Are Recessions Good for Everyone’s Health? The Association Between Mortality and Income by Race ∗ Matias Fontenla

Fidel Gonzalez

University of New Mexico

Sam Houston State University

[email protected]

fidel [email protected] Troy Quast

Sam Houston State University [email protected]

2009-01

Abstract While studies have found that economic expansions (contractions) are associated with increases (decreases) in mortality for the general U.S. population, little is known regarding this relationship for vulnerable populations. Given the relatively low income levels of these groups, the procyclical relationship found for the general population may not be present or may be reversed for vulnerable populations. This paper analyzes county-level data for the U.S. from 1999 though 2005 to investigate the relationship between the business cycle and mortality for white, latinos, and blacks. We find that the procyclical relationship is present for both latinos and blacks. Further, the relationship is three to four times as strong as the relationship for whites. Our results suggest that during economic expansions resources should be directed especially towards vulnerable ∗

The authors would like to thank the Robert Wood Johnson Foundation Center for Health Policy

at the University of New Mexico for their financial support and Luis Robles for his excellent research assistance.

1

populations to address the negative health effects that these groups tend to suffer during these periods.

Keywords: business cycles; mortality rates; vulnerable populations JEL: C33, E32, I1.

1

Introduction

An important aspect of public health policy is the effect of the economic cycle on general and specific causes of mortality. Recent literature for the U.S. (Rhum 2000, Tapia Granados 2005a) , Germany (Neumayer 2004), a group of five European countries (McAvinchey 1988), Spain (Tapia Granados 2005b), advanced OECD countries (Gerdtham and Ruhm 2006), and Mexico (Gonzalez and Quast 2009) have shown that short-run increases in economic activity are associated with short-run increases in total mortality. However, the aforementioned studies for the U.S. have focused their analysis on mortality rates for the general population leaving some unanswered questions about the possible effect of the business cycle on vulnerable populations such as latino and blacks. Since racial groups are impacted differently by the business cycle, it seems reasonable to suggest that results for the overall population do not necessarily apply in the same magnitude to each racial group. For example, Freeman (1973), Bradbury (2000) among others have shown that during recessions blacks tends to suffer higher increases in unemployment than whites. In addition, the access to health care by racial group tends to differ and can change unevenly during the business cycle. The purpose of this paper is to investigate whether the procyclical nature of mortality found in the general population of the U.S. found by Ruhm (2000) and Granados Tapia (2005a) is also present in the major racial groups of the U.S. In contrast to the previous literature, we do not use state-level data but rather we employ county-level data. The use of county-level data provides us with more disaggregation of employment status and mortality. This allows us to isolate with more precision the effect of the business cycle on vulnerable populations.1 1

The use of more disaggregated data comes at the cost of using a smaller set of control variables

2

We consider two possible channels through which changes in income affect mortality: behavior and health care utilization. The behavioral channel suggests that short-run increases (decreases) in income lead to greater (less) risk-taking by individuals and thus increased (decreased) mortality rates. The health care utilization channel implies that increases (decreases) in income increase (decrease) the amount of health care that individuals can afford and thus leads to lower (higher) mortality. Therefore, procyclical results would be consistent with the behavioral channel hypothesis whereas countercyclical results would favor the health care utilization channel hypothesis. We construct a large data set at the county level that covers the period 1999 through 2005 and contains overall mortality rates, a measure of the business cycle (unemployment rate), health care infrastructure (physicians per 1000), welfare assistance (federal payments to individuals) and crime (violent crime per 1000). We use a panel level regression with county and year fixed effects. We analyze all counties and subsets of the counties based on racial majority (white, black, and latino). We obtain two main results. First, we find that the procyclical relationship between income and mortality found for the general population also holds for counties where the majority of the population is black or latino. Second, in the black and latino majority counties, the effect of the business cycle on mortality is three to four times stronger than in the white majority counties. These results suggest that the procyclical relationship is stronger for blacks and latinos than for the white population. Regarding the behavioral and healthcare utilization channels hypotheses, our findings are consistent with the behavioral channel for blacks and latinos. Thus, public policy should address the pronounced negative effects that these vulnerable populations suffer during periods during economic expansions. The paper is organized as follows. The next section describes the data and the empirical specification. The third section presents and discusses the results. The final section concludes. compared to Ruhm (2000) and Tapia Granados (2005a) because some of the data are unavailable at the county level during our sample period.

3

2

Data and empirical specification

To perform the analysis, three types of variables are necessary: the mortality rate, the unemployment rate, and control variables. The sample period is 1999 through 2005. Given the inclusion of year and county fixed effects, all of the variables must vary by county and year. The mortality data come from the Compressed Mortality database maintained by the U.S. Centers for Disease Control. The mortality rate is the number of deaths per 100,000 residents. It is adjusted by age, to account for differences in mortality rates due to the age distribution of its residents. The unemployment rate comes from the Bureau of Labor Statistics. The control variables are included to account for other factors that vary over time within a county that may affect the mortality rate. For instance, mortality may be influenced by the level of medical infrastructure. As a proxy for this, the number of physicians per capita is included as a control variable.2 The number of violent crimes per capita is also included as a control variable. In addition to accounting for the potential for physical harm from criminal activity, this variable can also be thought of as a rough proxy for the education level of a county. Finally, government transfers may influence mortality, as payments to individuals will affect their income level. Two types of payments are included: those for retirement and disability and those related to other reasons. Table 1 summarizes the sample data for the full sample. There are 21,407 observations covering 3089 counties. The overall, age-adjusted mortality rate is 893, with a minimum of 231 and a maximum of 3331. Over the sample period for the counties included the mean unemployment rate is 5.3%. The coefficients are estimated via ordinary least squares. The natural log of the mortality rate is used as the dependent variable and the observations are weighted by 2

As the number of physicians is available for 2003 forward, the 1999 through 2002 values are linearly extrapolated. As a robustness test, the regressions were estimated without this variable and the results were largely unchanged.

4

the square root of the state population. The main estimating equation is: ln(morti,t ) =β0 + β1 unemploy ratei,t + β2 phys percapi,t + β3 crimes percapi,t

(1)

+ β4 pmts retdisi,t + β5 pmts otheri,t + γt + ηi + i,t where i indexes the county and t indexes the year. The γt terms are the year fixed effects, the ηi are the county year effects, and i,t is the error term. The error terms are clustered at the county level to account for the possibility of correlated disturbances within each state.

3

Results

3.1

Overview

As noted above, the sample is subset into those counties that are white majority, Latino majority, or black majority for each year in the 1999 - 2005 period. In this sample, 2799 counties are white majority, 48 are Latino majority, and 89 are black majority. Summary statistics for each subset are reported in Table 2. Given the preponderance of white majority counties, it is unsurprising that the white majority statistics are very similar to those for all counties. However, interesting differences are present across racial majorities. While the Latino mortality rate is less than the white mortality rate, the black mortality rate is greater. For both black and latino majority counties, the unemployment rate is greater than in white majority counties. The same pattern holds true for violent crimes, while black majority counties have a greater number of physicians per capita than white and latino majority counties. The regression results are provided in Table 3. As a robustness check, for each sample subset two sets of regression results are provided: those that include the control variables and those that do not. (The odd-numbered columns correspond to where the control variables are excluded, while the even-numbered columns correspond to where they are included.) For each sample subset, the coefficient on the unemployment variable is roughly the same regardless of whether the control variables are included. Going forward, the discussion will focus on the results where the control variables are

5

included. For the full sample, the coefficient on the unemployment rate is -.0015 and is significant at the 10% level. The interpretation of this coefficient is that a one percentage point increase in the unemployment rate is associated with an average decrease in the mortality rate of 0.15%. At the sample average of the mortality rate, this translates to decrease of 1.4 deaths per 100,000 residents. The coefficients on the control variables generally have the expected sign. One would expect that an increase in the number of physicians and a decrease in the number of violent crimes would lead to a decrease in mortality. The positive sign on federal payments for retirement and disability may reflect the health status of residents. The results by racial majority indicate that the procyclical relationship is stronger for the three racial majorities than for the full sample. Further, the relationship is of a larger magnitude in latino and black majority counties than in white majority counties. Specifically, the coefficient for latino majority counties is three times as large as that for white majority counties, while the coefficient is more that twice as large in black majority counties. Based on the sample averages, a one percentage point increase in the unemployment rate in latino majority counties is associated with a decrease of 5.0 deaths per 100,000. In black majority counties, the average effect is a decrease of 5.8 deaths. Two notes of caution are warranted regarding these results. First, the statistical significance of the unemployment rate in the black majority counties does not reach the 10% level (the p-value is 0.113). However, given the relatively small sample size of the latino and black majority counties, it is unsurprising that the statistical significance is not especially strong. Second, most of the control variables are not statistically different from zero. However, this too is unsurprising given the relatively small sample sizes.

3.2

Discussion

The above results potentially tell an interesting story regarding how mortality is affected by the business cycle. Generally, as is the case in white majority counties, the mortality rate in latino and black majority counties is procycical. Further, the 6

relationship is of a considerably larger magnitude than in white majority counties. Given the limited, aggregate data, it is not possible to draw conclusions as to what is driving these results. However, it is helpful to keep in mind Ruhm’s (2000) finding that the procyclical relationship found for the entire population was likely due to an increase in risky behavior. The results here are consistent with the possibility that, given the relatively lower discretionary income of blacks and latinos, increases in income for these groups are devoted to the risky behaviors described in Ruhm (2000).

4

Conclusions

In this paper, we analyze the relationship between the business cycle and mortality for vulnerable populations, specifically blacks and latinos. Previous studies for the U.S. have shown a procyclical relationship for the general population. However, blacks and latinos can have different experiences over the business cycle which may have a differentiated impact on mortality. In contrast to previous studies, we use county-level data which provides greater racial detail. The data set covers the period 1999 to 2005 and contains mortality rate, unemployment, and control variables. We use a panel level regression that controls for county and year fixed effects. Our results suggest that the mortality rates of blacks and latinos have a stronger procyclical relationship with the business cycle than the mortality rate of the white population. This also supports the notion that the channels through which income affect mortality may be behavioral for blacks and latinos. As such, resources to reduce mortality should perhaps be directed towards vulnerable populations during economic expansions. Nevertheless, our results should be taken with caution as more research is needed to confirm our findings. In particular, we use mortality data for the overall population at the county-level. Since we analyze counties with a majority black or latino population, this assumes that the overall mortality for these counties is representative of the mortality rates of the these racial groups. Moreover, we acknowledge that additional control variables (especially education and health care spending) and a longer sample

7

period would strengthen our analysis. Therefore, our study may be extended as more county-level data becomes available.

References [1] Bradbury, K. (2000) Riding Tides in the Labor Market: To What Degree Do Expansions Benefit the Disadvantaged? New England Economic Review, May/June: 3-33. [2] Freeman, R. (1973) Changes in the Labor Market for Black Americans, 1948-1972. Brookings Papers on Economic Activity Vol. 1: 67-131. [3] Gonzalez, F., Quast, T. (2009) Do Changes in Income Affect Mortality? The Case of Mexico, Sam Houston State University, Working Paper XXX. [4] Gerdtham U, Ruhm C (2006) Deaths rise in good economic times: evidence from the OECD. Econ and Human Biology, 4:298-316. [5] McAvinchey I (1988) A comparison of unemployment, income, and mortality interaction for five european countries. Appl Econ, 20:453-471. [6] Neumayer E (2004) Recessions lower (some) mortality rates: evidence from Germany. Soc Science and Medicine, 58: 1037-1047. [7] Ruhm C (2000) Are recessions good for your health? Q J of Econ, 115(2):617-650. [8] Tapia Granados J (2005a) Increasing mortality during the expansions of the US economy, 1900-1996. Intl J of Epidemiology, 34:1194-1202. [9] Tapia Granados J (2005b) Recessions and mortality in Spain, 1980-1997. European J of Pop, 21:393-422.

8

5

Tables

Table 1: Summary statistics - all counties (n = 21, 407) Variable Mean Std. Dev. Min. Max. Age-adjusted mortality rate (per 100000) 893.3 148.3 230.8 3331.2 Unemployment rate 5.3% 2.0% 0.7% 30.6% Number of violent crimes (per 1000) 2.5 2.6 0.0 80.9 Number of physicians (per 1000) 1.5 3.3 0.0 167.5 Federal payments to individuals (retire. & dis.) 2465 741 0 16,279 Federal payments to individuals (other) 1125 448 0 7550 Unless otherwise noted, variables are calculated on a per capita basis. Includes counties for which the mortality rate was not missing, excluded, or deemed unreliable.

9

Table 2: Summary statistics - by racial majority Variable Mean Std. Dev. All counties (n = 21, 407) Age-adjusted mortality rate (per 100000) 893.3 148.3 Unemployment rate 5.3% 2.0% Number of violent crimes (per 1000) 2.5 2.6 Number of physicians (per 1000) 1.5 3.3 Federal payments to individuals (retire. & dis.) 2.465 .741 Federal payments to individuals (other) 1.125 .448 White (not Latino) majority (n = 19, 416) Age-adjusted mortality rate (per 100000) 886.5 140.4 Unemployment rate 5.1% 1.8% Number of violent crimes (per 1000) 2.3 2.3 Number of physicians (per 1000) 1.5 1.6 Federal payments to individuals (retire. & dis.) $2.500 $.736 Federal payments to individuals (other) $1.117 $.437 Latino majority (n = 329) Age-adjusted mortality rate (per 100000) 809.4 130.9 Unemployment rate 8.5% 4.0% Number of violent crimes (per 1000) 3.7 2.5 Number of physicians (per 1000) 1.0 0.9 Federal payments to individuals (retire. & dis.) $1.871 $.561 Federal payments to individuals (other) $1.057 $.353 Black majority (n = 622) Age-adjusted mortality rate (per 100000) 1068.7 135.6 Unemployment rate 7.8% 2.4% Number of violent crimes (per 1000) 4.9 4.3 Number of physicians (per 1000) 3.1 17.4 Federal payments to individuals (retire. & dis.) $2.422 $.677 Federal payments to individuals (other) $1.424 $.471

Min.

Max.

230.8 0.7% 0.0 0.0 0 0

3331.2 30.6% 80.9 167.5 16.279 7.550

230.8 2768.1 0.7% 17.0% 0.0 80.9 0.0 28.2 $0 $16.279 $0 $7.489 362.3 2.8% 0.0 0.0 $.852 $.512

1200.7 30.6% 12.4 4.0 $4.031 $2.555

686.6 2.6% 0.0 0.0 $.770 $.303

1548.0 18.5% 24.7 169.2 $5.597 $3.479

Unless otherwise noted, variables are calculated on a per capita basis. A county is considered as having a racial majority if the percent of the population for that race is greater than 50% for each year in the sample period. Includes counties for which the mortality rate is not missing, excluded, or deemed unreliable. There are 2799 white (not Latino) majority counties. See table X in the appendix for a list of counties and years for which data are not available. There are 48 Latino majority counties. The 2003 data for Culberson county, TX are missing. For Hudspeth county, TX, only the 2002 data are available. There are 89 black majority counties. The 2005 data for Orleans Parish, LA are missing.

10

11

.0579*** (.0102) -.0006 (.0047)

Federal payments retire. & dis.

Federal payments other

-.0027*** (.0007)

(3)

-.0206*** (.0079)

-.0313*** (.0080)

-.0003 (.0005)

-.0043 (.0042)

-.0021*** (.0007)

(4)

(5) -.0040 (.0028)

* significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level

.0029** (.0013)

Number of violent crimes

-.0015* (.0009)

(2)

-.0063 (.0047)

-.0018* (.0011)

(1)

All counties

Number of physicians

Unemployment rate

Explanatory variable

.0219* (.0116)

.1091 (.0724)

.0053 (.0049)

.0317 (.0610)

-.0062** (.0029)

(6)

Table 3: Effect on overall mortality White majority Latino majority Coefficient (Standard error)

-.0050 (.0044)

(7)

.0235 (.0343)

.0384 (.0371)

-.0022*** (.0006)

-.0160 (.0202)

-.0054 (.0034)

(8)

Black majority