The effect of federal government spending on the national unemployment rate Micro or Macro

Am J Public Health. 2004 November; 94(11): 2004–2009.

Abstract

Objectives. We analyzed the relationship between macroeconomic conditions, measured as unemployment rate and social security spending, from 4 social security schemes and total spending due to sickness and disability.

Methods. We obtained aggregated panel data from 13 Organization for Economic Cooperation and Development member countries for 1980–1996. We used regression analysis and fixed effect models to examine spending on sickness benefits, disability pensions, occupational-injury benefits, survivor’s pensions, and total spending.

Results. A decline in unemployment increased sickness benefits spending and reduced disability pension spending. These effects reversed direction after 4 years of unemployment. Inclusion of mortality rate as an additional variable in the analysis did not affect the findings.

Conclusions. Macroeconomic conditions influence some reimbursements from social security schemes but not total spending.

Research into the determinants of social security spending has produced a large body of literature documenting a procyclical pattern in sickness absenteeism.1–3 This pattern is generally believed to result from negative selection of employment during good times (i.e., sicker individuals have an easier time finding jobs when the economy is robust) or increased “shirking” because of reduced fear of job loss.4–6 However, such a pattern is also consistent with deteriorating health when labor markets strengthen.7–9 Hazardous work conditions, the physical exertion of employment, and job-related stress can have negative health effects, especially when job hours are extended during short-lived economic expansions.10,11 Cyclically sensitive sectors, such as construction, also have high accident rates, and some byproducts of economic activity, such as pollution and traffic congestion, clearly present health risks. Gerdtham and Ruhm8 recently used aggregate data for 23 Organization for Economic Cooperation and Development (OECD) countries over the period 1960–1997 to study the relationship between macroeconomic conditions (unemployment) and fatalities. Their main finding, as well as findings from Ruhm and Neumayer, was that there is strong evidence that mortality increases when unemployment is low.7–9

Our principal objective was to analyze the relationship between macroeconomic conditions, as measured by unemployment rate and social security spending due to sickness and disability on the basis of information from 13 member nations of the OECD over the period 1980–1996. We analyzed social security spending on sickness benefits, disability pension, occupational injury benefits, survivor’s pension, and their aggregates. Knowledge about how spending by different social security programs changes through economic business cycles can affect social security policymaking and help to establish new social security rules that effectively address cost containment.

METHODS

Estimation Strategy

We used linear regression to estimate the relationship between labor market conditions and social security spending. Each of four types of social security spending, as well as the sum of this spending, was transformed to natural log before being used in the regression analysis. Using the subscripts j and t to index country and year, respectively, the basic model specification is as follows:

(1)

where y is the natural log of social security spending, E is the unemployment rate, X is a vector of regressors controlling for age and sex distributions in the population and other relevant variables, αt is year-specific fixed effects, ɛ is a Cj is country-specific effects, and disturbance term. The year effects hold constant factors that vary uniformly across countries over time (e.g., the effects of oil price shocks). The country effects account for factors that differ across locations but that are time invariant, such as certain country-specific social security institutions; that is, countries’ social security systems differ in terms of their principles of payments, population coverage, and historical basis of formation. Accordingly, the effect of the labor market conditions is identified by within-country variations in unemployment rates relative to the changes occurring in other nations. Importantly, our estimation procedure automatically controls for cross-country differences in social security systems that remain constant over time.

Most of the models used in this context also include a vector for country-specific linear time trends (C j *T ) to control for factors that vary over time within nations (such as level of education), which implies the following regression equation:

(2)

Finally, in some models, we controlled for per capita income to examine whether macroeconomic fluctuations in social security spending reflect changes in gross domestic product (GDP).

The regressions were estimated by weighted least squares (with observations weighted by the square root of the national population) to account for heteroscedasticity. We also tested whether the results changed when using unweighted data or when allowing for first-order autocorrelation with country-specific autocorrelation coefficients.

Data

We analyzed annual data from 13 OECD member countries (Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Japan, Luxembourg, Sweden, the United Kingdom, and the United States), during the period 1980–1996. The standardized unemployment rate was our primary proxy for labor market conditions. Information on unemployment was obtained from OECD annual statistics.12 Information on social security spending (disability pensions, occupational injury benefits, sickness benefits, and survivor’s pensions), in million national currency units and at current prices, was taken from the OECD Social Expenditure Database 1980–98.13 Disability cash benefits consisted mainly of disability pension benefits. In some cases, other benefits were included that were not separated, because preserving the data comparability between countries and over time was prioritized.

Data from 1990 onward were provided in European System of Integrated Social Protection Statistics methodology format for Belgium, Denmark, France, Greece, Ireland, Sweden, and the United Kingdom.14 The new European System of Integrated Social Protection Statistics methodology format dates from 1990. A match at the level of individual social security programs was attained on the basis of the years for which the format and its predecessor overlapped, which enabled a coherent series to be obtained. For certain programs and aggregate categories, breaks in series were unavoidable. For France, these breaks concerned benefits for occupational injury and disease; for Greece, these concerned benefits for sickness benefits and survivors’ allowances (to a lesser extent); and for the United Kingdom, these concerned benefits for survivor’s pensions. Such unavoidable breaks in series were excluded from the analysis. German expenditure data referred solely to Western Germany for 1990 and earlier; accordingly, data for Germany before 1991 were excluded from the analysis.

Data on country characteristics and per capita GDP came from the World Development Indicators and different OECD statistic series. All the regressions controlled for the proportion of the population that is female (per 100 persons in each country), for labor-force participation among people aged 55–64 years, for the proportion of people living in urban areas among the total population, and for the age of the social security system. As observed earlier, such variables were expected to affect social security spending independently.15,16 Data on control variables, with the exception of labor-force participation among people aged 55–64 years and age of social security system, were obtained from World Development Indicators 2000.17 OECD’s Labour Force Statistics 1979–199918 and the US Department of Health and Human Service’s Social Security Programs Throughout the World—199319 were our data sources for age of the social security system. The data source for GDP per capita, purchasing power parity and total standardized mortality was the OECD,20–22 and the data on consumer price index was collected from a document center in the University of Michigan.23 Country-fixed effects, general time effects, and (usually) country-specific time trends were included. Finally, some models controlled for the natural logarithm of national GDP per capita. These models were adjusted for purchasing power parity and were expressed in 1990 US dollars. To convert national income to US dollars, we used purchasing power parities instead of conventional market exchange rates, because exchange-rate conversions overstate (understate) real income in high- (low-) income countries.24

Descriptive Statistics

Summary statistics, weighted by the total population resident in each country, are shown in Table 1 ▶. Mean, standard deviation (SD), and number of observations (on the basis of which the statistics were calculated) are presented for both dependent and independent variables. Mean total social security spending per capita was US dollars 792.6, with a standard deviation of US dollars 535.4. Per capita sickness benefit was 134.5 US dollars (SD 127.2 US dollars).

TABLE 1—

Variables Employed in the Analysis

Note. PPP = purchasing power parity; GDP = gross domestic product per capita of US$PPP (1990). Data were obtained from 13 OECD member countries during the period 1980–1996. All observations were weighted by country population. Social security spending and gross domestic product were converted to 1990 PPP (US$). Standardized unemployment rate was according to the OECD definition.

Lowest average spending was on occupational injury benefit. Occupational injury benefit per capita was 74.1 US dollars (SD 71.7 US dollars). The highest spending was on disability pension, with an average spending of 324.5 US dollars (SD 258.8 US dollars). Spending on survivor’s pension was 262.3 US dollars (SD 220.6 US dollars). Standard deviations were very high for all kinds of per capita social security spending.

RESULTS

Total Social Security Spending

Table 2 ▶ displays the results of a variety of econometric models that relate the strength of the labor market to total social security spending. Model A shows the results of regressing the natural log of total social security spending on standardized unemployment rate, demographic characteristics of country (gender and age), location effects, year effects, and country-specific linear time trends. The unemployment coefficient, shown in the table, measures the effect of 1-percentage-point change in unemployment on the changes in spending, expressed in percent (difference between antilog of coefficient and 1, multiplied by 100) in unemployment. The results show no statistically significant relationship between total social security spending and unemployment. Note that the results are sensitive to whether year effects or country-specific time trends are controlled for (i.e., before control for these effects, unemployment was positively and significantly related to total social security spending). Model B adds control for per capita GDP, which is found to be insignificant and does not change the results of Model A.

TABLE 2—

Predicted Effects of a 1-Percentage-Point Change in the Unemployment Rate on Total Social Security Spending

Specific Sources of Social Security

Table 3 ▶ summarizes results concerning disbursements from each of the 4 social security schemes separately. In all models, we controlled for demographic variables, years effects and location effects, and country-specific time trends. Model B adds per capita GDP. The results in Model B show that unemployment is significantly associated with sickness benefit and disability benefit and is close to being significant on survivor’s pension. No significant association was found for occupational injury benefit (after controlling for GDP). For Model B, a 1-percentage-point decline in unemployment was associated with a 4% increase in sickness benefits, whereas disability pension and survivor’s pension fell by 2% and 1%, respectively. Thus, different sources of social security appear to be affected differently by changes in labor market conditions, and the separate effects tend to cancel each other out with regard to total spending.

TABLE 3—

Predicted Effects of a 1-Percentage-Point Change in the Unemployment Rate on 4 Social Security Disbursements

The unemployment elasticity of social security spending (i.e., how relative changes in percent of unemployment affect the relative changes in spending) is calculated by using regression coefficients. For sickness benefit, the statistically significant coefficient indicates that a 1-percentage-point fall in unemployment is estimated to raise sickness benefit by 4%. Because the average rate of unemployment is 7.6%, a 1-point decrease corresponds to a decrease of approximately 13%; the unemployment elasticity of sickness benefit is 4/13 or 0.308. It means that a 1% decrease in unemployment increases sickness benefits by 0.308%. The unemployment elasticity of disability pension is much lower (0.153).

Because both social security spending and income are transformed to natural logged values (log–log model), the regression coefficient of GDP directly measures the income elasticity with respect to social security spending. The income effects in Model B are evident in disability pension scheme and survivor’s pension. For all 3 payments, the coefficients for GDP per capita are positive and significant. An increase in GDP per capita by 1% increases occupational injury insurance benefits, disability pension, and survivor’s pension by 0.89%, 0.80%, and 0.65%, respectively.

We also used unweighted data or allowed for autocorrelation of the error term with country-specific first-order-autocorrelation processes to estimate the models (both in Table 2 ▶ and Table 3 ▶). These sensitivity analyses had no important influence on our results (data not presented in the table).

Dynamics

Table 4 ▶ shows findings with regard to adjustment patterns of total social security disbursements and from the 4 social security schemes. Unemployment has different effects on spending over time. A durable 1-percentage-point decline in unemployment is estimated to raise total spending by 0.4%, 0.8%, and 0.3%, after 0, 2, and 4 years, respectively (only significant after 2 years).

TABLE 4—

Predicted Effects of a Sustained 1-Percentage-Point Change in the Unemployment Rate

It appears that total social security spending rises significantly 2 years after a decline in unemployment because spending on disability pension rises whereas all the other social security schemes are unaffected. The accumulated effect of a 1-percentage-point drop in unemployment raises disability pension spending by 1.6% after 2 years. Further, total spending does not rise significantly after 4 years after a drop in unemployment because the effects of unemployment on sickness benefits and disability pensions appear almost to cancel out. For example, the accumulated effect of a 1-percentage-point drop in unemployment increases spending on sickness benefits by 2.1% but reduces spending on disability pensions by 1.6%.

Mortality and Social Security Spending

Ruhm,7 Gerdtham and Ruhm,8 and Neumayer9 concluded that mortality decreases in economic recessions. In our data, we found a similar relationship between mortality and unemployment. Therefore, we controlled for mortality rate to analyze whether social security spendings are affected through variations in mortality. Finally, we tested whether the effect of unemployment on total spending and on disbursements from the 4 social security schemes was sensitive to controlling for total standardized mortality. We found that our results were not sensitive to this change in specification (data not shown).

DISCUSSION

Our results showed that spending on sickness benefits and disability pensions was influenced by unemployment rates. As unemployment rates decrease, spending on sickness benefits increases while spending on disability pensions decreases. We also found that the effects switched direction over time. The results were robust in relation to changes in overall mortality. According to the labor market literature, sickness absenteeism increases immediately in periods of low unemployment.1–3 Our results with regard to sickness benefit were consistent with these findings. These results can be explained by the common phenomenon of negative selection of workers into the labor market during periods of economic expansion. That is, workers with ill health enter the labor market because the labor demand is high in economic expansion. Consequently, economic expansion increases sickness absenteeism. In addition, because of reduced fear of unemployment during periods of low unemployment, shirking may be more common.4–6 However, the reduction in sick leave spending during periods of high unemployment might be explained by 2 further interactions. First, during periods of low unemployment, work-related sickness increases, and thereby sick leave payments increase as well. During periods of low unemployment, overtime and the pace of work increase in some industrial and commercial sectors. During periods of economic expansion, with an increasing demand for labor, groups with temporary disability are attracted to the labor market by disability-adjusted job opportunities. A lengthy boom may entail the risk that these groups are incapable of managing physical and psychological job demands (over a long period of time). Accordingly, they may be expected to resort to taking sick leave or a short-term disability pension (either part time or full time). This phenomenon has been detected in an earlier Swedish study.25

Second, according to Johansson and Palme, the reduction in payments can be explained solely by public policy.26 During recessions, when government financing is negatively affected, restrictions on social security can be implemented, for example, in the form of an increased number of qualifying days for reimbursement or lower compensation levels. Naturally, this reduces overall social security spending. Such an explanation does not contradict our findings.

That the amount paid out in disability pension falls immediately when there is an upturn in the economy (unemployment reduces) may be explained, in a similar manner, by more people with reduced working capacity choosing to return to work. This gives rise to an immediate reduction in social security spending. Flexible regulations can facilitate a return to work among people with permanent functional impairments if they can quickly obtain work where job demands are tailored to such impairment. Just as is the case with people on sick leave, these groups are affected by the length of the economic boom and by whether the relationship between job demands and performance is changed. This may explain the changes in reimbursements reported in Table 4 ▶, namely that spending on disability pension falls at first but then rises after a lag of 2 to 4 years.

The relationship between GDP per capita and social security spending from occupational injury insurance, disability pensions, and survivor’s pensions are significant. Countries with a better economic level have higher spending on these schemes.

Total mortality has been shown to be low during periods of low unemployment.27–31 However, Ruhm,7 Gerdtham and Ruhm,8 and Neumayer9 have demonstrated that overall mortality increases during times of expansion and boom. In our analysis, we detected a similar association between unemployment and overall mortality in the OECD countries as a whole. Our findings concerning the effect of unemployment on disbursements in terms of social security and its components were not affected by whether we controlled for mortality in the estimated models.

When making cross-country comparisons, the effects of variations and changes in social security regulations between and within countries over time must be taken into account. By using the fixed-effect model, the influence of such effects can be reduced.7,8,32 Further, unobserved heterogeneity between countries can be removed from any model by using dummy variables for countries, years, and country-specific time trends. Nevertheless, individual data might provide extensive information about the reasons underlying variations in spending. Aggregated data tend to obscure the scope to observe such variations. For example, individual-level data on mortality as a proxy for a health condition might be replaced by information on utilization of health care or self-reported health.

The concept of work absenteeism, often used by labor-market economists, does not empirically correspond fully to the concept of sickness benefit. For example, the number of qualifying days or payments made by an employer directly to the insured over the first part of a sick-leave period may not be included in spending data produced by social insurance authorities. Conversely, the number of days reported from a questionnaire survey on absence may not be fully comparable with the number of days paid for by the insurance system because of recall error. In addition, absenteeism is influenced by changes in the concept of illness compared with disease in the population over time, ability to cope with psychosocial factors at work (especially during recession), and the changing demographic characteristics of aging populations.33

CONCLUSIONS

Social security spending from 4 social security schemes acted differently during economic business cycles. Both magnitude and direction of response in spending in addition to changes in economic business cycles varied. Knowledge from this study may contribute to policymaking decisions on the basis of prevailing or expected economic cycles.

Macroeconomic conditions influence most social security payments, but total spending is not affected. We recommend further study based on individual data to better understand the relationship between the macroeconomic business cycle and social security spending.

Acknowledgments

This study was supported by the Swedish Social Insurance Board Research Fund (contract 3406/01-UFU), the Swedish Council for Working Life and Social Research (contract 2002–0376) and the Swedish National institute of Public Health.

We also thank 2 anonymous referees for their valuable comments.

Human Participation Protection
The ethical committee at Karolinska Institutet, Stockholm, Sweden, approved this study.

Notes

Peer Reviewed

Contributors
J. Khan wrote the article and analyzed the data. U.-G. Gerdtham planned and conducted the data analysis and contributed to interpretation of the data and writing of the article. B. Jansson conducted the study, selected the relevant social security reimbursements, and contributed to interpretation of the data and writing of the article.

References

1. Leigh JP. The effects of unemployment and business cycle on absenteeism. Journal of Economics and Business. 1985;37:159–70. [Google Scholar]

2. Kaivanto K. An alternative model of procyclical absenteeism. Econ Lett. 1997;54:29–34. [Google Scholar]

3. Arai M, Thoursie PS. Incentives and Selection in Cyclical Absenteeism. FIEF Working Paper Series 167, 2001. Stockholm, Sweden: Stiftelsen Fackföreningens Institut för Ekonomisk Forskning.

4. Calvo GA. Quasi-Walrasian theories of unemployment. Am Econ Rev Proceedings. 1979;9:102–106. [Google Scholar]

5. Salop SC. A model of natural rate of unemployment. Am Econ Rev. 1979;69:117–125. [Google Scholar]

6. Shapiro C, Stiglitz JE. Equilibrium unemployment as a worker discipline device. Am Econ Rev. 1984;74: 433–444. [Google Scholar]

7. Ruhm CJ. Are recessions good for your health? Q J Econ. 2000;115:617–650. [Google Scholar]

8. Gerdtha U-G, Ruhm, CJ. Death rise in good times: Evidence from the OECD. Working Paper 9357. Cambridge, Mass: National Bureau of Economic Research; 2002.

9. Neumayer E. Recessions Lower (Some) Mortality Rates: evidence from Germany [mimeo]. London, England: London School of Economics and Political Science; 2002.

10. Karasek R, Theorell T. Healthy work: stress, productivity, and reconstruction of working life. New York, NY: Basic Books; 1990.

11. Sokejima S, Kagamimori S. Working hours as a risk factor for acute myocardial infarction in Japan: case-control study. BMJ. 1998; 317:775–780. [PMC free article] [PubMed] [Google Scholar]

12. Quarterly Labor Force Statistics. Paris, France: Organization for Economic Cooperation and Development; 2000.

13. OECD Social Expenditure Database 1980–98. Paris, France: Organization for Economic Cooperation and Development; 2001.

14. Hagemejer K. Social protection expenditure and performance reviews methodological note on definitions, classifications and performance indicators. Geneva, Switzerland: International Labor Office, Financial, Actuarial and Statistical Services; 2000. Available at: http://www.ilo.org/public/english/protection/socfas/research/sper/spersmet.pdf. Accessed January 2002.

15. Harold LW. Economic level, ideology and social structure in the welfare state equality: structural and ideological roots of public expenditure. In: Robert E, Deborah M, eds. The Foundations of the Welfare States. Cheltenham, UK: Edward Elgar Publishing Limited; 2000: 15–52.

16. Hicks AM, Swank DH. Politics, institutions and welfare spending in industrialized democracies, 1960–82. American Political Science Review. 1992;86: 658–674. [Google Scholar]

17. United Nations. World Development Indicators. Washington, D.C.: The World Bank; 2000.

18. Labour Force Statistics 1979–1999. Paris, France: OECD; 2000.

19. US Department of Health and Human Services. Social Security Programs Throughout the World—1993. Washington, DC: Social Security Administration Office of Research and Statistics; 1994. SSA Publication No. 13–11805. Research Report 63.

20. OECD National Accounts, 1960–96. Paris, France: OECD; 1998.

21. OECD National Accounts, Volume 1. Paris, France: OECD; 1998.

22. OECD Health Data 2000; Comparative Analysis of 29 Countries. Version 07/15, Paris, France: OECD; 2000.

24. Kravis IB, Heston AW, Summers R. Real GDP per capita for more than one hundred countries. The Economic Journal. 1978;88:215–241. [Google Scholar]

25. Bäckman O. Longitudinal Studies on Work Absence in Sweden [dissertation]. Stockholm, Sweden: Swedish Institute for Social Research; 1998.

26. Johansson P, Palme M. Assessing the Effect of Policy on Worker Absenteeism. Uppsala, Sweden, Institute for Labour Market Policy Evaluation; 2002. Working Paper 2002:13.

27. Gerdtham UG, Johannesson M. Business Cycles and Mortality: Results from Swedish Micro Data [mimeo]. Lund, Sweden: Lund University; 2002. [PubMed]

28. Wagstaff A. Time series analysis of the relationship between unemployment and mortality: a survey of economic critiques and replications of Brenner’s studies. Soc Sci Med. 1985;21:985–996. [PubMed] [Google Scholar]

29. Brenner MH. Mortality and the national economy: a review and the experience of England and Wales, 1936–76. The Lancet. 1979; 2(8142):568–573. [PubMed] [Google Scholar]

30. Gravelle HS, Hutchinson G, Stern J. Mortality and unemployment: a critique of Brenner’s time-series analysis. The Lancet. 1981; 2(8248):675–679. [PubMed] [Google Scholar]

31. Smith R. “Without work all life goes rotten”: Unemployment kills, ruins health and destroys families. BMJ. 1992;305:972. [PMC free article] [PubMed] [Google Scholar]

32. Greene WH. Econometric Analysis, 4th ed. New York, NY: Prentice Hall International, Inc.; 2000.

33. Currer C, Stacey M, eds. Concepts of Health, Illness and Disease: A Comparative Perspective, Leamington Spa, UK: Berg; 2001.

Articles from American Journal of Public Health are provided here courtesy of American Public Health Association