Comparing the performance of health care systems: an alternative approach.

I. Introduction

The health care system in the United States has been diagnosed as a moribund institution by a large number of media commentators and health policy analysts. Continually rising health care costs and inadequate health insurance coverage for approximately thirty-seven million people are cited as two of the most visible symptoms supporting their diagnosis. Most of these individuals see little chance for the U.S. health care system to recover given its present design. Full recovery, they believe, requires the successful transplant of a health care system from another country. The health care systems in Europe and Canada, where government is assigned a much larger role than in the United States, are argued to offer universal health care coverage and simultaneously contain health care costs.

Simple comparative statistics appear to support the view that the performance of the U.S. health care system could be improved through a new health care system design. Infant mortality in the United States ranked twentieth among twenty-four member countries of the Organization for Economic Cooperation and Development (OECD) in 1988. Yet, health care expenditures in the United States comprised 11.8 percent of gross domestic product in 1989, while the comparable average for all OECD countries was only 7.4 percent |28~. These statistics portray the U.S. health care system as being deficient and, therefore, unable to offer quality medical services at a reasonable price.

Given the limited stock of information that exists on this issue, however, it is ill-advised to place full blame for the seemingly dismal performance on the U.S. health care system. The simple international comparisons fail to take into account that differences in performance are not solely due to variations in health care systems. While the design of a health care system, including the financing scheme, reimbursement method and organization of production, is indeed important because it influences how various economic, technological and demographic characteristics of a country are transformed into health outcomes, the performance of the health care system also depends on the magnitude of these national characteristics as well. A proper analysis would try to determine how the health care system itself specifically influences the performance of health care markets. An analysis of that kind could provide useful information concerning any change in performance that might result from redesigning the U.S. health care system.

Despite the value of the information, only a very few studies |15; 13; 27~ have formally compared the performance of the U.S. health care system to the systems existing in a number of other OECD countries. For the most part, however, these studies naively, but understandably, assume that the multifaceted nature of a health care system can be reduced to a single variable; the health care expenditure share of government. But clearly, the diverse institutional and regulatory features of a health care system cannot be captured fully by a simple measure of that kind.

Because of this major shortcoming in the literature, we adopt an entirely different approach to compare the underlying performance of the health care system in the United States with other OECD countries. An approach of this kind has been successfully used to examine gender discrimination in labor markets |19~, rent control laws in housing markets |17~ and structural gaps between market and nonmarket economies |10~. First, we use a large panel data set of twelve OECD countries (other than the United States) to estimate an infant mortality and a health care expenditure regression equation. The regression analysis establishes how various socio-demographic and environmental determinants influence the performance of the health care-sector in the OECD countries.

Next, we use the estimated regression equations to generate predicted values for infant mortality and health expenditures in the United States. A comparison of the predicted and actual values allows us to identify any "residual" performance differences attributable to health care system and immeasurable "nonsystem" factors such as lifestyle and climate not already captured in the regression equations. Finally, we assess whether the residual performance difference arises because of health care system or immeasurable "nonsystem" factors.

II. Empirical Specification

Sample, Data and Methodology.

Following previous studies comparing the performance of health care systems, we assume that the infant mortality rate and per capita health care spending capture important differences in health outcomes across countries. Infant mortality data are more complete over time than any other potential measure of health status. Infant mortality is a reasonable health indicator "because it is generally accepted that, where infant mortality rates are high, health levels in all segments of the population are likely to be low |8, 63~."

Of course, international comparisons are not exempt from data measurement problems |23; 16~. It should be kept in mind, however, that data measurement problems have not deterred health policy analysts or the popular press from drawing substantive policy conclusions about the performance of health care systems based solely and simply on the relative rankings of various countries in an infant mortality or health care spending table. This is despite the fact that a host of factors, in addition to the health care system, such as lifestyle patterns and environmental conditions, affect health outcomes. While we are unable to seriously overcome the data measurement problems, the methodology employed in this paper represents a substantial improvement upon the earlier work--data measurement problems or not. This study often a technique for making more systematic and controlled comparisons of health outcomes in the United States with other countries. Nevertheless, we caution the reader that the results of this study are conditioned on the reliability of the data just like the other discussions on this topic in the popular press and academic journals.

A panel data set of OECD countries is used in the empirical analysis. To be included in the sample, each OECD country must have at least eight years of continuous data for all of the necessary variables over the period 1960 to 1987. We insist on a consistent time-series for each country in the panel to ensure that important time-series features of the model are properly captured. With this restriction, we obtain 183 observations for a sample of twelve (non-United States) OECD countries. The countries (and number of observations) included are Austria (20), Belgium (15), Canada (19), Denmark (10), Finland (26), France (22), Greece (9), Italy (11), Netherlands (12), Spain (8), Sweden (20) and United Kingdom (11).(1) Most data come from Health Care Systems in Transition which was published by the OECD in 1990, OECD National Accounts (various years) and OECD Labor Force Statistics (various years).(2)

The infant mortality and health care expenditure equations are estimated with multiple regression analysis for the twelve non-U.S. OECD countries in the following general form:

|Y.sub.it~ = |Beta~|X.sub.i,t~ + ||Mu~.sub.it~, (1)

where X represents a vector of measurable economic, environmental, lifestyle and demographic variables, i = 1, 2, . . ., 12 non U.S. countries and t = 1, 2, . . ., n for the n years of observations for each country. The error terms ||Mu~.sub.it~, it are assumed to be independent and identically distributed across the i countries.

The vector of parameters, |Beta~, reflects how values of X are transformed into a final value for Y. This transformation process depends, in part, on the health care system and also on any nonsystem factors influencing health outcome that we are unable to measure and include in the regression equation. The more important features of a health care system include the mode of production (private or public), financing method (taxes or insurance premiums) and reimbursement scheme (fixed payment or fee for service, single payer versus multipayer system) |29~. In addition, physician referral practices (indirect or direct access to specialists) and level of decision-making (centralized versus decentralized) help to shape the overall design of the health care system. Most European countries and Canada provide universal health insurance coverage and rely on regional governments and global budgets to guide resource allocation in the health care sector |13~.

In contrast, infant mortality and health care expenditures in the United States may depend upon an entirely different transformation process given the greater reliance on the market mechanism for allocating health care resources, or:

|Y.sub.jt~ = |Alpha~|X.sub.jt~ + |v.sub.jt~ (2)

where |Alpha~ may not be equal to |Beta~. We can test whether |Alpha~ is different from |Beta~ by substituting the appropriate vector of explanatory variables for the United States, |X.sub.jt~, in the place |X.sub.it~ to obtain predicted values for infant mortality and health care expenditures, |Mathematical Expression Omitted~. These predicted values identify what the infant mortality rate and health care expenditures would be given the economic, environmental, demographic and technological characteristics of the United States, but assuming the "structure" of the other OECD countries. It is important to realize at this point that structure, in this context, refers to both health care system and any immeasurable nonsystem factors such as lifestyle and diet.(3) If |Mathematical Expression Omitted~, it is taken as evidence of structural differences between the United States and the other OECD countries. Finally, if the predicted values for infant mortality and per capita health spending in the U.S. are different from the actual values, we use published information about various health care systems to assess whether the performance differential is because of uniquely common features of the health care system in the various OECD countries or some other unobservable nonsystem factors.

Specification of the Infant Mortality Equation

To guide the selection of factors influencing infant mortality, we rely on health production theory |2~. Following previous work on international comparisons |15; 3; 27~, we specify infant mortality as a function of the quantity of medical services, education, lifestyle and environmental variables. More specifically, the infant mortality equation is specified in the following logarithmic form:

ln |M.sub.it~ = |a.sub.1~ ln |PHYS.sub.it~ + |a.sub.2~ ln |GDP.sub.it~ + |a.sub.3~ ln |DENS.sub.it~ + |a.sub.4~ ln |ED.sub.it~ + |a.sub.5~ ln |FLFPR.sub.it~ + |a.sub.6~ ln |ALC.sub.it~ + |a.sub.7~ ln |TOB.sub.it~ + |a.sub.8~ ln |TIME.sub.it~ + |Sigma~|A.sub.i~|CNTRY.sub.i~ (3)

where:

M = infant mortality rate (deaths as a percent of live births)

PHYS = number of physicians per capita (measure of medical services)

GDP = real gross domestic product per capita (measure of income or socioeconomic environment).

DENS = population density (environmental variable)

ED = real education expenditures per capita (reflects the technology of health production in the home)

FLFPR = female labor force participation rate (home production variable)

ALC = per capita real expenditures on alcoholic beverages (lifestyle variable)

TOB = per capita real expenditures on tobacco products (lifestyle variable)

TIME = time trend (technology variable)

|CNTRY.sub.i~ = country dummy variable (country-specific nonsystem and health care system variable).

Drawing upon health economic theory and the results of previous empirical studies, the expected relation between most of the independent variables and infant mortality can be inferred. According to health production theory, increased medical services, as reflected by the number of physicians per person, should lead to lower infant mortality as long as the marginal productivity of medical services is greater than zero. Real gross domestic product per capita (GDP) controls for differences in nutrition and the home environment, among other factors |6~.(4) Thus, higher levels of real GDP should lead to lower infant mortality.

The relation between the degree of urbanization or density and infant mortality depends on the net result of two conflicting impacts. Urbanization leads to lower infant mortality if consumers are in closer proximity to medical care providers in urban areas. On the other hand, pollution and other environmental problems impose greater health hazards in urban environments which results in higher infant mortality |15~. It is argued that the efficiency of home-produced health care can be captured by the level of education and that greater efficiency in home production leads to better health status |9~. Thus, according to this theory, a negative coefficient estimate is expected on education. Real per capita educational expenditures are used as a proxy for the level of education.

The female labor force participation rate may also result in two opposing effects |27~. In short, a higher female labor force participation rate may mean less home production of health and a higher infant mortality rate, or, increased income in two wage earner families and a lower infant mortality rate. While the female labor force participation rate of married women is preferred to test these particular hypotheses, data of these kind are unavailable so the aggregate figure is used for each country.

The next two variables, per capita real expenditures on alcohol and tobacco, control for adverse lifestyles across countries |3~. These two variables should directly impact infant mortality to the extent they capture maternal tobacco and alcohol consumption or mirror general lifestyles in the country. A positive coefficient estimate is expected on both of these lifestyle variables.

The time trend variable is specified because panel data are used. Mortality rates may decline over time with advances in knowledge and technology |6~. Finally, using a fixed effects framework, dummy variables, taking on a one or zero value, for each country are also specified in the multiple regression equation to control for any unique aspects of the health care system across the twelve countries and other national differences, including immeasurable lifestyle variables (e.g., diet and exercise), medical technology and climate.

Specification of the Health Spending Equation

A reduced form equation is also estimated to explain differences in health care expenditures per capita, HEXP, across the different countries and over time. Gross domestic product per capita, degree of urbanization, age structure of the population, and life style variables are typically argued to influence health care expenditures on a national level. The exact form of the estimation equation is:

ln |HEXP.sub.it~ = |b.sub.1~ ln |GDP.sub.it~ + |b.sub.2~ ln |DENS.sub.it~ + |b.sub.3~ ln |ED.sub.it~ + |b.sub.4~ ln |FLFPR.sub.it~ + |b.sub.5~ ln |TOB.sub.it~ + |b.sub.6~ ln |ALC.sub.it~ + |b.sub.7~ ln |OLD.sub.it~ + |b.sub.8~ ln |YOUNG.sub.it~ + |b.sub.9~ ln |TIME.sub.it~ + |Sigma~ |B.sub.i~|CNTRY.sub.i~. (4)

where all variables are described above except OLD and YOUNG which represent the percentage of population over sixty-five and under fifteen, respectively.

The first variable, GDP, has been found to be the single most important determinant of health spending differences across countries. Simply put, since medical services represent a normal good, health care expenditures increase with income. In fact a large number of studies have confirmed that the aggregate income elasticity of demand for health care services is greater than zero |14; 18; 15; 24; 1; 27; 7; 12~. This result is consistently obtained even though the various studies use different samples of OECD countries and time periods to conduct their investigations.

Density is expected to increase health care expenditures for two reasons. One, an increased demand for medical services results from the greater likelihood of contagious diseases and industrial pollution in urban areas. Two, the lower time costs of travel imply that consumers are willing to pay a higher monetary price for medical services. The net relation between the level of education and health care costs depends on two opposing impacts. On the one hand, the opportunity cost of sickness is greater for educated individuals. Given the greater opportunity cost of being sick, more health care services are demanded in the marketplace by educated individuals. Hence, health care costs increase with a more educated population. On the other hand, if efficient home health care is provided in more educated households, lower health care costs result in the marketplace.

The female labor force participation rate should directly affect health care spending since less home-production of health takes place and greater household income results when a greater percentage of married women enter the job market. The next two variables, real spending per capita on alcohol and tobacco, are specified to capture the impact of an adverse lifestyle on health care costs. Therefore, the coefficient estimates on these two variables are expected to be positive. The variables reflecting the age structure of the population, percent older than sixty-five and percent less than 16, should both be associated with increased health expenditures given that the young and elderly are typically the main consumers of health care services |14~.

Finally, the time trend variable controls for the effects of changing technology and tastes over time. As mentioned above, the country dummies are designed to capture unique aspects of the health care systems and other national differences. As a result, we form no a priori expectations.

III. Regression Results and U.S. Predictions

Regression Results

Ordinary least squares estimation of equations (3) and (4) finds evidence of autocorrelation so each equation is estimated with the Beach-McKinnon maximum likelihood procedure adjusted for estimation of pooled cross-section and time-series data. The results of the maximum likelihood estimation are shown in Table I. Columns 2 and 3 show the regression results for equations (3) and (4), respectively. The coefficient estimate and corresponding t-statistic (in parentheses) are shown opposite each explanatory variable.

Focusing first on the infant mortality regression equation in column 2, note that approximately 86 percent of the variation in infant mortality is explained by the right hand side variables. The coefficient estimates on five of the eight continuous variables in the infant mortality equation, including the number of physicians, GDP per capita, alcohol and tobacco consumption and the time trend, possess the correct sign and are statistically significant. In addition, seven of the twelve country dummy variables have parameter estimates that are significantly different from zero. Density, education and the female labor force participation rate are found to be unrelated to infant mortality, however.

Turning attention to the estimated health care expenditure equation in column three of Table I, we note that the independent variables explain 90 percent of the variation in health care spending. GDP per capita, population density, the alcohol and tobacco consumption rate, the time trend and country dummies provide most of the explanation. Moreover, the parameter estimates are statistically different from zero on five of the country dummy variables. While the coefficient estimates on GDP per capita, tobacco consumption and the time trend possess their expected signs, the parameter estimates on density and alcohol consumption have the wrong sign.(5) Concluding our discussion, we point out that education, the female labor force participation rate and age structure of the population are all statistically unrelated to health care spending.

Table I. Estimated Infant Mortality and Health Spending Equations
(t-Statistics in Parentheses)

                                                          HEALTH
                                                         SPENDING
                                             INFANT         PER
                                           MORTALITY      CAPITA

PHYSICIANS PER CAPITA                     -0.302(**)
                                          (2.58)

GDP PER CAPITA                            -0.386(**)      0.710(**)
                                          (3.99)         (7.92)

DENSITY                                   -0.867         -0.824(*)
                                          (1.40)         (1.66)

EDUCATION EXPENDITURES                     0.007          0.002
                                          (1.31)          (.427)

FEMALE LABOR FORCE PARTICIPATION RATE     -0.086         -0.048
                                          (0.633)        (0.387)

ALCOHOL CONSUMPTION                        0.099(*)      -0.115(**)
                                          (1.63)         (2.27)

TOBACCO CONSUMPTION                        0.145(**)      0.123(**)
                                          (2.84)         (2.43)

PERCENT ELDERLY                                          -0.560
                                                         (0.425)

PERCENT YOUNG                                             0.305
                                                         (0.511)

TIME TREND                                -0.145(**)      0.480(**)
                                          (2.03)         (8.69)

AUSTRIA                                    0.737(*)       0.510(*)
                                          (2.32)         (1.99)

BELGIUM                                    1.778          1.370
                                          (1.59)         (1.55)

CANADA                                    -2.669         -2.42
                                          (1.37)         (1.55)

DENMARK                                    0.448          0.606
                                          (0.923)        (1.59)

FINLAND                                   -1.630(*)      -1.162(*)
                                          (1.95)         (1.73)

FRANCE                                     0.624(*)       0.583(*)
                                          (1.70)         (1.99)

GREECE                                     0.858(**)     -0.513(**)
                                          (3.96)         (2.92)

ITALY                                      1.313(*)       0.930
                                          (1.71)         (1.48)

NETHERLANDS                                1.538          1.789(*)
                                          (1.21)         (1.75)

SPAIN                                      0.574(**)     -0.132
                                          (2.79)         (0.79)

SWEDEN                                    -1.297(*)      -0.469
                                          (1.92)         (0.85)

UNITED KINGDOM                             1.252          0.905
                                          (1.44)         (1.28)

Adjusted |R.sup.2~                          .864           .904

D.W.                                       1.95           1.84

RHO                                         .81            .76
                                           (.05)          (.05)

** significant at the one percent level or better

* significant at the ten percent level

Predicting the U.S. Infant Mortality Rate and Health Care Expenditures

All in all, the regression results explain a sizeable amount of the variation in infant mortality and health care expenditures across the non-U.S. OECD countries. In addition, the more important variables possess the correct signs and are statistically significant. We now use the estimated regression equations in Table I, to generate predicted values for infant mortality and health expenditures in the United States. This is accomplished by substituting the vector of characteristics for the United States, |X.sub.jt~, in the place of |X.sub.it~. We compare the difference between predicted and actual values for each year over the period 1973 to 1987. Yearly comparisons are made because, for example, either the financing or delivery of health care may have changed in the various countries during the fourteen year period.

Table II reports the yearly comparison between predicted and actual infant mortality rates for the United States given the structure, both health care system and immeasurable nonsystem structure, of the typical OECD country.(6) A comparison for the entire period 1973 to 1987 is also shown at the bottom of the table.(7) The predicted mortality rate is consistently higher than the actual mortality rate for each individual year. While no one yearly difference is statistically different from zero, the joint test for the prediction of all years finds that, on average, predicted infant mortality rates over the entire period are significantly greater than the actual infant mortality rates.

The implication is that the U.S. infant mortality would have averaged approximately 17.2 rather than 12.8 deaths per 1,000 live births over the period 1973 to 1987 if the United States possessed the health care system and unobservable "nonsystem" structure of the typical OECD country. Clearly, the OECD countries in the sample have a wide-range of nonsystem differences whereas they all have a health care system involving a much greater government role than in the United States. Since we are comparing the United States to the typical OECD country, nonsystem differences should average out among the OECD countries. This leads to the inevitable conclusion that the differences found between actual and predicted U.S. values should be dominated by the consistent difference in the health care systems between the United States and other OECD countries. Consequently, it is very likely that the design of the health care system in the typical OECD country accounts for the high predicted U.S. infant mortality rate.

Table II. Actual and Predicted Infant Mortality Rates for the U.S. (Infant
Deaths per 1,000 Live Births)

Time Period     Actual      Predicted     t-test

1973             17.7          24.1       -1.29
1974             16.7          23.1       -1.35
1975             16.1          22.1       -1.32
1976             15.2          21.0       -1.37
1977             14.1          20.0       -1.48
1978             13.8          18.7       -1.28
1979             13.1          17.9       -1.32
1980             12.6          17.1       -1.31
1981             11.9          16.3       -1.35
1982             11.2          15.7       -1.47
1983             10.9          14.8       -1.35
1984             10.7          14.1       -1.21
1985             10.6          13.5       -1.07
1986             10.4          13.0       -0.99
1987             10.0          12.4       -0.96
(1973-1987)      12.8          17.2       -4.95

Table III lists actual and predicted health care expenditure per capita for the same periods as above, assuming the United States adopts the structure of the typical OECD country. Again, the predicted value is consistently higher than the actual value. For most of the individual years, the predicted amount is greater than the actual amount of health care spending at the ten percent level of significance or better. For the entire period, predicted health care expenditure exceeds the actual amount of health care spending by a highly significant margin. Following the logic above, the results indicate that health care expenditure per capita would be considerably higher if the United States designed its health care system similar to the one in the typical OECD country.

Since the health care system in some particular OECD countries may lead to better performance than the others, we also allowed for country specific effects, and generated predicted values for the United States based on the slope parameter estimates and each individual country dummy variable (even if it was not statistically different from zero). Since any resulting discrepancy between predicted and actual values may be due to nonsystem or health care system effects, we examined whether any pattern in |Mathematical Expression Omitted~ across the different OECD countries could be explained by uniquely common elements of their health care systems. For example, if we found that the structure of two particular countries produced better performance for the United States and these two countries shared some unique health care system characteristics not found in the other countries that generate worse performance for the United States, we were willing to accept the evidence as suggesting that the health care system accounts for the performance differential rather than nonsystem factors.

Overall, however, this experiment provided further evidence that health care system differences could not account for the comparative performance of the United States. Specifically, the empirical findings showed that the United States would have a significantly lower infant mortality rate if it possessed a structure more like Canada, Finland or Sweden. In contrast, infant mortality would be higher if the United States was structured similar to Austria, Belgium, Denmark, France, Greece, Italy, Netherlands, Spain and United Kingdom. A similar pattern emerged for health care expenditures. Health care expenditures would be lower if the United States was structured more like Canada, Finland, Greece or Sweden. On the other hand, health care expenditures would have been significantly higher if the United States possessed the structure that exists in any of the other eight countries.(8)

Table III. Actual and Predicted Health Expenditures for the U.S. (Spending per
Capita in Constant Dollars)

Time Period    Actual($)        Predicted($)       t-test

1973              881             1271             -1.90
1974              911             1281             -1.77
1975              939             1282             -1.62
1976              978             1344             -1.65
1977             1002             1381             -1.69
1978             1025             1430             -1.74
1979             1057             1454             -1.68
1980             1089             1447             -1.51
1981             1122             1477             -1.47
1982             1146             1454             -1.28
1983             1170             1480             -1.30
1984             1179             1567             -1.56
1985             1167             1600             -1.75
1986             1197             1630             -1.72
1987             1229             1670             -1.71
(1973-1987)      1067             1446             -6.29

For a number of reasons, however, it does not appear that the superior performance of Canada, Finland and Sweden can be attributed to the design of the health care system. First, while all three of these countries have a compulsory health insurance system financed by taxation and entrust health care decision-making to nonfederal governments, the health care systems in most of the European countries also possess these two features |13~. As a result, if these two features of the health care system matter significantly, the health care systems in the other countries should also render superior performance for the United States. But this is not the case.

Second, the Canadian health care system has a different production arrangement and reimbursement scheme than Finland and Sweden |25~. Health care services are primarily produced by the government in the two Scandinavian countries. Physicians are typically paid on a salary basis. On the other hand, health care services are privately produced in Canada where physicians are paid on a fee for service basis. Fee schedules are determined by negotiation with physician representatives (their version of the American Medical Association) and provincial governments. Given these two differences, it seems unlikely that the superior performance can be attributed to uniquely common elements of the health care system.

Third, since the health care system in Denmark (for example) is very similar to Finland and Sweden, one would suspect that the Denmark country specific effects would also generate low predicted values for the United States if the country dummy variable is truly capturing health care system differences. But, according to the results, the infant mortality rate and health care expenditures would be higher if the United States was structured similar to Denmark. So obviously, this is not the case.

If differences in health outcomes cannot be explained by the design of the health care system, then what accounts for the superior performance of Canada, Finland and Sweden? Fuchs argues in a related context, that any variation in infant mortality within and across countries is most likely because of "differences in the demand for life and the ability to produce life, which are unrelated to income |6, 199~ ." Fundamental to these demand and production differences is the underlying structure of tastes (lifestyle) and preferences (especially time preference). Following this logic, we suspect that it is tastes and preferences, rather than the health care system, that simultaneously accounts for the higher relative mortality rates and health care expenditures in the United States. For example, Eberstadt argues that it is parental life styles, rather than income or access to health care providers, that account for infant survival chances |4~.

IV. Conclusion

This analysis offers an empirical methodology to compare the relative performance of the U.S. health care system to the systems in the other OECD countries and examines how a redesigning of the U.S. health care system might affect infant mortality and health care spending. The findings indicate that both infant mortality and health care expenditures would be significantly higher if the United States possessed a health care system like the one in the typical OECD country. Further analysis, however, reveals that the structure (both health care system and unobservable nonsystem structure) of some particular OECD countries are associated with favorable health outcomes. In particular, the results suggest that the infant mortality rate and health care expenditures would both be lower if the United States adopted the structure of Canada, Finland or Sweden. We are unable to link this superior performance to uniquely common elements of the health care system, however.

Consequently, this study finds little to gain from redesigning the U.S. health care system along the lines of the European and Canadian health care systems. While the results fail to validate the claims of the popular press and many analysts |13; 11~, they should in a general sense, come with little surprise. Given various fundamental differences, the European and Canadian health care systems should not be expected to offer the solution to health care problems in the United States. As Uwe Reinhardt notes:

I have also learned after some twenty-five years in the United States that there are two things Americans just cannot seem to do, for masons only an anthropologist can understand. First they cannot make a railroad run on time; second, they cannot legislate, let alone operate, a lean, streamlined, publicly financed human services system. They demonstrably cannot do it in education; they demonstrably cannot do it in jurisprudence; and they probably could not do it in health care either |26, 175~.

Because individualism is allowed to flourish and there is a general distrust of government |5~, the United States continues to rely more on the market mechanism for allocating health care resources. On the other hand, the European countries and Canada only reluctantly rely upon the market system. They instead trust government to implement global budgets as a way of containing health care costs and guiding resources in the health care sector. Simple statistics suggest that health care costs and infant mortality are lower as a result of the dominant role of government in the health care sector of European countries and Canada. In fact, many health care policy analysts believe that increased government involvement can produce similar results in the United States. Given our findings, however, there is reason to question whether increased government intervention can solve the problems that exist in the mixed-market health care system in the U.S.

1. Given the relatively small number of observations, we also performed the various tests without Greece and Spain in the sample. The results below remain generally the same, however.

2. A detailed list of data sources will be provided by the authors on request.

3. If all nonsystem factors influencing health outcomes could be measured and specified in the regression equations, the residual difference would entirely reflect the performance difference attributable to the health care system.

4. To obtain real dollar figures for the various monetary variables (GDP, ED, ALC and TOB), the nominal domestic values are divided by the GDP deflator and converted to U.S. dollars by using the purchasing power parity exchange rates |20~.

5. Almost all studies find an aggregate income elasticity estimate greater than one for health care services. It should be pointed out, however, that our estimate of .710 represents a short-run income elasticity given the time series nature of the estimation procedure.

6. The t-statistic for each comparison between actual and predicted values represents the test-statistic for the null hypothesis of no difference. The test-statistic is based on the standard error of the forecast.

7. The t -statistic for the comparison of the entire period is the test-statistic for the joint hypothesis that the difference between all actual and predicted values is zero |10~.

8. These results will be provided by the authors on request.

References

1. Aaron, Henry J. Serious and Unstable Condition. Washington, D.C.: The Brookings Institution, 1991.

2. Auster, Richard, Irving Leveson, and Deborah Sarachek, "The Production of Health: An Exploratory Study." Journal of Human Resources, 4, 1969, 411-36.

3. Cochrane, A. L., A. S. St. Leger and F. Moore, "Health Service 'Input' and Mortality 'Output' in Developed Countries." Journal of Epidemiology and Community Health, 32, 1978, 200-205.

4. Eberstadt, Nicholas, "America's Infant-Mortality Puzzle." Public Interest, 105, 1991, 30-47.

5. Fuchs, Victor R., "National Health Insurance Revisited." Health Affairs, Winter 1991, 7-17.

6. -----. "Some Economic Aspects of Mortality in Developed Countries," in The Health Economy, edited by V. R. Fuchs. Cambridge, Mass.: Harvard University Press, 1986, pp. 181-99.

7. Gerdtham, U-G. and Bengt Jonsson, "Conversion Factor Instability in International Comparisons of Health Care Expenditure." Journal of Health Economics, 10, 1991, 227-34.

8. Goldman, Fred and Michael Grossman, "The Impact of Public Health Policy: The Case of Community Health Centers." Eastern Economic Journal, Jan./March 1988, 63-72.

9. Grossman, Michael, "On the Concept of Health Capital and the Demand for Health." Journal of Political Economy, March/April 1972, 223-55.

10. Grubaugh, Stephen G., Andrew J. Stollar, and Rodney G. Thompson, "Socialist Structural Gaps: A Simultaneous Inference Analysis." The Review of Economic and Statistics, November 1989, 693-98.

11. Himmelstein, David U. and Steffie Woolhandler, "Cost Without Benefit." New England Journal of Medicine, 314, 1986, 441-45.

12. Hitiris, Theo and John Posnett, "The Determinants and Effects of Health Expenditure in Developed Countries." Journal of Health Economics, 11, 1992, 173-81.

13. Jonsson, Bengt, "What Can Americans Learn from Europeans," in Health Care Systems in Transition. Paris: Organization for Economic Cooperation and Development, 1990, pp. 87-101.

14. Kleiman, Ephraim. "The Determinants of National Outlay on Health," in The Economics of Health and Medical Care, edited by Mark Perlman. International Economic Association, 1974, pp. 67-81.

15. Leu, Robert E. "The Public-Private Mix and International Health Care Costs," in Public and Private Health Services, edited by A. J. Culyer and Bengt Jonsson, Basil Blackwell, 1986, pp. 41-63.

16. Liu, Korbin, Mailyn Moon, Magarette Sulvetta, and Juhi Chawla, "International Infant Mortality Rankings: A Look Behind the Numbers." Health Care Financing Review, Summer 1992, 105-18.

17. Marks, Denton, "The Effect of Rent Control on the Price of Rental Housing: A Hedonic Approach." Land Economics February 1984, 81-94.

18. Newhouse, Joseph, "Medical Care Expenditures." Journal of Human Resources, Winter 1977, 115-25.

19. Oaxaca, Ronald, "Male-Female Wage Differentials in Urban Labor Markets." International Economic Review, October 1973, 693-709.

20. Organization for Economic Cooperation and Development. Health Care Systems in Transition, Paris, 1990.

21. -----. National Accounts, Paris, various years.

22. -----. Labor Force Statistics, Paris, various years.

23. Parkin, David, "Comparing Health Service Efficiency Across Countries." Oxford Review of Economic Policy, 5, 1989, 75-88.

24. -----, Alistair McGuire, and Brian Yule, "Aggregate Health Care Expenditures and National Income: Is Health Care a Luxury Good?" Journal of Health Economics, 6, 1987, 109-27.

25. Raffel, Marshall W. Comparative Health Systems. University Park, Penn.: The Pennsylvania University Press, 1984.

26. Reinhardt, Uwe, "Health Care Woes of American Business: Reinhardt Responds." Health Affairs, Spring 1990, 174-77.

27. Santerre, Rexford E., Stephen G. Grubaugh, and Andrew J. Stollar, "Government Intervention in Health Care Markets and Health Care Outcomes: Some International Evidence." Cato Journal, Spring/Summer 1991, 1-12.

28. Schieber, George J., Jean-Pierre Poullier, and Leslie M. Greenwald, "Health Care Systems in Twenty-Four Countries." Health Affairs, Fall 1991, 22-38.

29. Stahl, Ingemar. "Sweden," in Advances in Health Economics and Health Services Research, Supplement 1: Comparative Health Systems, edited by Richard M. Scheffler and Jean-Jacques Rosa. Greenwich, Conn.: JAI Press, 1990, pp. 197-210.