RETIREMENT RESEARCH April 2012, Number 12-8 WHY DO MORE PEOPLE DIE DURING ECONOMIC EXPANSIONS? By Ann Huff Stevens, Douglas L. Miller, Marianne Page, and Mateusz Filipski* Introduction Consider this seeming paradox: when economic The first section of this brief discusses the prior times are good, deaths in the United States increase. research on the relationship between declining un- The reasons for this economic impact on mortality employment and rising death rates and tests it using are not well understood, but the negative health ef- new sources of data. The second section explains a fects of over-work, stress, and work-related behavior process of elimination that isolates additional deaths are often cited as culprits. However, this explanation among the elderly as the main reason for rising death is not completely persuasive, because other evidence rates during good economic times. The conclusion is shows that losing a job when the economy sours that an expanding economy generates a greater scar- causes individuals’ health to deteriorate.1 If that were city of front-line caregivers in nursing homes, which the case, it would seem that, during good economic may cause more deaths among the elderly. times, mortality would decline. These conflicting theories about the effect of individuals’ work on their own health suggest that the mechanisms driving the The Unemployment-Mortality unemployment-mortality link are more complex. This brief, adapted from a recent paper, departs Link from earlier studies suggesting individuals’ own Research conducted a decade ago first documented work behavior causes the unemployment-mortality the core relationship: the death rate rises during link and summarizes an empirical scavenger hunt to periods of low unemployment rates, and it falls when find an alternative explanation.2 A key initial finding the unemployment rate goes up.3 Economists and is that most of the additional deaths due to a falling health researchers have tried to explain this finding by unemployment rate occur among the elderly, par- looking at changes in the behavior of individuals that ticularly elderly women. Since the elderly are usually would affect their health. During boom times, when retired, factors unrelated to their work are most likely more people are employed, job-related stress may to explain rising mortality. And since the elderly use increase obesity and smoking. High employment and medical services more as they age, the health care long hours on the job also limit individuals’ ability labor force is a logical place to search for clues. to find time for diet and exercise, causing health to * The authors are all with the University of California, Davis. Ann Huff Stevens is an economics professor and director of the Center for Poverty Research. Douglas L. Miller is an assistant professor of economics. Marianne Page is a professor of economics. Mateusz Filipski is a postdoctoral scholar in the agricultural and resource economics department. This brief is adapted from a longer working paper published by the National Bureau of Economic Research, Stevens et al. (2011). 2 Center for Retirement Research deteriorate. Economists explain this effect, which care facilities? And, third, could the impact of falling seems intuitive, by saying that individuals reallocate unemployment rates somehow affect the quality of their time to accommodate their jobs – and away from health services for vulnerable populations? leisure and healthy living practices.4 But other factors may be involved that are unrelat- Elderly Women Prove Key ed to individuals’ own work behaviors. One example is that motor vehicle accidents increase when unem- The search begins by testing alternatives to the tra- ployment is low, likely the result of more economic ditional hypothesis that higher employment erodes activity and driving during economic booms. Such health and increases mortality. New regression accidents, however, are relatively rare and so cannot analyses again use the unemployment rate as the key come close to fully accounting for the overall cyclical- explanatory variable, but they test whether these rate ity of deaths. Perhaps, instead, broader job-market changes affect deaths differently for different gender effects can better explain why deaths increase when and age groups. The findings provided the first clue unemployment falls. Before searching for the un- that the effect of individuals’ own employment status derlying cause, it is necessary to test the basic link on their health was not driving mortality changes. between the death rate and the unemployment rate First, men and women were analyzed separately. that was established in earlier research. A 1-percentage-point decline in the overall unemploy- The research study underlying this brief uses a ment rate increased male mortality by 0.25 percent. more complete set of data to test the unemployment- But the effect on women was stronger: 0.40 percent; mortality relationship. It also analyzes a longer time it was equally strong for women over age 65. Elderly period than prior research – 1978 through 2006 – to women are typically retired and have little attachment include recent economic booms and increases in lon- to the labor force, indicating that mechanisms apart gevity.5 The dependent variable is each state’s mortal- from their work explain why deaths rise when unem- ity rate in a given year, using a new and improved ployment declines. measure contained in the Vital Statistics data at the A second set of analyses tested changes in mortal- National Centers for Disease Control and Prevention.6 ity among narrowly defined age groups consisting The main independent variable is each state’s unem- of one- and five-year age increments. Deaths among ployment rate in a given year, using data from the U.S. young adults proved most sensitive: a one-point Bureau of Labor Statistics.7 The rich Vital Statistics decline in the unemployment rate sharply drove up dataset includes death rates by age, gender, and cause their death rates. But while their mortality is sensi- of death that allow a more detailed exploration of how tive to business cycles, aggregate deaths among young changes in unemployment affect death rates. adults were not large enough to substantially increase After first replicating the results of prior research, the analysis then tried variations on the basic equa- tion, changing the time period and the mortality Figure 1. Age Composition of Additional Deaths measure and testing the equation using regressions Caused by Decline in Unemployment Rate, 2006 both unweighted and weighted by state-year popula- tion.8 The results of this exercise confirmed that the core relationship holds: a falling unemployment rate Under 65 leads to a higher mortality rate.9   Finding the Cause of the Unemployment-Mortality Link Having confirmed the unemployment-mortality link, the next step is to search for alternative explana- Age 65 and tions for its cause. Three related questions may help older provide an answer. First, is the rise in the mortality rate concentrated among a particular population, for Source: Stevens et al. (2011). example, by gender or age? Second, can we learn anything about where the deaths occur, namely health Issue in Brief 3 mortality in the overall population. Working-age men Nursing Homes and Mortality and women accounted for just 9 percent of the 6,700 additional deaths in 2006 caused by a 1-point decline Past research has shown that the ability of health care in the unemployment rate. facilities to hire staff is particularly sensitive to the Instead, the elderly dominated here: Deaths economy. For example, employment levels in the among people ages 65 and older accounted for 75 health-care sector decline during economic expan- percent of the 6,700 additional deaths (see Figure 1 sions as low-paid, low-skill health workers find better on the previous page). Notably, women over 65 alone jobs elsewhere. This migration worsens perennial accounted for 55 percent of the additional deaths. shortages of direct caregivers, such as nursing aides These findings confirm that factors other than indi- and home health workers.11 Also, the elderly heav- viduals’ own work behavior influence overall mortal- ily utilize hospitals, where labor shortages also occur ity patterns. during strong job markets.12 Perhaps a connection Given this finding, it is not surprising that mea- can be found among health care workers, employ- suring the impact of an age-specific unemployment ment levels in the economy overall, and deaths rate on mortality within that same age group did not among the elderly. generate significant effects. However, testing the The first step is to pinpoint the primary locations relationship across different age groups in the sample of elderly deaths using Vital Statistics data to distin- turned up another important clue: changes in death guish people who died in nursing homes – where rates for older people were primarily driven by em- women typically out-number men – from those who ployment changes among younger individuals. died in other locations.13 The analysis again used This clue led to an examination of whether chang- deaths as the dependent variable, but broke them ing employment patterns in health-care facilities, down by deaths in nursing homes compared with which are heavily used by the elderly, might explain deaths at other locations. The effect of a 1-point the connection. Elderly women typically outlive their decline in the unemployment rate on nursing home husbands and are more likely to reside in nursing deaths was large – a 0.56 percent increase – and statis- homes or other care facilities at the end of their tically significant (see Figure 2). The impact of the lives.10 Putting all of this evidence together, the unemployment rate on deaths at locations other than health-care labor force, and perhaps nursing-home nursing homes was not statistically significant.14 staffing, emerged as places to look for why more of A second piece of information might shed ad- the elderly die when unemployment is falling. ditional light on the importance of nursing-home Figure 2. Impact of 1-Percentage Point Decline in Unemployment Rate on Deaths, by Location, 1983-2002 Deaths at all places 0.004 Statistically significant Non-nursing home deaths -0.005 Not statistically significant Nursing home deaths 0.056 -0.02 0.00 0.02 0.04 0.06 -0.02 0.00 0.02 0.04 0.06 Source: Stevens et al. (2011). 4 Center for Retirement Research deaths on overall mortality: the share of each state’s Medicare data provide employment levels in virtu- population living in nursing homes. If changes in ally all U.S. hospitals and nursing homes.15 They also nursing-home staffing can cause higher mortality, include categories of workers in these facilities: physi- then states with larger nursing home populations cians, nurses (registered nurses and licensed practical would see more deaths in an improving economy nurses), and certified aides. and job market. The U.S. Census does not report on In the next set of regression analyses, nursing- nursing home residents per se but provides a good home employment – not the mortality rate – became proxy for each state: the number of individuals over the dependent variable; the unemployment rate re- age 65 living in group homes or facilities. mained the main independent variable. The analysis The analysis found that, indeed, mortality does revealed that a 1-percentage-point decline in the un- increase more in states with larger shares of their employment rate causes more than a 3-percent drop populations in nursing homes. This finding indicates in overall full-time employment at nursing homes. that nursing homes may be key to answering the cen- Looking at specific occupations, a 1-point drop in the tral question: why does falling unemployment cause unemployment rate caused employment to decline death rates to rise? by 3 percent for aides and by more than 2 percent for nurses (see Figure 3).16 Nursing Home Workforce In tight labor markets, it is easier for nurses and aides in nursing homes to migrate to better jobs and the Economy elsewhere, and the impact could be large when the economy is growing and jobs are plentiful. For So, what is occurring inside nursing homes that example, amplifying the results of the first analysis, might cause more elderly deaths? One possible expla- a 4-point drop in the overall unemployment rate – nation is that the quality of care declines in a strong which approximates the change that accompanies an economy when demand for workers grows through- economic boom to bust cycle – could cause nursing- out the economy. Using nursing-home staffing levels home staff to decline by about 12 percent. as a proxy for quality of care, the analysis explored Taken together, these findings point to a connec- two potential effects of a tight labor market that would tion between falling unemployment rates and more bear directly on elder care: 1) total employment in all deaths. Specifically, tight labor markets constrain the nursing homes; and 2) employment levels for various already scarce number of workers available for hire occupations found in nursing homes. by nursing homes, where older people in vulnerable Figure 3. Effect of 1-Percentage-Point Decline in Unemployment Rate on Employment in Nursing Homes, by Occupation, 1983-2002 Certified aides -0.034 Nurses -0.026 Statistically significant Not statistically significant Physicians -0.013 Total employment -0.038 -0.02 0.00 0.02 0.04 0.06 -0.04 -0.02 0.00 Note: These results are from a linear regression model that includes year and state fixed effects and state-specific trends. All models are weighted by number of beds. While the impact on physician employment was not statistically significant, doctors make up an extremely small percentage of nursing home employment. Source: Stevens et al. (2011). Issue in Brief 5 health require direct and constant care. A greater Endnotes scarcity of these front-line caregivers may have a direct impact on the elderly, causing them to die in 1 Sullivan and von Wachter (2009). greater numbers when the unemployment rate is declining. 2 Stevens et al. (2011). 3 Ruhm (2000, 2003, 2005a, 2005b). Conclusion 4 Job-related health problems may also result from This research reaffirms a robust link between mortal- hazardous working conditions. In addition, people ity and the unemployment rate, but offers an alterna- migrate geographically to new jobs when the econo- tive explanation for the cause. The analysis found my is strong, which may increase crowding or lead to that rising mortality during good economic times is other lifestyle changes that could raise mortality. See largely driven by additional deaths among people over Ruhm (2000). age 65 in nursing homes, particularly women. To find the cause, the analysis explored the impact of a 5 Ruhm analyzed 1972-1991 using unemployment falling unemployment rate on nursing home staff- rates from the Bureau of Labor Statistics, mortality ing and found significant declines in the number of data from Vital Statistics publications, and control nurses and health aides caring for the elderly during variables from the U.S. Census. economic booms. This finding suggests that such shortages may be an important focus of future efforts 6 The new Vital Statistics measure of the mortality to improve the quality of health care. rate provides micro-record “multiple cause of death” files with a denominator from population counts collected by the National Cancer Institute’s Surveil- lance Epidemiology and End Results (Cancer-SEER) program. Cancer-SEER estimates are based on an algorithm that incorporates information from Vital Statistics, IRS migration files, and Social Security data; this method makes them more accurate than population estimates interpolated from between years in U.S. Census surveys. 7 Monthly data from the Current Population Survey (CPS) were pooled to construct employment and unemployment rates by state and demographic group; estimates for states prior to 1978 are unavailable in the CPS. Demographic controls include education, race, and the fraction of the population who are less than 5 years old and greater than 65 years old. A vec- tor of year-specific fixed effects captures national time effects and a vector of state-specific indicator variables controls for time-invariant state characteristics. State- specific time trends are also included. 8 Using population weights is appropriate to estimate the degree to which economic conditions contribute to overall fluctuations in U.S. mortality. On the other hand, the unweighted regressions address the impact on a typical state’s mortality rate. 6 Center for Retirement Research 9 For a more detailed discussion of the methodology References and results, see Stevens et al. (2011). Goodman, William C. 2006. “Employment in Hos- 10 Murtaugh, Kember, and Spillman (1990). pitals: Unconventional Patterns Over Time.” Monthly Labor Review 129 (6):3-14. 11 Yamada (2002) cites a New York State study that suggests that between 70 percent and 90 percent of Murtaugh, Christopher, Peter Kember, and Brenda home health care agencies and nursing homes are Spillman. 1990. “The Risk of Nursing-Home Use short on direct-care staff. in Later Life.” Medical Care October 28 (10): 952- 962. 12 Goodman (2006). Ruhm, Christopher J. 2000. “Are Recessions Good for 13 Vital Statistics data show whether deaths occurred Your Health?” Quarterly Journal of Economics May in hospitals, nursing homes, personal residences or 115(2): 617-650. other locations; non-nursing home locations were grouped together into “other locations.” This analy- Ruhm, Christopher J. 2003. “Good Times Make You sis, for the time period 1983-2006, found that the Sick.” Journal of Health Economics 22(4): 637-658. coefficient for nursing-home deaths was large but was not statistically significant. Dropping the years after Ruhm, Christopher J. 2005a. “Mortality Increases 2002, when the overall effect of unemployment on During Economic Upturns.” International Journal mortality weakens and when a change in data coding of Epidemiology 34(6): 1206-1211. occurred, substantially increased the magnitude of the coefficient for the full sample. Ruhm, Christopher J. 2005b. “Healthy Living in Hard Times.” Journal of Health Economics 24: 341-363. 14 Repeating the analysis using U.S. Census data, which supplies another measure of the place of death, Stevens, Ann Huff, Douglas L. Miller, Marianne confirmed a relationship between the unemployment Page, Mateusz Filipski. 2011. “The Best of Times, rate and nursing home deaths: a 1-percentage-point the Worst of Times: Understanding Pro-cyclical decline in the unemployment rate caused mortality to Mortality.” Working Paper 17657. Cambridge, MA: increase by 0.03-0.06 percent. National Bureau of Economic Research. 15 Medicare data, taken from the Online Survey Sullivan, Daniel and Till von Wachter. 2009. “Job Certification and Reporting Database (OSCAR), cover Displacement and Mortality: An Analysis Using employment in 97 percent of U.S. hospital facilities, Administrative Data.” Quarterly Journal of Economics which include skilled nursing facilities. August 2009: 1265-1306. 16 Using Medicare data, a similar exercise for hos- Yamada, Yoshiko. 2002. “Profile of Home Care Aides, pitals found little evidence that changes in hospital Nursing Home Aides, and Hospital Aides: Histori- staffing levels were related to the business cycle. A cal Changes and Data Recommendations.” The separate analysis, using CPS data, was also used to Gerontologist 42(2): 199-206. test the connection between the unemployment rate and health care staffing patterns. This analysis in- cluded all types of employers, not just nursing homes, and found that employment levels of health and nurs- ing aides followed a similar pattern indicated by the Medicare data presented in Figure 3. RETIREMENT RESEARCH About the Center Affiliated Institutions The Center for Retirement Research at Boston The Brookings Institution College was established in 1998 through a grant Massachusetts Institute of Technology from the Social Security Administration. The Syracuse University Center’s mission is to produce first-class research Urban Institute and educational tools and forge a strong link between the academic community and decision-makers in the public and private sectors around an issue of Contact Information Center for Retirement Research critical importance to the nation’s future. To achieve Boston College this mission, the Center sponsors a wide variety of Hovey House research projects, transmits new findings to a broad 140 Commonwealth Avenue audience, trains new scholars, and broadens access to Chestnut Hill, MA 02467-3808 valuable data sources. Since its inception, the Center Phone: (617) 552-1762 has established a reputation as an authoritative source Fax: (617) 552-0191 of information on all major aspects of the retirement E-mail: crr@bc.edu income debate. Website: http://crr.bc.edu The Center for Retirement Research thanks AARP, InvescoSM, LPL Financial, Mercer, MetLife, MFS Investment Management, National Reverse Mortgage Lenders Association, Prudential Financial, Putnam Investments, State Street, TIAA-CREF Institute, T. Rowe Price, and USAA for support of this project. © 2012, by Trustees of Boston College, Center for Retire- The research reported herein was supported by the Center’s ment Research. All rights reserved. Short sections of text, Partnership Program. The findings and conclusions ex- not to exceed two paragraphs, may be quoted without ex- pressed are solely those of the authors and do not represent plicit permission provided that the authors are identified and the views or policy of the partners or the Center for Retire- full credit, including copyright notice, is given to Trustees of ment Research at Boston College. Boston College, Center for Retirement Research.