Income inequality, democracy and health: 

 

A global portrait

 

 

Draft version

Not to be cited without the permission of the author.

 

30 March 2002

 

 

Michael E. Shin

University of California, Los Angeles

Department of Geography

1255 Bunche Hall

Box 951524

Los Angeles, California  90025-1524

USA

t.  001.310.825.3727

f.  001.310.206.5976

shinm@geog.ucla.edu

 

 

 

 

 

 

Abstract

 

Throughout the world there exists a notable relationship between income inequality and health.  Political factors and circumstances, which arguably affect the distribution of income, may also be related to population health.  This paper explores the geographic dimensions of these relationships in an attempt to determine whether or not levels of democracy complement, influence or confound the association between income inequality and health. 

 

 

 

 

 

 

This paper was prepared for presentation at the conference, Responding to Globalization: Societies, Groups and Individuals, sponsored by the Globalization and Democracy Research and Training Program, Program on Political and Economic Change, Institute of Behavioral Science, University of Colorado at Boulder, 4-7 April 2002.  The data used in this paper are available via ftp at pacific.ssnet.ucla.edu/pub/tmp/shinm.

Introduction

 

Income inequality is one of the predominant features of global society.  Moreover, income inequalities between and within countries are increasing (Milanovic 1999; Prichitt 1997; Jones 1997).  One of the more perplexing social costs related to income inequality is its impact upon health, or more specifically, the finding that high levels of income inequality are associated with lower life expectancy, and higher infant and childhood mortality rates throughout the world (Rodgers 1979; Waldmann1992).  The relationship between income inequality and health, referred to as the ‘relative income hypothesis’, remains robust even when absolute levels of income are controlled and when making comparisons between and within countries (e.g., Kennedy et al. 1996; Ben-Shlomo et al. 1996).

            One factor that is possibly related to both income inequality and health is democracy.  There is a long tradition of research on the relationship between economic development, inequality and democracy (e.g., Lipset 1959; Muller 1995; Midlarsky 1997), and it is plausible that health is implicated in this relationship.  Determining whether or not and to what extent democracy tempers or fortifies this association between income inequality and health is the primary objective of this research.  A geographic perspective is used to explore and examine the relative income hypothesis at the global scale.  I contend that situating analyses of the relative income hypothesis within such a geographic framework will help to clarify the nature of the linkages between income inequality, democracy and health.

 

 

Theoretical considerations

 

Over the course of the last century, life expectancy and levels of economic prosperity increased dramatically and concurrently around the world.  Increases in life expectancy can, in part, be attributed to the reduction of death rates among the young, or in other words, more and more people are now reaching older ages than ever before.  Though increases in absolute income are undoubtedly linked to health, especially in underdeveloped countries, it is interesting to note that the beneficial returns on health appear to diminish after a per capita income of approximately $5,000 is achieved.  In fact, after this $5,000 threshold is surpassed, the relationship between life expectancy and wealth becomes quite unclear.  For example, life expectancy in the United States, the richest country in the world in terms of per capita income at the turn of the new century, falls short of that for several countries such as Japan and Italy, and infant mortality rates in America are higher than those in the Czech Republic (World Bank 2001).  It may appear that the limits to human longevity have been reached in many developed countries, but the rate of increase in life expectancy (i.e., 2-3 years per decade) remained remarkably stable over the course of the twentieth century (Wilkinson 1994; 1996).

            Country to country variations in life expectancy and infant mortality are frequently understood in terms of the epidemiological transition model (see Omran 1971; 1983; Wilkinson 1994).  This model documents a given society’s pathway to the reduction of death rates, particularly those deaths associated with communicable diseases.  Demographic characteristics of countries that have yet to complete the epidemiological transition include:  high levels of mortality primarily due to infectious diseases, high levels of fertility and a predominantly young population which is especially vulnerable to infectious disease.  Once the detrimental effects of communicable disease can be overcome by a society, mortality and fertility rates gradually decrease and stabilize, the population ages and non-communicable diseases such as heart disease, stroke, obesity and cancer become more common as causes of death (Wilkinson 1996).

            Improvements in agricultural production, transportation and access to health care are all factors that can undoubtedly help a country complete this critical transition.  The transformation of various social and economic practices, political structures and cultural, traditions, such as the active and passive repression of women, children and other groups in many societies is also recognized to be important to population health.  Countries that have successfully completed the epidemiological transition often experience similar patterns and trends, for example, periods of massive rural to urban migration and the emergence of a distinct urban hierarchy (Omran 1971; 1983).  However, as Jones and Moon (1992) and Picheral (1989) point out, the geographic and temporal manifestations of the epidemiological transition are not necessarily identical in different places nor are they consistent over time.

            The epidemiological transition marks an important stage in the process of economic development for a society, and once the transition is completed, increases in absolute income impact health very little (Wilkinson 1994; 1996).  Another striking post-transition health trend is that diseases once more commonly diagnosed in the affluent (e.g., heart disease, obesity, cancer, etc.) become more prevalent in the under-classes and the poor (Marmot et al. 1978; Wing 1988).  Moreover, a significant body of research shows that population health becomes more a function of relative deprivation than of absolute deprivation after the epidemiological transition (Waldmann 1992; Wilkinson 1992).  In other words, increases in absolute income are arguably less important than disparities in income levels insofar as population health is concerned.  Evidence that supports this relationship, which is referred to as ‘the relative income hypothesis’ exists when making:  i) international comparisons (e.g., Rodgers 1979; Wilkinson 1996; LeGrand 1987; van Doorslaer et al. 1997), ii) comparisons between less developed countries (e.g., Rodgers 1979; Flegg 1982), iii) comparisons between developed economies (e.g., Duleep 1995; Wennemo 1993), and iv) it is also present within societies (e.g., Kaplan et al. 1996; Walberg et al. 1998; Merva and Fowles 1993; Kahn et al. 1998; Chiang 1999; Hales et al. 1999).  Though there are multiple scales from which the relative income hypothesis can be examined, this research attempts to clarify the nature of this relationship at the global level.

            Several explanations have been posited with regard to how income inequality affects health, and range from the argument that inequality leads to problematic variations in access to health care, education and social services to the argument that income polarization strains the beneficial components of social capital, which in turn adversely affects population health (e.g., Rodgers 1979; 1996; Berkman 1996; Kawachi et al. 1997).  The relationship between income inequality and health has also been subject to much scrutiny and skepticism.  Concerns about the relative income hypothesis range from methodological issues about data measurement to demands to identify more clearly the pathway between income inequality and health.  Such concerns will be addressed and discussed further as they arise within the scope of this analysis.

One dimension of the relative income hypothesis that has not thoroughly been examined relates to those factors that are associated with and possibly affect the distribution of income within a country.  In other words, does income inequality itself impact population health, or is income inequality a possible surrogate for other political or economic factors?  It is not difficult to envisage that the distribution of income within a society may indeed be a function of political considerations, such as levels of democracy.  Conversely, a substantial body of research exists with respect to the economic determinants of democracy (e.g., Lipset 1959; Muller 1988; 1995).  Economic development arguably facilitates democracy through increases in education and the enlargement of the middle classes (Lipset 1959).  The former arguably promotes socio-political tolerance while the latter helps to temper social conflict by reducing the size of the population that is attracted to anti-democratic ideologies and anti-system parties.  Though strong evidence exists to support the relationship between economic prosperity and democracy in certain geographic regions and over certain periods of time, several exceptions illustrate well the complexity underlying this relationship (Huntington 1991:  59-72; O’Loughlin et al. 1998). 

            The view that democracy alleviates social disparities and income inequalities stems from the argument that political enfranchisement in the form of suffrage empowers those who are on the margins of society (Lipset 1959; Lenski 1966; Muller 1995).  Once granted a political voice, the marginalized and the disadvantaged will attempt to expropriate and redistribute wealth from the advantaged through, for example, support for more radical parties.  The assertion that inequality, and income inequality in particular, is perceived as unjust and elicits a political response is disputed.  Bollen and Jackman (1995) argue that there is little substantive evidence to suggest that income inequality is a suitable proxy for political resentment, and reiterate that no significant relationship exists between income inequality and democracy, and vice versa (Bollen and Jackman 1985; 1995). 

            Paralleling the possible relationship between income inequality and democracy is that between democracy and health.  The logic that underpins the relationship between health and democracy is similar, and may be related, to that discussed above.  If democracy indeed reduces socio-economic inequalities, it may also improve population health.  Within egalitarian and democratic societies, healthcare is arguably more easily obtained and there are more opportunities to improve socio-economic status, which in turn provides greater access to the means by which health can be improved (e.g., food, shelter, medicine, healthcare, etc.).  Furthermore, at the regional and global scales, democracies tend to be pacific with each other, thus reducing the deleterious effects of war (Russett 1993; Ray 1993).   War not only inflates mortality rates through battlefield and collateral deaths in the short term, but war can also have several health consequences over the long term (Ghobarah 2001; Murray 2002).  War creates refugees, destroys vital social and health infrastructures and dismantles local, national and regional economies to the extent that both direct and indirect victims of war (e.g., successive generations) become susceptible to malnourishment, infectious disease and premature death.

Peaceful inter-state relations are typically necessary for mutually beneficial interactions and exchanges such as trade to develop and grow, which subsequently can contribute to and improve economic development and likewise population health.  It is understood, however, that becoming democratic often entails violence and conflict.  Though relatively smooth democratic transitions diminish the likelihood of war, the more irregular the transition to democracy, the greater the chance of warfare (Ward and Gleditsch 1998; Gleditsch and Ward 2000).  It is also possible that very high levels of wealth and democracy create diseconomies of affluence and freedom that may negatively impact population health (Rodgers 1979: 343).  The crisis of expensive, incomplete and spotty health care in the United States, especially for the less privileged, may be related to American “hyperdemocracy” where the profit motive coupled with the freedom of choice, and their preservation, takes precedence over the common good and the provision of universal health care (Heclo 1999).  Notwithstanding the debate surrounding the relation between democracy and income inequality, the possible role of democracy within the framework of the relative income hypothesis merits further investigation.

A causal model of the relationships between income, income inequality, democracy and health is provided in Figure 1. 

 

 

 

 



Figure 1.  Relationships between income, income inequality, democracy and health.

 

It is widely accepted that economic prosperity improves population health, and as discussed previously, a positive relationship is expected between democracy and health, with levels of democracy a function of absolute income.  Income inequality is also believed to be a function of income, but this relationship is believed to take the form of an inverted-U.  According to Kuznets (1955), higher incomes reduce inequalities in developed countries, but in less developed countries, the early stages of economic development can exacerbate inequalities.  The nature of the relationship between income inequality and democracy remains in question (see Muller 1995; Bollen and Jackman 1995), and it is plausible that this association is bi-directional.  Finally, a negative relationship is expected to exist between income inequality and health. 

           

 

Data and methodology

 

A robust statistical relationship has been found to exist between income inequality and a variety of indicators of population health, such as life expectancy, infant mortality rates and childhood mortality rates at the international level (e.g., Rodgers 1979;Waldmann 1992).  Life expectancy, and somewhat paradoxically, mortality rates have proved to be concise and reliable indicators for population health in light of the difficulties in determining, defining and gauging the concept of health (Wilkinson 1996:  55-56).  Within the scope of this analysis, 1999 life expectancy data are used to measure health and are taken from the 2001 World Development Indicators published by the World Bank.  Life expectancy at birth is defined as “the number of years a newborn infant would live if prevailing patterns of mortality were to stay the same throughout its life” (World Bank 2001: 117).

            Income inequality data in the form of the Gini index are also drawn from the 2001 WDIs.  Derived from the Lorenz curve that portrays graphically the cumulative shares of income at successive income intervals, the value of the Gini index falls between zero, which indicates perfect equality, and 100, which indicates perfect inequality (see Sen 1973; Amiel and Cowell 2000).  Though other measures of inequality exist, and are evaluated with respect to the relative income hypothesis elsewhere (e.g., Kawachi and Kennedy 1997), the Gini index is the statistic used most frequently and is strongly correlated with other such measures.  Note that use of the Gini index presents several comparability issues (Deininger and Squire 1996; Milanovic 1999).  Inequality measures such as the Gini index are often based upon household surveys that are difficult to adjust and standardize over time and between countries.  In order to provide the most complete set of inequality measures, the World Bank reports Gini index values based on both income and consumption expenditures, the former tending to capture greater inequalities (World Bank 2001: 73).  Additionally, variations in household size and levels of income preclude strict country-to-country comparisons.  Nevertheless, recent efforts to overcome such comparability issues make these data, at the very least, as accurate and as valid as those used in previous analyses examining the relationship between income inequality and health.

            The measure for democracy is taken from the Polity 98D database (November 2000 version) compiled by Kristian Gleditsch (Gleditsch 2002).  Democracy scores are extracted for 1998 and range from zero to ten.  These data are based upon the comprehensive Polity III database that provides comparable data about the political characteristics of countries since 1800.  The democracy scores used are derived from a combination of measures such as the extent of constraints placed upon the chief executive and the competitiveness of political participation (for details, see Gleditsch and Ward 1997).  Though the use of a single year’s data fails to capture whether or not a polity is experiencing a political transition, change or trend that could possibly affect population health, the scores provide a very useful proxy of democracy within subsequent statistical analyses.  Finally, 1992 GDP per capita data, in constant 1985 U.S. dollars, are extracted from the Penn World Tables, Mark 5.6a (Summers and Heston 1991).  Controlling for the relationship between absolute income and health, which tends to diminish as income increases, will elucidate the possible associations between democracy, income inequality and health.  Though more recent data on per capita GDP are available, 1992 values are used to capture the possible lagged effects of income upon health.

            The complete data set contains a total of 88 countries, and is provided in Appendix A.  Inconsistencies between data sources force the exclusion of several countries, but a sufficiently large and geographically diverse pool of observations remains for examination.

Summary statistics for the four variables of interest are provided in Table 1 and scatterplots and the correlation matrix is provided in Figure 2. 

 

 

 

mean

σ

minimum

maximum

 

life expectancy, 1999

 

      65.93

 

     12.48

 

37

Sierra Leone

 

 

    81

Japan      

 

real GDP/capita, 1992

$4931.32

$4991.60

$274.00

Ethiopia

$17,945.00

United States

 

Gini index

      39.99

      10.71

  19.5

Slovakia

 

      62.9

Sierra Leone

Democracy score, 1998

       6.69

       3.51

  0

9 countries

 

    10

25 countries

 

 

 

 

 

 

Table 1.  Summary statistics for the relative income data set.

 

 

 

 

 

 

 


 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 


lifex99

rgdp92

gini

dem98

lifex99

 1.00

 0.67

-0.42

 0.61

rgdp92

 0.67

 1.00

-0.48

 0.58

gini

-0.42

-0.48

 1.00

-0.21

dem98

 0.61

 0.58

-0.21

 1.00

 

 

Figure 2.  Scatterplot and correlation matrix for the relative income data set.

 

 

 

 

 

 

 

 

All correlations between the variables of interest are statistically significant (i.e., p < 0.05), with the strongest relationship between per capita GDP and life expectancy, and the weakest relationship between democracy and the Gini inequality index.  Note also the curvilinear relationships between income and life expectancy, and income and democracy, with the former more apparent.

Most examinations of the relative income hypothesis utilize ordinary least squares (OLS) regression, and follow the general model specification,

 

y = α + βX + ε                                                                                          (1)

 

where y is life expectancy or some other measure of health, α is the intercept term, X is the matrix of independent variables (i.e., per capita GDP, Gini index, etc.), and ε is the error term is distributed normally and independently (i.e., εi ~ NID(0,σ 2)).  A series of models based upon Rodger’s (1979) original work (i.e., equations 1-4), which is the basis for much research on the relative income hypothesis (see Kawachi et al. 1999), are estimated and serve as benchmarks for comparison to subsequent equations that include the democracy variable (i.e., equations 5-9) in Table 2. 

 

 

           

dependent variable:  life expectancy at birth, 1999

 

1

2

3

4

5

6

7

8

9

 

constant

 

 

75.88

 

-13.05

 

83.32

 

-9.39

 

51.43

 

68.55

 

-8.80

 

75.73

 

-3.89

 


 

 

-17,132

(1,179)

 

 

 

-16,027

(1,188)

 

 

 

 

 

-14,593

(1,263)

 

 

-13,699

(1,248)

 

 

 

ln(gdp)

 

 

 

    9.91 

(0.65)

 

 

    9.69

(0.73)

 

 

 

    9.00

(0.82)

 

 

   8.68

(0.90)

 

Gini index

 

 

 

 

  -0.20

(0.07)

 

   -0.05

(0.07)

 

 

 

 

   -0.19

(0.06)

 

  -0.06

(0.07)

 

democracy

 

 

 

 

 

 

    2.18

(0.30)

 

    0.88

(0.22)

 

    0.44

(0.25)

 

    0.83

(0.21)

 

   0.46

(0.25)

 

adj. R2

 

 

      0.71

 

0.73

 

0.73

 

0.72

 

0.37

 

0.75

 

0.73

 

0.77

 

0.73

 

F-statistic

 

 

211.30

 

232.14

 

119.65

 

115.52

 

51.10

 

131.63

 

120.43

 

98.53

 

80.26

 

Table 2.  OLS estimates of the relative income hypothesis, standard errors in parentheses.

     Estimates in boldface are significant at p < 0.01 level.

 

 Results from equations 1 through 4 are more or less in line with previous research findings (e.g., Rodgers 1979; Gravelle et al. 2002), even though a newly constructed data set is used.  Across all estimated equations, only the absolute measures of income are consistently significant, and the initial equations support the argument that life expectancy increases with income, but at a diminishing rate.  As hypothesized, the Gini index is negatively related to life expectancy, but is only significant when the reciprocal of per capita GDP, rather than logged income, is used.  Note that the equations with the best fit, or highest adjusted R2 values, include the democracy variable.  A rather strong positive relationship exists between the level of democracy and life expectancy, and the inclusion of democracy only slightly diminishes the strength of the association between the Gini index and health (i.e, comparing equations 3 and 8).  Moreover, in equations using the reciprocal of per captia GDP, the values of the constant terms fall within a reasonable range for life expectancy, especially the constant in equation 6 that is slightly greater than the average life expectancy for the entire data set.  Regression diagnostic tests indicate that the error terms from all estimated equations in Table 2 are contaminated by a significant amount of heteroskedasticity, the consequences of which are incorrect parameter and model inferences.  Heteroskedasticity is often the by-product of the geographic distribution of data values, which is seldom random, and which may also shed light upon the relationship between health, wealth and democracy.

Frequently overlooked in the debate about the significance of the relative income hypothesis is the role of geography.  This is a glaring oversight, especially at the global level, since the geography of mortality is not evenly nor randomly distributed across the world.  I contend that a geographic approach to examining the relative income hypothesis can clarify, inform and extend discussions about the relationships between income inequality, democracy and health.  Health is not only a function of individual traits, but is also related to where one lives.  Broadly speaking, geography is implicated with health in several ways.  For example, the combinations and conjunctures of tropical environments which tend to harbor and promote infectious disease, economic underdevelopment, authoritarian regimes and civil and inter-state conflict seriously challenge the population health of many countries in sub-Sahara Africa (WHO 1992).  On the contrary, Western Europe is home to economically developed and democratic countries that enjoy elevated and relatively stable standards of living.  Such difficulties and privileges are not necessarily contained by international borders, but are regional in nature and extent (Phillips et al. 1997).  These regional contexts vary considerably across the world, as do the inter-state interactions (e.g., war, trade, monetary union, etc.) that tend to color and characterize regions.  Therefore, recognizing the geographic context in which countries are situated, and the possibility and consequences of interactions between proximate countries, may clarify the pathways to poor or better health.  

The assertion that the geography of mortality is not evenly nor randomly distributed across the world can be verified by looking through any introductory world regional geography textbook.  Most texts contain maps that shade countries different colors according to infant mortality rates or life expectancy, and familiar patterns emerge (see Maps 1 and 2).  The Moran’s I statistic, which tests for spatial autocorrelation within a data set, provides a means to evaluate formally such geographic patterns.  Spatial autocorrelation refers to the clustering or association of similar values of a variable, such as life expectancy, in the spatial domain.  Utilizing a spatial weights matrix, W, that summarizes geographic relations between observations (e.g., 1 = adjacency, 0 = non-adjacency), Moran’s I provides a general indication of whether or not spatial dissimilarity or clustering exists in a set of data (for details, see Cliff and Ord 1974; 1981 Griffith 1987; Shin and Ward 1999; Gleditsch and Ward 2001)1.  Values of Moran’s I fall between negative one and positive one, the former suggesting a checkerboard pattern of dissimilarity and the latter indicating that the value of the variable of interest in one observation can effectively be predicted by averaging the values found in neighboring observations.

 



 

Maps 1 and 2. World life expectancies and infant mortality rates, 1999.


 

 

All four variables of interest in this analysis exhibit statistically significant levels of positive spatial autocorrelation (i.e., z­-scores for all variables > 5.0, based upon the

assumption of randomness).  Life expectancy has the highest Moran’s I value of 0.85, followed by that for per capita GDP (0.84), the Gini index (0.71) and the democracy score (0.53).  These results are not necessarily unexpected, and the heteroskedasticity detected in previous models may be related to these elevated levels of spatial autocorrelation in the data.  If spatial autocorrelation indeed contaminates the error term, thus violating the assumptions of normality and independence, parameter estimates will be inefficient.  This inefficiency can lead to misleading assessments of t and F statistics, and evaluations of a model’s fit using R2 will also be incorrect.   If spatial autocorrelation affects the dependent variable, y, then parameter estimates will be biased and inferences based upon the regression will be incorrect altogether (Anselin 1995).  

To combat the effects of spatial autocorrelation, the inclusion of a spatial component on the right-hand side of the model is necessary.  Incorporating of a vector of observations that is equivalent to the weighted average of adjacent values for each observation can help to resolve and to control spatial autocorrelation issues related to the dependent variable (for details, see Anselin 1988).  Using the same spatial weights matrix, W, used in the calculation of Moran’s I, multiplying the dependent variable, y, and the row-standardized spatial weights matrix, W, returns what is referred to as the spatial lag of y, or Wy.  The regression is now specified,

 

y = α + ρWy + βX + ε                                                                              (2)

 

where ρ is the spatial autoregressive parameter for the spatial lag, Wy, and the remaining terms are defined as in Equation 1.  Since spatial autocorrelation relates to the dependent variable, this problem is considered substantive because it may have important theoretical implications, and it requires careful interpretation of the spatial autoregressive parameter. 

For the case of a spatially autocorrelated error term, Equation 1 can be re-specified with a spatial constraint placed upon the error term

 

y = α + βX+ε                                                                           (3)

 

ε = λWε + u                                                                              (4)

 

where Wε is the spatial lag of the error term, λ, is the corresponding parameter estimate and u is the homoskedastic and independent error term.  Since the parameter estimates are not biased in Equation 3, the spatial autocorrelation that contaminates the error term is considered more of a nuisance than a substantive problem (for details, see Anselin 1988; 1995).  Table 3 provides the maximum likelihood estimates of the best-fitting OLS model in Table 2 (i.e., equation 8), controlling for various spatial effects and using various combinations of independent variables.

            Beginning with the spatial lag models (i.e., equations 1-4), all parameters are statistically significant and parallel the OLS estimates in both sign and value provided in Table 2.  In fact, the parameter estimates for democracy are identical or differ only slightly between two of the corresponding equations in Tables 2 and 3.  The first of these spatial regression models support the argument that life expectancy in one country is related to that in surrounding countries, that relative income is negatively linked to population health, and

           

 

dependent variable:  life expectancy, 1999

 

1

2

3

4

5

6

7

8

9

 

constant

 

 

70.03

 

77.78

 

62.62

 

69.90

 

71.39

 

81.83

 

67.30

 

77.33

 

79.76

 


 

 

-15,716

(1,217)

 

  -15,151

(1,204)

 

-13,157

(1,276)

 

-12,664

 

-10,028

(1,282)

 

-9,187

(1,228)

 

-10,252

(1,253)

 

-9,352

(1,203)

 

-16,664

(1,775)

 

 

Gini index

 

 

 

      -0.18

(0.07)

 

 

 

  -0.16

(0.06)

 

 

  -0.27

(0.08)

 

 

  -0.25

(0.07)

 

  -0.15

(0.05)

 

democracy

 

 

 

 

    0.88

(0.21)

 

    0.84

(0.20)

 

 

 

   0.61

(0.20)

 

   0.57

(0.19)

 

   0.39

(0.18)

 

λ  (error)

 

 

 

 

 

 

    0.66

(0.07)

 

    0.65

(0.07)

 

    0.57

(0.08)

 

   0.57

(0.08)

 

   0.31

(0.11)

 

 (lag)

 

 

    0.08

(0.03)

 

       0.07

(0.03)

 

    0.08

(0.03)

 

    0.07

(0.03)

 

 

 

 

 

 

likelihood

  ratio

 

 

-289.10

 

-285.74

 

-281.04

 

-277.79

 

-282.05

 

-276.38

 

-278.04

 

-272.46

 

 

-252.40

 

Table 3.  Maximum likelihood estimates of the relative income hypothesis, spatial effects

                  included, standard errors in parentheses.  Estimates in boldface are significant at

      the p < 0.01 level.

 

that higher levels of democracy have an important positive effect upon life expectancy.  Note that the absolute effect of democracy is over five times greater than that of relative income according to estimates in equation 4.  Log likelihood ratios indicate that the four spatial lag models are preferred over their OLS equivalents, with equation 4 providing the best overall fit.  Regression diagnostics for the four spatial lag models indicate that heteroskedasticity remains a problem, and that a significant amount of spatial autocorrelation continues to contaminate the error term. 

            Equations 5 through 8 in Table 3 provide the parameter estimates of the spatial error model, with the same combination of variables used previously.  All coefficients are statistically significant, with the sign of estimates resembling those found in previous equations.  When the log likelihood value from each spatial error equation is compared to its corresponding value from the spatial lag equation, it is clear that the latter spatial error specification is preferred.  Of particular interest is that the inclusion of the spatial error term, λ, increases the strength of the relationship between the inequality measure and life expectancy, but it reduces the overall impact (i.e., strength and significance) of democracy upon health.  The size and significance of the spatial error term also confirms that the spatial autocorrelation within the error term needed to be controlled. 

Notwithstanding the explicit inclusion of the spatial error term, λ, on the right hand side of the best fitting model in Table 3 (i.e., equation 8), regression diagnostics indicate that a significant amount of heteroskedasticity remains.  Heteroskedasticity is common in cross-sectional analyses such as this, and often arises when the data exhibit a large range of observations (Gujarati 1995).  It is likely that the variable used to measure income is the primary culprit of the non-constant variance of the error terms in previously estimated regressions given its broad range.  In order to correct for heteroskedasticity, each observation is weighted to eliminate the effects of larger variance terms relative to smaller ones.  Equation 9 of Table 3 contains estimates from a weighted model that jointly controls for heteroskedasticity and spatial error dependence2.  Relative income is again significant and negatively related to health, but its overall strength is reduced considerably when compared to previously estimated equations.  The effect of democracy is also diminished in the final spatial error/heteroskedastistic model.  The democracy parameter is about half as strong compared to previous models and is only significant at the p < 0.05 level.  Overall, however, this final model provides the best fit and returns the highest log likelihood ratio out of all estimated equations in all tables.

The results from these analyses generally support the relative income hypothesis, or that greater levels of income inequality negatively impact population health.  More importantly perhaps are the findings that levels of democracy are positively related to life expectancy, and that a geographic perspective can help to clarify the linkages between wealth, health and democracy.  With regard to the former, it is interesting to note that the effect of democracy was at least twice as strong as that of income inequality measured in absolute terms in all estimated models.  Furthermore, unreported multicollinearlity condition numbers from the OLS regressions, combined with the relatively low correlations between the democracy score and Gini index, suggest that there is little interaction between these two variables.  Though the primary objective of this analysis was to determine the effects of geography, democracy and inequality upon health, this result is more or less in line with the arguments put forth by Bollen and Jackman (1995) who are skeptical of the association between income inequality and democracy.

It should be noted that the relative income hypothesis, and the methodologies used to verify its influence, are not without their critics (e.g., Judge 1995; Gravelle 2002).  Some skeptics view the relationship between income inequality and health as a statistical artifact arising from the ecological fallacy, or making inferences about individuals from aggregate data (Robinson 1950).  Since income inequality is a societal characteristic and not an individual one, skeptics contend that inferring health outcomes from population-level variables is problematic to begin with and unreliable at best (e.g., Gravelle 1998).    Within the scope of this analysis, the same critique can be leveled against the inclusion of the democracy measure; societies are measured for their degrees of democracy, not individuals.  Similarly, as mentioned earlier, issues related to comparability and consistency arise when measuring inequalities both within and between societies.  Despite such criticism and the ongoing debates surrounding the relationship between income inequality and health, the potential social and policy implications of the relative income hypothesis are too significant to disregard.

This analysis furnished several insights into the relationships between wealth, health and democracy, and the importance of a geographic perspective is also underscored.  Though population health is measured on a country-by-country basis, regional circumstances often play an important role with respect to the opportunities for better or worse health and need to be explored further.  Since such geographic contexts are not identical across space or stable over time, their possible influence can vary considerably.  Failure to recognize and to consider the effects of geography, or in other words, that wealth, democracy and health in one country are related to that found in neighboring countries, can lead to misleading statistical inferences and inaccurate conclusions.  Therefore, situating examinations of the relative income hypothesis within a geographic framework can help to clarify and extend our understanding of the relationships between income inequality, democracy and health.  

 

 

Conclusions

 

The relative income hypothesis presents itself as a salient research puzzle to academic and policy circles alike.  These analyses support previous work on the relative income hypothesis, and demonstrate that greater degrees of income inequality negatively affect population health, even when considering levels of democracy and geographic effects.  The consistently strong, positive and independent association between levels of democracy and life expectancy merits further investigation in itself, as well as within the scope of the relative income hypothesis.  With regard to the latter, the existing body of research on the linkages between economic development, democracy and inequality is relevant to the relative income hypothesis (e.g., Midlarsky 1997), and may help to articulate the pathways between income inequality and health. 

            In addition to assessing the importance of democracy to health, these analyses highlight the utility and perhaps the necessity of a geographic perspective and a spatial analytic framework when examining the relative income hypothesis.  Few studies acknowledge the geographic dimensions of income inequality and health, and there are two important implications of this oversight.  First, in methodological terms, failing to test and to control for spatial effects, which are present in the data used here and are likely to exist in other cross-sectional data sets, can lead to misspecified models and incorrect inferences.  The exploratory spatial techniques confirm that levels of wealth, democracy and health in one country are strongly correlated to levels found in surrounding countries.  Second, disregarding the geographic nature of health (e.g., refer to Maps 1 and 2), wealth and democracy is analogous to assuming that the world consists of autarkic, isolated and solitary countries.  A geographic approach can help to clarify and identify regional obstacles and pathways to better health.  For example, though mortality rates are most frequently reported by country, regional environmental, political, economic and social factors arguably play an important role in elevating or depressing such rates.

The lead feature of the 16-22 June 2002 edition of the Economist asks the question, “Does inequality matter?”.  Though a considerable body of literature exists on the subject, particularly in the field of economics, a definitive answer to this question is difficult to find and to articulate.  The finding that income inequality may very well be linked to health provides a considerable amount of impetus to both academics and policy-makers to try to understand better the social consequences of inequality.  However, as noted by Sen (1992:  88), “the evaluation of inequality has to take note of both the plurality of spaces in which inequality can be assessed, and the diversity of individuals”.  By situating questions and analyses about inequality, democracy and health within geographic context, ranging from the global to the local, our understanding of inequality as a social concept and its impact upon individuals will ultimately be expanded. 

 

 

 


Notes

 

1.  A spatial weights matrix based upon an arbitrarily selected distance threshold of 100 kilometers is used to summarize the geographic relationships between countries in 1998.  For details, see Gledtisch and Ward (2001).  Note that the spatial matrix used is row-standardized (i.e., each row in the binary matrix of linkages is divided by the total number of linkages in that row), which corresponds to a form of spatial smoothing (for details, see Anselin 1988).

 

2.  Estimates of the joint heteroskedastistic and spatial error dependence model should be accepted with caution.  Though the estimation of the model is valid, it is based upon the assumption that heteroskedastiscity is caused exclusively by the weighting variable, which may not be the case.
References

 

Amiel, Y. and F. Cowell.  1999.  Thinking about inequality.  Cambrige University Press, Cambridge.

 

Anselin, L.  1988.  Spatial Econometrics:  Methods and Models.  Kluwer Academic, Dordrecht.

 

Anselin, L.  1995.  SpaceStat, A Software Program for the Analysis of Spatial Data, Version 1.80.  Regional Research Institute, West Virginia University, Morgantown, WV.

 

Ben-Slomo, Y, I. White and M. Marmot.  1996.  Does the variation in the socioeconomic characteristics of an area affect mortality?  British Medical Journal 312:  1013-1014.

 

Berkman, L and L. Breslaw.  1983.  Health and ways of living:  The Alameda county study.  Oxford University Press, Oxford.

 

Bollen, K. and R. Jackman.  1985.  Political Democracy and the Size Distribution of Income.  American Sociological Review 50:  438-457.

 

Bollen, K. and R. Jackman.  1995.  Income Inequality and Democracy Revisited:  A Comment on Muller.  American Sociological Review 60:  983-989.

 

Chiang, T.  1999.  Economic transition and changing relation between income inequality and mortality in Taiwan:  regression analysis.  British Medical Journal 319:  1162-1165.

 

Cliff, A. and J. Ord.  1981.  Spatial Processes: Models and Applications.  Pion, London.

 

Cliff, A. and J. Ord.  1974.  Spatial Autocorrelation.  Pion, London.

 

Deininger, K. and L. Squire.  1996.  A new data set measuring income inequality.  World Bank Economic Review 10:  565-591. 

 

Duleep, H.  1995.  Mortality and income inequality among economically developed countries.  Social Security Bulletin  58:  34-50.

 

Flegg, A.  1982.  Inequality of income, illiteracy and medical care as determinants of infant mortality in underdeveloped countries.  Population Studies 36:  441-458.

 

Ghobarah, H., P. Huth and B. Russett.  2001.  Civil wars kill and maim people – long after the shooting stops.  Draft version.  Center for Basic Research in the Social Sciences.  www.cbrss.Harvard.edu/programs/hsecurity/papers/civilwar.pdf

 

Gleditsch, K. 2002.  Polity Data Archive.  http://weber.ucsd.edu/~kgledits/Polity.html (accessed 12 February 2002).

 

Gleditsch, K. and M.D. Ward.  1997.  Double Take:  A Reexamination of Democracy and Autorcracy in Modern Polities.  Journal of Conflict Resolution 41:  361-383.

 

Gleditsch, K. and M.D. Ward.  2000.  War and Peace in Space and Time:  The Role of Democratization.  International Studies Quarterly 44:  1-29.

 

Gleditsch, K. and M.D. Ward. Measuring Space: A Minimum Distance Database and Applications to International Studies.  Journal of Peace Research 38:739-58

 

Gradstein, M. and B. Milanovic.  2000.  Does Liberté = Egalité?  A survey of the empirical evidence on the links between political democracy and income inequality.  World Bank working paper.

 

Gravelle, H.  1998.  How much of the relation between population mortality and unequal distribution of income is a statistical artefact?  British Medical Journal 316:  382-385.

 

Gravelle, H.  2002.  Income, income inequality and health:  what can we learn from aggregate data?  Social Science & Medicine 54:  577-589.

 

Griffith, D.  1987.  Spatial Autocorrelation:  A Primer.  Association of American Geographers Resource Publications, Washington, D.C.

 

Gujarati, D.  1995.  Basic Econometrics, 3rd edition.  McGraw-Hill, Inc., New York.

 

Hales, S. P. Howden-Chapman, C. Salmond.  1999.  National infant mortality rates in relation to gross national product and distribution of income.  Lancet 354:  2047.

 

Heclo, H.  1999.  Hyperdemocracy.  The Wilson Quarterly.  Winter.

 

Jones, C.  1997.  On the Evolution of the World Income Distribution.  Journal of Economic Perspectives 11:  19-36.

 

Jones, K. and G. Moon.  1992.  Medical geography:  global perspectives.  Progress in Human Geography 16:  563-572.

 

Judge, K.  1995.  Income distribution and life expectancy:  a critical appraisal.  British Medical Journal 311:  1282-1285.

 

Kahn, H. L. Tatham, E. Pamuk, C. Heath.  1998.  Are geographic regions with high income inequality associated with risk of abdominal weight gain?  Social Science and Medicine 47:  1-6.

 

Kaplan, G., E. Pamuk, J. Lynch, R. Cohen and J. Balfour.  1996.  Inequality in income and mortality in the United States:  analysis of mortality and potential pathways.  British Medical Journal 312:  999-1003

 

Kawachi, I. and B. Kennedy.  1997.  The relationship of income inequality to mortality:  Does the choice of indicator matter?  Social Science and Medicine 45:  1121-1127.

 

Kawachi, I., B. Kennedy, K Lochner and D. Prothrow-Stith.  1997.  Social capital, income inequality and mortality.  American Journal of Public Health 87:  1491-1498.

 

Kawachi, I., B. Kennedy, and R. Wilkinson, eds.  1999.  The Society and Population Health Reader:  income inequality and health.  The New York Press, New York.

 

Kennedy, B., I. Kawachi and D. Prothrow-Stith.  1996.  Income distribution and mortality:  cross-sectional ecological study of the Robin Hood Index in the United States.  British Medical Journal 312:  1004-1007.

Kuznets, S.   1955.  Economic growth and income inequality.   American Economic Review, 45: 1-28.

 

Le Grand, J.  1987.  Inequalities in health.  Some international comparisons.  European Economic Review 31:  182-191.

 

Lenski, G.  1966.  Power and privilege:  A theory of social stratification.  McGraw-Hill, New York. 

 

Lipset, S.M.  1959.  Some Social Requisites of Democracy:  Economic Development and Political Legitimacy.  American Political Science Review 53:  69-105. 

 

Marmot, M. A. Adelstein, N. Robinson and G. Rose.  1978.  Changing Social Class Distribution of Heart Disease.  British Medical Journal II:  1109-1112.

 

Merva, M. and R. Fowles.  1993.  Effects of diminished economic opportunities and social stress:  Heart attacks, stroke and crime.  Economic Policy Institute, Washington, D.C.

 

Midlarsky, M.I.  1997.  Inequality, democracy and economic development.  Cambridge University Press, Cambridge.

 

Milanovic, B.  1999.  True world income distribution, 1988 and 1993:  First calculation based on household surveys alone.  Working paper World Bank, Development Research Group, Washington, D.C.

 

Muller, E.  1988.  Democracy, Economic Development and Income Inequality.  American Sociological Review 53:  50-68.

 

Muller, E.  1995.  Economic Determinants of Democracy.  American Sociological Review 60:  805-821.

 

Murray, C., G. King, A. Lopez, N. Tomijima and E. Krug.  2002.  Armed conflict as a public health problem.  British Medical Journal 324:  346-349.

 

O’Loughlin, J., M.D. Ward, C. Lofdahl, J. Cohen, D. Brown, D. Reilly, K. Gledtisch and M. Shin.  1998.  The diffusion of democracy, 1946-1994.  Annals of the Association of American Geographers 88:  545-574.

 

Omran, A.  1971.  The epidemiological transition:  a theory of the epidemiology of population change.  Milbank Quarterly 64:  355-391.

 

Omran, A.  1983.  The epidemiological transition theory:  a preliminary update.  Journal of Tropical Pediatrics 29:  305-316.

 

Phillips, D., P. Groenewegen and Y. Verhasselt.  1998.  Health, environment and development:  issues in developing and transitional countries.  Geojournal 44:  97-102.

 

Picheral, H.  1989.  Géographie de la transition epidémiologique.  Annales de Géographie 546:  129-151.

 

Prichitt, L.  1997.  Divergence, Big Time.  Journal of Economic Perspectives 11:  3-17.

 

Robinson, W.  1950.  Ecological correlations and the behavior of individuals.  American Sociological Review 15:  351-357.

 

Rodgers, G.  1979.  Income and Inequality as Determinants of Mortality:  An International Cross-Section Analysis.  Population Studies 33:  343-351.

 

Ray, J. L.  1993.  Wars between democracies:  rare, or nonexistent?  International Interactions 18:  251-276.

 

Russett, B.  1993.  Grasping the Democratic Peace.  Princeton University Press, Princeton.

 

Sen, A.  1973.  On Economic Inequality.  Clarendon Press, Oxford.

 

Sen, A.  1992.  Inequality re-examined.  Harvard University Press, Cambridge, MA.

 

Shin, M. and M.D. Ward.  1999.  Lost in Space:  Political Geography and the Defense-Growth Trade-Off.  Journal of Conflict Resolution 43:  793-817.

 

Summers, R. and A. Heston.  1991.  The Penn World Table (Mark 5) – and expanded set of international comparisons, 1950-1988.  Quarterly Journal of Economics 106:  327-368.

 

van Doorslaer, E., A. Wagstaff, H. Bleichrodt, S. Calonge, U. Gerdtham and M. Gerfin.  1997.  Income-related inequalities in health:  some international comparisons.  Journal of Health Economics 16:  93-122.

 

Walberg, P., M. McKee, V. Shkolnikov, L. Chenet, D. Leon.  1998.  Economic change, crime and mortality crisis in Russia:  regional analysis.  British Medical Journal 317:  312-318.

 

Waldmann, R.  1992.  Income distribution and infant mortality.  The Quarterly Journal of Economics 107:  1283-1302.

 

Ward, M.D. and K. Gleditsch.  1998.  Democratizing for Peace.  American Political Science Review 92:  51-62.

 

Wennemo, I.  1993.  Infant mortality, public policy, and inequality – a comparison of 18 industrialised countries  1950-1985.  Sociology of Health and Illness 15:  429-446.

 

Wilkinson, R., ed.  1986.  Class and health:  Research and Longitudinal Data.  Tavistock, London.

 

Wilkinson, R.  1992.  Income distribution and life expectancy.  British Medical Journal 304:  165-168.

 

Wing, S.  1988.  Social Inequalities in the Decline of Coronary Mortality.  American Journal of Public Health 78:  1415-1416.

 

World Bank.  2001.  2001 World Development Indicators.  World Bank, Washington, D.C.

 

World Health Organization.  1992.  Our planet, our health.  WHO, Geneva.


Appendix A. 

 


cowcode

lifex99

rgdp92

gini

dem98

ALG

71

2719

35.3

1

AUL

79

14458

35.2

10

AUS

78

12955

23.1

10

BNG

61

1510

33.6

6

BEL

78

13484

25

10

BOL

62

1721

58.9

9

BRA

67

3882

59.1

8

BUL

71

5208

26.4

8

BFO

45

514

48.2

3

CAN

79

16362

31.5

10

CEN

44

514

61.3

6

CHL

76

4890

57.5

8

CHN

70

1493

40.3

0

COL

70

3380

57.1

7

COS

77

3569

45.9

10

CDI

46

1104

36.7

0

CZR

75

3303

25.4

10

DEN

76

14091

24.7

10

DOM

71

2250

47.4

8

ECU

69

2830

43.7

9

EGY

67

1869

28.9

1

SAL

70

1876

50.8

9

ETH

42

274

40

3

FIN

77

12000

25.6

10

FRN

79

13918

32.7

9

GAM

53

784

47.8

0

GRG

73

1837

37.1

5

GFR

77

14709

30

10

GHA

58

956

39.6

5

GRC

78

6769

32.7

10

GUA

65

2247

55.8

8

GUI

46

740

40.3

3

HON

70

1385

59

7

HUN

71

4645

24.4

10

IND

63

1282

37.8

8

INS

66

2102

31.7

0

IRE

76

9637

35.9

10

ISR

78

9843

35.5

9

ITA

78

12721

27.3

10

JAM

75

2455

36.4

9

JPN

81

15105

24.9

10

JOR

71

2951

36.4

1

KEN

48

914

44.5

3

ROK

73

7550

31.6

10

 

cowcode

 

lifex99

 

rgdp92

 

gini

 

dem98

LAO

54

1442

37

0

LAT

70

2692

32.4

8

MAG

54

608

46

8

MAL

72

5746

49.2

0

MLI

43

522

50.5

6

MEX

72

6253

51.9

6

MON

67

1454

33.2

9

MOR

67

2173

39.5

1

MZM

43

711

39.6

6

NEP

58

1091

36.7

8

NTH

78

13281

32.6

10

NIC

69

1221

60.3

8

NIG

47

978

50.6

0

NOR

78

15518

25.8

10

PAK

63

1432

31.2

7

PAN

74

3332

48.5

9

PNG

58

1606

50.9

10

PAR

70

2178

57.7

7

PER

69

2092

46.2

3

PHI

69

1689

46.2

9

POL

73

3826

31.6

9

POR

75

7869

35.6

10

RUM

69

1555

28.2

8

RWA

40

762

28.9

0

SEN

52

1117

41.3

3

SIE

37

734

62.9

5

SLO

73

4463

19.5

8

SAF

48

3068

59.3

9

SPN

78

9802

32.5

10

SRI

73

2215

34.4

6

SWD

79

13986

25

10

SWZ

80

15887

33.1

10

TAZ

45

493

38.2

3

THI

69

3942

41.4

9

TRI

73

7218

40.3

9

TUN

73

3075

41.7

1

TUR

69

3807

41.5

8

UGA

42

547

37.4

2

UK

77

12724

36.1

10

USA

77

17945

40.8

10

URU

74

5185

42.3

10

VEN

73

7082

48.8

8

ZAM

38

667

52.6

3