Job Loss, Retirement and the Mental Health of Older Americans

1. Job Loss, Retirement and the Mental Health of Older Americans Bidisha Mandal Brian Roe The Ohio State University

2. Outline • Motivation • Literature • Data • Model • Results • Conclusion • Future Research

3. Motivation • Increasing percentage of older individuals in the population. • General decline in job security in U.S. labor market. • Physical limitations, cognitive changes, bereavement are commonly associated with aging. • Does work displacement cause additional distress? Are there any long-term effects? • Job loss – skills may not be transferable, loss of income • Retirement – lifestyle changes. • Policy implication – increased private medical expenditure, increased public spending for government medical programs.

4. Relevance • Mental health affects social behavior, morale, as well as work productivity. • Deteriorating mental health can manifest in weakened physical health and increase likelihood of suicide. • Declines in the mental health may negatively influence the well-being of other household members. • Older Americans may be less inclined to seek help for psychological problems (as compared to physical decrements). • Job loss affects the quality of life

5. Literature • Retirement • Kim and Moen (2002): 458 New York employees; 1994, 1996, 1998 waves of Cornell Retirement and Well-being study. Results: short-term boost in morale, and long-term increase in distress levels for men. • Drentea (2002): 2 different cross-sectional national surveys Mixed results: lower sense of control, but lower anxiety levels among retirees. • Midanik et al. (2005): 595 members of a health maintenance organization; short-term effect. Result: lower stress levels among retirees. • No clear trend; No long-run national panel have been studied yet.

6. Related Literature on Retirement • Kerkhofs et al. (1999): health and retirement are endogenously related. • Dwyer and Mitchell (1999), Disney et al. (2006): health problems influence retirement plans more strongly than economic variables. • Involuntary Job Loss (business shut-down or lay-off) • Gallo et al. (2000): 1992 and 1994 waves of Health and Retirement Study. • Different methodology to handle endogeneity • Reverse causality and unobserved heterogeneity • OLS vs. HT/IV and 2SLS • Alternative coding

7. Framework

8. CESD Score • Mental health measure • Developed by Radloff (1977) – short, self-reporting scale (20 items) for general population. • HRS only includes 8 items – 6 negative and 2 positive binary indicators • Negative items – felt depressed, everything an effort, sleep was restless, felt lonely, felt sad, could not get going • Positive items – was happy, enjoyed life • CESD = sum (negative items) – sum (positive items) Thus, higher score (0 to 8) means worse mental health. • Both versions commonly used in other studies to measure distress and psychological well-being.

9. Summary – CESD Score • Reliability • Cronbach’s alpha coefficient for 20 items = 0.85 • Cronbach’s alpha coefficient for 8 items = 0.71 • Mean change in CESD score among those who suffered involuntary job loss is 0.19 • Mean change in CESD score among retirees is 0.17 • Maximum increase in CESD score is reported between the first two waves (1992 to 1994), when job loss rates were high. • CESD scores improve during latter waves for all.

10. Coding Involuntary Job Loss and Retirement • Unbalanced panel data • 6 waves – 1992, 1994, 1996, 1998, 2000 and 2002 • N=7,780 (all those employed in 1992, 51-61 years old) • Coding • Survey does not ask R if suffered involuntary job-loss, but reason for unemployment. • Involuntary job loss • If R reports business closure or layoff, and started looking for job immediately. • Retirement (voluntary) • If R accepts early retirement incentives, and does not look for job immediately. These individuals also call themselves – ‘self-retired’. • Plus, those who report retirement as labor market status. • Data limitations

11. Data Mean change in CESD score between 1992 and 1994 Distribution of HRS respondents in different labor market situations

12. Summary Statistics (selected variables)

13. Unobserved Heterogeneity • Compare fixed effects, random effects and Hausman-Taylor IV random effects model using Hausman specification test FE: where, are time-varying independent variables RE: where, are time-invariant independent variables and, denotes individual-specific effects HT-IV: where, the subscripts distinguish between exogenous and endogenous variables • First difference model:

14. Comparing Model Properties • FE • Subtracts off group means • Along with time-invariant regressors, latent effects are left out • FD • Similar, but subtracts off last period’s observations • Again, gets rid of both time-invariant factors and latent effects • Unbiased, consistent estimates from both FE and FD • RE • Can use time-invariant variables, as long as independent of latent effects • Efficiency gain • HT-IV RE • Allows time-invariant variables under lesser constraints • Correct specification produces consistent, unbiased and efficient estimates • Limitation – single-equation model; model misspecification

15. FE, RE and HT-IV RE Models Estimates (SE) from different models for selected variables Dependent variable: CESD score * p < 0.01; ** p < 0.05

16. Model Choice • Latent effects – motivation, productivity • Time-varying endogenous variables – involuntary and voluntary exits, marriage/remarriage, separation/divorce, ADLA index, physical health condition • Time-invariant endogenous variables – age, education, white/blue-collar job • Choice of model (FE vs. RE) depends on cost of efficiency gain • Only one time-invariant variable significant - gender • First difference model is adequate in controlling for latent effects and is able to capture the change in mental health due to a shock

17. Compare with Previous Study • Gallo et al. (2000) use data from 1992 and 1994 HRS. • Sample selection is sufficient to take care of latent effects – exclude retirees, self-employed individuals, disabled, and those who left their jobs for reasons other than plant closure and lay-off. • Involuntary job loss – plant closure and lay-off • Method – OLS regression. • Replicate their coding and methodology, and obtain estimate of involuntary exit similar to theirs. • Problem – unobserved heterogeneity still exists

18. Reverse Causality • Suspect endogenous variables • Involuntary exit • Voluntary exit • Separation/Divorce • Marriage/Re-marriage • Instruments (excluded exogenous variables) • Unemployment rate • Age at the beginning of each survey • Parents’ level of education • R’s level of education • If R’s parents are/were married to each other or to step-parents • Number of divorces and widowhoods reported in 1992

19. Validity • Three basic tests to • Check if endogeneity actually exists • Ho: suspect endogenous variables are exogenous Compare 2 regressions – one where suspect regressors are treated as endogenous, and the other where they are exogenous. Test statistic is distributed χ2 with df = number of endogenous regressors • Check for weak instruments • LR test - Ho: equation is underidentified To check if the instruments are poor proxies for the endogenous variables. Test statistic is distributed χ2 with df = total number of exogenous regressors - endogenous regressors + 1 • Check the validity of the instruments • J statistic - Ho: instruments are uncorrelated with error Test statistic is distributed χ2 with df = number of instruments - 1

20. Results from Labor Market Exit 2SLS regression: Dependent variable – ΔCESD score E - endogenous * p < 0.01

21. Results from Re-entry after Job Loss 2SLS regression: Dependent variable – ΔCESD score E – endogenous; * p < 0.01; ** p < 0.05

22. Results from Re-entry after Retirement 2SLS regression: Dependent variable – ΔCESD score E – endogenous; * p < 0.01; ** p < 0.05

23. Involuntary Exit vs. Reemployment 2SLS regression: Dependent variable – ΔCESD score E – endogenous; * p < 0.01; ** p < 0.05

24. Summary of Steps • First Difference Model • Accounts for Unobserved Heterogeneity • To capture the effect of change in labor market status on change in mental health • Reverse Causality • Mental health may decide labor market status • Two Stage Least Squares (2SLS) • Endogenous Regressors • Labor market status • Marital status (except widowhood) • Effect of Reemployment

25. Conclusion • Endogenous regressors are labor market exit and re-entry, separation or divorce, and marriage or remarriage • Involuntary job loss negatively impacts mental health • Similar in magnitude and direction to effect of death of child • Highest negative effect due to death of spouse • Retirement has a positive effect on mental health (short-term) • Re-entering labor market has a positive effect on the mental health for all • Re-entry recaptures the previous mental health status of those who lost job involuntarily