exit Ewan Carr and Jenny Head University College London October 19th 2015 1 / 35
2. renEWL project (2014-17) Research into determinants of Extended Working Lives • ESRC/MRC funded project on the determinants of extended working (beyond age 50) • Collaboration of researchers from UCL, QMUL, Stockholm University and the Karolinska Institute. • Four domains: 1 Work environment 2 Family arrangements 3 Area effects 4 Cross-national comparisons www.ucl.ac.uk/renewl @EWLresearch 2 / 35
3. renEWL project (2014-17) Research into determinants of Extended Working Lives • ESRC/MRC funded project on the determinants of extended working (beyond age 50) • Collaboration of researchers from UCL, QMUL, Stockholm University and the Karolinska Institute. • Four domains: 1 Work environment 2 Family arrangements 3 Area effects 4 Cross-national comparisons Cross-national comparisons www.ucl.ac.uk/renewl @EWLresearch Cross-national comparisons 2 / 35
4. Population ageing, 1851-2011 1908: State pension introduced at age 70 1851 1871 1891 1911 1931 1951 1971 1991 2011 10 12 14 16 18 20 22 Women Men Life expectancy at 65 Source: Decennial Life Tables, ONS 3 / 35
5. Employment rates keeping up? 4 / 35
6. Employment rates among older workers (55-64) 5 / 35
7. Effective vs. official retirement ages, 2012 6 / 35
8. Aims of the study 1. To derive a consistent measure of age of labour market exit across 9 longitudinal datasets. 7 / 35
9. Aims of the study 1. To derive a consistent measure of age of labour market exit across 9 longitudinal datasets. 2. To identify the type of exit (e.g. health-related). 7 / 35
10. Aims of the study 1. To derive a consistent measure of age of labour market exit across 9 longitudinal datasets. 2. To identify the type of exit (e.g. health-related). 3. To consider associations with sex, education, occupational grade. 7 / 35
11. This talk 1 Why is consistent measurement of labour market status hard to achieve? 8 / 35
12. This talk 1 Why is consistent measurement of labour market status hard to achieve? 2 How have we measured age of work exit and health-related exit? 8 / 35
13. This talk 1 Why is consistent measurement of labour market status hard to achieve? 2 How have we measured age of work exit and health-related exit? 3 Some results, based on these measures. 8 / 35
14. This talk 1 Why is consistent measurement of labour market status hard to achieve? 2 How have we measured age of work exit and health-related exit? 3 Some results, based on these measures. 4 What are the problems with these measures? 8 / 35
15. This talk 1 Why is consistent measurement of labour market status hard to achieve? 2 How have we measured age of work exit and health-related exit? 3 Some results, based on these measures. 4 What are the problems with these measures? 5 Conclusions 8 / 35
16. 1. Why is consistent measurement hard? 9 / 35
17. 1: Retirement is a poor indicator of work status 10 / 35
18. 1: Retirement is a poor indicator of work status 10 / 35
19. 1: Retirement is a poor indicator of work status If we’re interested in the timing of labour market exit. . . Retirement is not a useful analytical concept • Often just a reflection of the institutional context (e.g. statutory pension age) 11 / 35
20. 1: Retirement is a poor indicator of work status If we’re interested in the timing of labour market exit. . . Retirement is not a useful analytical concept • Often just a reflection of the institutional context (e.g. statutory pension age) • Retirement often combined with work • Partial retirement, gradual retirement, bridge employment • Post-retirement employment (‘unretirement’) • Retired from ‘main career job’, but still working 11 / 35
21. 1: Retirement is a poor indicator of work status If we’re interested in the timing of labour market exit. . . Retirement is not a useful analytical concept • Often just a reflection of the institutional context (e.g. statutory pension age) • Retirement often combined with work • Partial retirement, gradual retirement, bridge employment • Post-retirement employment (‘unretirement’) • Retired from ‘main career job’, but still working • Multiple roles during retirement (caring, volunteering) 11 / 35
22. Working hours by self-reported job status 12 / 35
23. Employment income by self-reported job status 13 / 35
24. 2. Increased variation in retirement timing “we are seeing increased variability in the ages at which people self-define as being re- tired, with some opting for early retirement and others working into their 70s and 80s.” — Kim and Wang (2013) 14 / 35
25. 2. Increased variation in retirement timing 15 / 35
26. 3. Range of data sources and collection methods Finland 1. Finnish Public Sector Study France 2. GAZEL study Sweden 3. Swedish Longitudinal Occupational Survey of Health UK 4. MRC National Study of Health and Development 5. Whitehall II 6. English Longitudinal Study of Ageing 7. British Household Panel Survey USA 8. Health and Retirement Study Europe 9. Survey of Health, Ageing and Retirement in Europe 16 / 35
27. 2. How have we measured age of labour market exit? 17 / 35
28. Measure: age of work exit Repeated panels (e.g. ELSA, BHPS and HRS) • A reduction in working hours between two consecutive waves (from > 0 hours/week → 0 hrs/week). • We derive an exit age from: a) The mid-point between two interview dates; or b) Information collected at the next non-missing interview, after work exit (“When did your last job end?”) 18 / 35
29. Measure: age of work exit 0 0 1 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 Wave Working? Last wave in work First wave not working 19 / 35
30. Measure: age of work exit Actual matching 45 50 55 60 65 Age of exit (retrospective data) 45 50 55 60 65 Age of work exit (based on mid-point) Exact matching Source: ELSA 20 / 35
31. Measure: age of work exit Other studies vary • Register-linked studies (e.g. FPSS) have exact dates of employment spells • Some occupational cohorts provide specific exit dates • Others (e.g. birth cohorts) just have work status at a given age (e.g. 53, 64) 21 / 35
32. Measure: type of work exit 1 Health-related 2 Non-health-related 3 Other 22 / 35
33. Measure: type of work exit 1 Health-related 2 Non-health-related 3 Other 1. Health-related work exit 1 The individual cites own ill-health as a “reason for stopping work” or “reason for retiring”; or 2 The individual starts to receive a health-related pension or benefit within 12 months of stopping work. 22 / 35
34. Measure: type of work exit 1 Health-related 2 Non-health-related 3 Other 2. Non health-related work exit The individual does not meet the criteria for (1), and: 1 Self-reports as being ‘retired’; or 2 When asked “at what age did you retire” the date given is within 12 months of age of work exit’; or 3 Starts receiving old-age or statutory retirement pension, within 12 months of stopping work. 22 / 35
35. Measure: type of work exit 1 Health-related 2 Non-health-related 3 Other 3. Other work exit The individual does not meet the criteria for (1) or (2). 22 / 35
36. 3. Results (so far. . . ) 23 / 35
37. Available sample sizes Eligible for sample Right censored Work exit observed BHPS 5,691 3,351 2,340 ELSA 4,180 2,897 1,193 FPSS 73,670 50,913 22,757 GAZEL 19,118 515 18,603 HRS 14,689 5,446 9,243 Whitehall II 7,913 1,787 6,126 Total 125,261 64,999 60,262 24 / 35
38. Exit types by study 25 / 35
39. Cox PH for health-related exit: Education BHPS ELSA FPSS GAZEL HRS W2 HR HR HR HR HR HR Low 3.76 1.90 3.00 4.27 3.13 1.67 [2.68-5.28] [1.33-2.72] [2.81-3.22] [3.18-5.72] [2.73-3.56] [1.36-2.06] Middle 1.71 1.29 2.36 3.71 1.72 1.17 [1.20-2.43] [0.93-1.81] [2.22-2.51] [2.82-4.89] [1.51-1.96] [0.91-1.51] High Reference. N 5,032 4,172 73,531 7,680 13,204 7,913 Adjusted for sex. 95% C.I. in brackets. 26 / 35
40. Cox PH for health-related exit: Occupation BHPS ELSA FPSS GAZEL HRS W2 HR HR HR HR HR HR Low 3.77 2.67 0.86 9.16 3.46 4.14 [2.42-5.88] [1.90-3.74] [0.79-0.93] [6.79-12.36] [2.84-4.20] [3.19-5.36] Middle 1.21 1.47 0.95 4.88 2.14 2.19 [0.75-1.94] [0.92-2.33] [0.90-1.01] [3.79-6.26] [1.77-2.60] [1.76-2.73] High Reference. N 5,067 3,332 70,555 19,100 8,766 7,913 Adjusted for sex. 95% C.I. in brackets. 27 / 35
41. 4. What’s wrong with these measures? 28 / 35
42. Returns to work, after initial exit • We’ve derived an age of work exit, based on changes in weekly working hours. 29 / 35
43. Returns to work, after initial exit • We’ve derived an age of work exit, based on changes in weekly working hours. • Working hours are widely and consistently reported; not simply a reflection of institution context. 29 / 35
44. Returns to work, after initial exit • We’ve derived an age of work exit, based on changes in weekly working hours. • Working hours are widely and consistently reported; not simply a reflection of institution context. • Yet, we know our measure is wrong. 29 / 35
45. Returns to work, after initial exit • We’ve derived an age of work exit, based on changes in weekly working hours. • Working hours are widely and consistently reported; not simply a reflection of institution context. • Yet, we know our measure is wrong. • In the absence of mortality data, respondents can always return to work. 29 / 35
46. Returns to work, after initial exit • We’ve derived an age of work exit, based on changes in weekly working hours. • Working hours are widely and consistently reported; not simply a reflection of institution context. • Yet, we know our measure is wrong. • In the absence of mortality data, respondents can always return to work. 0 0 1 1 1 1 0 0 0 Working? 47 48 49 50 51 52 53 54 55 Age 29 / 35
47. Returns to work, after initial exit • We’ve derived an age of work exit, based on changes in weekly working hours. • Working hours are widely and consistently reported; not simply a reflection of institution context. • Yet, we know our measure is wrong. • In the absence of mortality data, respondents can always return to work. 1 1 0 1 1 1 0 0 0 Working? 47 48 49 50 51 52 53 54 55 Age 29 / 35
48. Sensitivity analyses 1 1 0 1 1 1 0 0 0 Working? Episode 1 Episode 2 47 48 49 50 51 52 53 54 55 Age 30 / 35
49. Sensitivity analyses 1 1 0 1 1 1 0 0 0 Working? Episode 1 Episode 2 47 48 49 50 51 52 53 54 55 Age 1 Must remain out of labour market for N years before considered to have exited (N = 1..5). 30 / 35
50. Sensitivity analyses 1 1 0 1 1 1 0 0 0 Working? Episode 1 Episode 2 47 48 49 50 51 52 53 54 55 Age 1 Must remain out of labour market for N years before considered to have exited (N = 1..5). 2 Must be minimum age at start of episode. 30 / 35
51. Results for BHPS, by year remaining out of work Years remains Remaining Low occupation Low education not working sample HR HR 1 2,384 4.34 4.13 [2.65-7.11] [2.89-5.92] 2 1,801 4.22 3.93 [2.58-6.92] [2.74-5.63] 3 1,456 4.48 3.94 [2.71-7.42] [2.74, 5.65] 4 1,142 4.42 3.85 [2.67-7.33] [2.68-5.52] 5 966 4.38 3.86 [2.64-7.25] [2.68-5.55] Adjusted for sex. 95% C.I. in brackets. 31 / 35
52. Results for BHPS, by age at episode start Age at start Remaining Low occupation Low education of episode sample HR HR 46 1,660 4.04 3.79 [2.46-6.61] [2.63-5.44] 48 1,346 3.93 3.47 [2.40-6.44] [2.41-4.99] 50 1,065 3.81 3.17 [2.32-6.24] [2.20, 4.55] 52 779 4.05 3.14 [2.45-6.71] [2.18-4.53] 54 483 4.07 2.90 [2.45-6.74] [2.01-4.19] Adjusted for sex. 95% C.I. in brackets. 32 / 35
53. 5. Conclusions 33 / 35
54. Conclusions • All Western populations are ageing, but outcomes filtered by differing socio-economic and cultural contexts. • Cross-national comparison is important, but consistent measurement is hard. • Our approach is based on working hours (widely and consistently reported), combined with registry data (where available). • Initial results suggest low education and low occupational grade are predictive of health-related work exit. • But this approach has drawbacks. • Next. . . 1 Test sensitivity to measure of health-related exit 2 Harmonise the remaining studies (NSHD, SLOSH, etc.) 34 / 35
55. Thank you ewan.carr@ucl.ac.uk @ewancarr 35 / 35