Abstract
Although the health effect of retirement has important policy implications, few economists have researched the topic. This paper utilizes longitudinal data from the Health and Retirement Study and instruments retirement status using exogenous variation in public and private pensions. Subjective health change models indicate retirement preserves the health of both men and women, although insignificant results for objective health change models suggest the preservation may be more perceived than real. The same pattern of results is found using continuous measures of annual hours. At the very least, the results give strong evidence against the anecdotal idea that retirement harms health.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
Other disciplines have examined the topic more extensively but have reached no consensus about how retirement affects health, or even if an effect exists. A significant issue with most of the studies is a failure to account for the endogeneity of the retirement decision, thus biasing the results. For a more extensive literature review analyzing studies in all disciplines see Neuman (2004).
Although the shortcomings of the linear probability model have been well documented, in this situation utilizing the estimation technique may actually be more appropriate and does not appear to affect the results. The problem with the IV probit technique for my model lies in the fact that the dummy variable for retirement choice enters into the model rather than the predicted probability of retirement. Once the predicted likelihood of retirement is used to instrument the retirement decision in the IV estimation procedure the error term is no longer normally distributed, making probit estimation inappropriate and leading to inconsistent results. Given that the linear probability model is more robust to specification errors I chose to use it in place of the probit model which is much more sensitive to these types of issues. There is also no reason to expect that the error is logistically distributed, ruling out IV logit estimation. In addition, examination of the predicted probabilities generated by the linear probability model showed that virtually all predictions lay within the 0/1 bounds of probability, alleviating concerns over a noted problem of the model. Finally, I did estimate the model using the IV probit technique as a check and found that the significance and marginal effects of the two models were virtually identical suggesting that the results are not driven by the choice of the linear probability model.
Using a 1,000-h threshold for retirement corresponding to a common work restriction for disbursements in DB pensions yielded virtually identical results.
Selecting multiple observations of the same individual conditional on working full time in the first wave of observation may potentially lead to sample selection bias if those individuals who continue working, and thus show up multiple times in the sample, are healthier than their retiree counterparts. I checked for this bias in two ways. First, I estimated the model separately for each three-wave cross section. The results showed similar patterns for both men and women but were insignificant, although this was likely due to the sample size being reduced to roughly a fifth of the total sample. Second, I estimated the model on a sample composed of individuals who appeared only once in the full sample, plus a single randomly selected observation from each individual who appeared more than once in the full sample. Repeated estimation using this random selection procedure consistently produced results that were extremely similar to the full sample results both in terms of magnitude and significance. Both tests suggest this potential selection bias has not affected the full sample results significantly.
A possible problem with defining the health change variables in this manner is if health as measured by the specific variable is already at its lowest possible point in the initial wave and therefore the individual can not possibly register a negative health change. I check for problems related to the health change variable construction by estimating the sample on only those who can register both a non-negative and a negative health change. The results were virtually identical as very few individuals reported the worst possible health in the initial wave.
The classification follows work by Quinn (1996). The ‘white-collar high skilled’ category includes managers and professionals, while the ‘white-collar other’ category includes sales and clerical occupations. Occupations in the ‘blue-collar high skilled’ category include protection services, mechanics and repair, construction and extraction, precision production, and machine and transport operators. The default category of ‘blue-collar other’ includes non-protection services; farming, forestry, and fishing; and operators and handlers.
I am not able to report the control variable coefficients due to the number of dependent variables and control variables actually used. Results are freely available upon request.
A related point that should be highlighted about the first stage results concerns the consistency of the control variable coefficients across the different instrument combinations. If the instruments are well behaved you would not expect the basic results of the model to change with the inclusion of additional instruments. Examination of the control variable coefficients across instrument combinations (results unreported) shows that for both sexes the control variables display remarkable consistency for both magnitude and significance. As could be hoped, adding additional exogenous instruments does not seem to significantly alter the conclusions.
I also conducted a more general test of the validity of using age thresholds as an identification strategy estimating the model using single year ages as instruments. The idea is that if the age threshold identification strategy is truly effective, the instrument coefficients in the first stage retirement equation should follow the same pattern as in the reduced form health change equations. If this is the case it suggests that the only way that the age thresholds are influencing health change is through their effect on retirement. Overall, the patterns of coefficients match closely for the models with consistently significant retirement effects and clearly exogenous instruments, suggesting that the age threshold approach is a valid identification strategy.
Once again I am not able to report the control variable coefficients due to the number of dependent variable/instrument combinations tested. Results are available upon request. For each health change model the control variable coefficients were very consistent across the various instrument combinations used showing that the particular instrument set did not change the basic results of the model.
I also estimated models restricting the sample to having worked at least ten and thirty years. For men neither of these restrictions affected the results. For women the ten year results were the same, while the thirty year results stayed positive but retirement lost significance in a few models, likely due to the sample being cut almost in half. I also estimated the model including those individuals who died between the first and second waves of the study and were thus excluded from the original sample, hypothesizing that if these dead individuals were all retired the baseline model might be missing the sickest retirees. Including these deaths as retirees coded as having bad health changes predictably lowered the magnitude of the coefficients, but by a very small amount. Even this worst case scenario did nothing to alter the overall pattern of results.
The annual hour estimates were subjected to the same first stage analyses, instrument exogeneity tests, and reduced form exercises as the retirement specifications with virtually identical results, supporting the validity of the estimates.
References
Baker M, Stabile M, Deri C (2001) What do self-reported, objective, measures of health measure? NBER Working Paper 8419 (August 2001)
Bound J, Schoenbaum M, Stinebrickner TR, Waidmann T (1999) The dynamic effects of health on the labor force transitions of older workers. Labour Econ 6:179–202 (June 1999)
Charles KK (2002) Is retirement depressing?: Labor force inactivity and psychological well-being in later life. NBER Working Paper 9033 (July 2002)
Coile CC (2003) Retirement incentives and couples’ retirement decisions. NBER Working Paper 9496 (February 2003)
Coile CC (2004) Health shocks and couples’ labor supply decisions. NBER Working Paper 10810 (October 2004)
Coile CC, Gruber J (2000) Social security and retirement. NBER Working Paper 7830 (August 2000)
Currie J, Madrian BC (1999) Health, health insurance and the labor market. In: Ashenfelter O, Card D (eds) Handbook of labor economics, vol 3, part 3. Elsevier B.V., Oxford, pp 3309–3415
Dave D, Rashad I, Spasojevic J (2006) The effects of retirement on physical and mental health outcomes. NBER Working Paper 12123 (March 2006)
Dwyer DS, Mitchell OS (1999) Health problems as determinants of retirement: are self-rated measures endogenous? J Health Econ 18:173–193 (April 1999)
Ekerdt DJ, Bosse R, LoCastro JS (1983) Claims that retirement improves health. J Gerontol 38:231–236 (March 1983)
Ettner SL (1996) New evidence on the relationship between income and health. J Health Econ 15:67–85 (February 1996)
Ferraro KF (1980) Self-ratings of health among the old and the old–old. J Health Soc Behav 21:377–383 (December 1980)
Gustman AL, Steinmeier TL (2000) Retirement outcomes in the health and retirement study. Social Secur Bull 63:57–70 (September 2000)
Gustman AL, Steinmeier TL (2002) The Social Security Early Entitlement Age in a Structural Model of Retirement and Wealth. NBER Working Paper 9183 (September 2002)
Gustman AL, Steinmeier TL (2005) Imperfect knowledge of social security and pensions. Ind Relat 44:373–397 (April 2005)
Gustman AL, Mitchell OS, Steinmeier TL (1995) Retirement in the health and retirement study. J Hum Resour 30:s57–s83 (Special Issue 1995)
Hurd MD, McGarry K (1997) The Predictive Validity of Subjective Probabilities of Survival. NBER Working Paper 6193 (September 1997)
Idler EL, Benyamini Y (1997) Self-rated health and mortality: a review of twenty-seven community studies. J Health Soc Behav 38:21–37 (March 1997)
Jewell RT (1992) The effect of retirement on health as reported in the retirement history survey. Ph.D. dissertation. Santa Barbara, CA: University of California—Santa Barbara, 1992
Kerkhofs M, Lindeboom M (1997) Age related health dynamics and changes in labour market status. Health Econ 6:407–423 (July 1997)
Kerkhofs M, Lindeboom M, Theeuwes J (1999) Retirement, financial incentives and health. Labour Econ 6:203–227 (June 1999)
Kiecolt-Glaser JD, Newton TL (2001) Marriage and health: his and hers. Psychol Bull 127:472–503 (July 2001)
Lillard LA, Panis CWA (1996) Marital status and mortality: the role of health. Demography 33:313–327 (August 1996)
Maddox GL, Douglas EB (1973) Self-assessment of health: a longitudinal study of elderly subjects. J Health Soc Behav 14:87–93 (March 1973)
Mossey J, Shapiro E (1982) Self-rated health: a predictor of mortality among the elderly. Am J Public Health 800–808 (August 1982)
Neuman KD (2004) The health effects of retirement: a theoretical and empirical investigation. Ph.D. dissertation. Notre Dame, IN: University of Notre Dame, 2004
Quinn J (1996) The role of bridge jobs in the retirement patterns of older Americans in the 1990s. Paper presented at the International Association of Research in Income and Wealth annual research conference
Ruhm CJ (1995) Secular changes in the work and retirement patterns of older men. J Hum Resour 30:362–385 (Spring 1995)
Staiger D, Stock JH (1997) Instrumental variables regression with weak instruments. Econometrica 65:557–586 (May 1997)
Starr-McCluer M, Sundén A (1999) Workers’ knowledge of their pension coverage: a reevaluation. Federal Reserve Board of Governors Report. Federal Reserve Board of Governors, Washington, DC, 1999
Wallace RB, Herzog AR (1995) Overview of the health measures in the health and retirement study. J Hum Resour 30:s84–s107 (Special Issue 1995)
Acknowledgement
I would like to thank Teresa Ghilarducci, Kevin Lang, and an anonymous referee for their insightful comments on earlier versions of the paper.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Neuman, K. Quit Your Job and Get Healthier? The Effect of Retirement on Health. J Labor Res 29, 177–201 (2008). https://doi.org/10.1007/s12122-007-9036-8
Published:
Issue Date:
DOI: https://doi.org/10.1007/s12122-007-9036-8