Social Inequality and Health in México 1990-1997: Individual and Policy Pathways – “Former: Impact of Income Inequalities on Mexico’s Popoulation Health”
Adolfo Martínez Valle, The Johns Hopkins University
7 de Diciembre de 2001
ABSTRACT
AIMS: To study how social inequality has influenced the health status of the Mexican population at the individual level from a social class perspective as well as the state level from a policy perspective between 1990 and 1996. Explore the individual and policy pathways through which social inequality affects health.
METHODS: To empirically examine the social class gradient in perceived health in Mexico a secondary cross-sectional analysis was designed using logistic multiple regression models. To empirically examine the association between social inequality and health across states, pooled cross-sectional data was used. Secondary analysis was conducted for the 32 Mexican states for 1990 and 1996 using general estimation equation (GEE) models. Path analysis was performed to explore how the electoral strength of opposition political parties shape health care resource allocation decisions which in turn influence population health.
RESULTS: The cross-sectional individual-level analysis provided empirical evidence that the lower the social class, the poorer the perception of health. The results of the pooled cross-sectional analysis indicated that social inequality across states was as expected, positively and significantly associated with maternal and infant mortality rates. The path analysis findings suggest that the electoral strength of PRD affects maternal mortality through its impact on the distribution of primary care physicians per capita via targeted health care expenditures.
CONCLUSIONS: Overall, the findings of this study provided empirical evidence that social inequality negatively influences health both at the individual and state level. The results also suggest that social inequality may be influencing health through both material and psychosocial mechanisms at the individual level and through policy mechanisms at the state level.
KEY WORDS: social inequality, socioeconomic determinants of health, social class, policy process, politics, Mexico.
TABLE OF CONTENTS
Introduction
1
PART I: INDIVIDUAL EFFECTS OF SOCIAL INEQUALITY 6
METHODOLOGY 7
RESULTS 13
DISCUSSION
19
PART II: POLICY EFFECTS OF SOCIAL INEQUALITY 23
METHODOLOGY 24
RESULTS 27
PATH ANALYSIS 32
DISCUSSION
37
GENERAL CONCLUSIONS
References
39
43
Introduction
One of the most intriguing findings in the recent public health literature is the strong influence income inequality seems to have on health. The greater the income differences within populations, the worse their health differentials. This evidence suggests that worsening health conditions are associated with widening income disparities both across countries (Rodgers, 1979; Flegg, 1982; Waldmann, 1992; Wilkinson, 1992; Wennemo, 1993; Duleep, 1995; McIssac and Wilkinson, 1997; Judge et al, 1998; Armada, 2000) and within countries (Bronfman, 1992; Wilkinson, 1992 &1996; Lynch et al, 1998; Hollstein et al, 1998; Szwarchwald CL et al, 1999; Kawachi et al, 1999, 1997; Kennedy et al, 1998, 1996; Soobader and Leclere, 1999; Shi et al, 1999).
Examining the associations between socioeconomic inequalities and health differentials is not new. Research from the Anglo-Saxon tradition (Marmot and Wilkinson, 1999; Berkman and Kawachi, 2000) as well as from the Latin American school of social epidemiology (Almeida-Filho, 1999) has made extensive and important contributions in the past. The Anglo-Saxon tradition has provided empirical evidence of how socioeconomic status measured by social class, occupation, education, and income positively influences health across the social spectrum (Marmot and Wilkinson, 1999; Berkman and Kawachi, 2000).
However, this research tradition has focused almost exclusively on individual social risk factors. This dissertation proposes to go beyond this individualistic approach by focusing on the social structure of inequality as an influencing factor on health. From this perspective, the underlying assumption is that health differentials across the social spectrum are closely related to the structure of a society, which in turn is shaped by social, economic and political processes.
This approach is similar in a way to the Latin American school of social epidemiology, which has emphasized the importance of class and the context under which social inequalities and its health consequences are manifested. However, Latin American research has mostly followed a theoretical approach and has not been able to provide strong empirical evidence of their important theoretical and conceptual contributions except for a few exceptions (Bronfman, 1992; Lozano et al, 2001).
This dissertation sought to reduce this empirical gap in the Latin American literature.
The more recent income inequality approach has sparked a renewed interest in this field of inquiry with its revealing empirical findings. Despite its theoretical limitations it has drawn the attention of many public health researchers. There are three main reasons for this. First, income inequality has grown considerably both within and across most countries. Second, health differentials persist, despite overall health improvements in terms of life expectancy and other population health indicators. Third, social inequality including income can be a target for public policies seeking to improve both health and living conditions.
Most of these income inequality studies have been conducted in industrialized countries, mainly in Great Britain and the United States. Furthermore, many have failed to explicitly define a theoretical framework to guide their empirical approach. Therefore, the novelty of this dissertation lies in exploring if a similar association exists in Mexico, a middle-income country that has experienced both greater income disparities and population health inequalities over the past two decades. Studies have shown that income inequalities have worsened since the 1980s and continued through the 1990s, though at a lesser growth rate (Cortés, 1998; De la Torre, 1995; Lustig, 1998).
Household survey data indicate that between 1984 and 1989 the income share of the bottom 90 percent of the total population decreased, while only the share of the top 10 percent increased (De la Torre, 1995).
Using the Gini[ The Gini coefficient is a summary measure of the degree of inequality. Its values range from 0 to 1 where 0 means perfect equality and 1 perfect inequality. For developed countries the Gini coefficient of earned pretax income tends to be around 0.4.] coefficient, income inequalities decreased from 0.53 in 1977 to 0.466 in 1984[ This improvement in the distribution of income, however, should be interpreted with caution because the two sources of information are not comparable due to methodological differences.], increased from 0.504 in 1989 to 0.528 in 1994, decreased to 0.488 in 1996, and increased again to 0.509 in 1998 (INEGI, 1996; 1998; 2000).
Overall, the degree of income inequality has increased by nearly 10 percent between 1984 and 1994. However, income is just one indicator of social inequality. The marginality index[ The area-based index of marginality is a policy-oriented indicator developed by the Mexican government (CONAPO, 1998) that measures the lack of basic public infrastructure as well as education and material living conditions at the state and county levels. More specifically, this inequality indicator measures the percentage of people over 15 years who are illiterate; the percentage of people without complete basic education; the percentage of the employed labor force earning less than twice the minimum wage (6 US$ per day); the percentage of people living in households in a town with less than 2,500 inhabitants without running water, without electricity, without sewage facilities, without solid floor materials and overcrowding living conditions. Principal component statistical analysis was performed to construct this area-based indicator of relative deprivation (CONAPO, 1998).
The marginality index is a normalized Z-score ranging between -3 and 3 standard deviations that correspond respectively to very low and very high marginality (CONAPO, 1998).], an area-based indicator of relative deprivation also shows that social inequality has not improved substantially in Mexico. According to this index, marginality has been reduced in 17 out of the 32 Mexican states between 1990 and 1995, while the remaining 15 states have experienced higher marginality figures (CONAPO, 1998).
This indicator was designed by the Mexican government to assess the degree of access to basic public services and infrastructure across both states and counties.
In Mexico population health has improved in the past two decades in terms of reducing mortality rates and increasing life expectancy. However, these average nationwide improvements hide the worsening of health conditions in at least two ways. First, they do not show the regional disparities within Mexico. The reductions in infant mortality across states were not the same between 1985 and 1995. The state with the highest reduction was Tlaxcala, where infant mortality was reduced from 38.2 to 30.4 children deaths less than 1 year of age per thousand live births. This 20 percent reduction contrasts with 5 percent reduction in Baja California, the lowest, which was reduced from 26.5 to 25.1 children deaths less than 1 year of age per thousand live births. The poorest states (Guerrero, Chiapas, and Oaxaca) reduced their infant mortality rates 14 percent on average between 1985 and 1995, but still experienced the highest mortality rates (ranging from 48.4 in Chiapas to 44.6 in Guerrero in 1995) in Mexico among its 32 states (FUNSALUD, 1998).
Furthermore, a previous study (Lozano, 1997) showed that child mortality in 1994 was almost two times higher in the poorer states, 523 per thousand children under 5 years of age (Guerrero, Chiapas, and Oaxaca) than in the richest ones, 278 per thousand (Distrito Federal, Nuevo Leon, Tamaulipas, Coahuila and Baja California Sur).
Health inequalities are also manifested in the adult population. The poorer states have a mortality rate of 139 deaths per thousand adults between 15-59 years of age, while the richest states have a rate of 105 per thousand (Lozano, 1997).
Second, some indicators reveal deterioration in the health of the population. For example, infant and preschool mortality caused by nutritional deficiencies increased since 1982 after years of steady decline. This specific cause of mortality accounted for 1.5 percent of total infant mortality and 2.4 percent of total preschool mortality in 1982, increasing its percentages to 5.2 and 9.1 respectively in 1988 (Lustig, 1998).
A more recent study shows that malnutrition inequalities have not been reduced significantly in the past decade (Roldán JA et al, 2000).
Poor states like Oaxaca and Chiapas show a malnutrition index[ The index of malnutrition was calculated through the principal component statistical technique based on information regarding stunting, infant and child mortality, marginality, indigenous population and deaths due to respiratory and gastrointestinal diseases.] of 35 in 2000, while the richest states such as Nuevo Leon, Baja California and Distrito Federal present figures below 8 (Roldán JA et al, 2000).
Thus, there is evidence that inequality may have an impact on health in the Mexican context. Furthermore, there is no consensus on why social inequality may affect health even if an empirical association is found. The main aim of this dissertation, therefore, was to study how social inequality has influenced the health status of the Mexican population at the individual and the state level between 1990 and 1996.
To achieve this goal this dissertation was organized as follows. To empirically examine the association between social inequality and health in the Mexican context this dissertation was divided in two parts. The first part analyzed how social inequality negatively influences health at the individual level. Using data from the Second National Health Survey (ENSA II, 1994) this level of cross-sectional analysis aims to investigate the degree and factors associated with the social class gradient in health status. Furthermore, it sought to explore the material and psychosocial pathways through which income inequality may affect health. The second part of this dissertation analyzes the association between social inequality and population health at the state level using several data sources including electoral data. This state level of analysis aimed to explore the ecological and political pathways through which social inequality may influence health in Mexico. Finally, this study sought to draw some conclusions from its empirical findings including the policy implications of targeting social inequality to improve the health of the Mexican population.
PART I
INDIVIDUAL EFFECTS OF SOCIAL INEQUALITY
The purpose of this first part of the study is to examine the association between social inequality and individual health using both individual measures and an area-based indicator of relative deprivation. This level of analysis also seeks to explore the material and psychosocial mechanisms that associate mortality and morbidity gradients with social class differentials. To study the social class gradients in morbidity, the sample was restricted to the occupied population aged 12 years and older.[ This age range was chosen because it is based on the official definition of the economically active population (INEGI, 1990).] This sample was divided in two groups: the urban working force and the rural working force. The first are those who worked in urban settings, while the rural were made up of people in agricultural related activities including the forestry and livestock industries. The final sample size was 13,062 individuals working in the urban sector and 4,614 working in the agricultural sector. The distinction was made because previous studies (INEGI, 1990; 1997; Bronfman, 1992; Bronfman et al, 1990; Bartra, 1991) have shown that urban and rural populations experience different socioeconomic living conditions.[ Exploratory data analysis also showed that urban and rural populations live under different socioeconomic conditions. ] For example, approximately 93 percent of the urban population in Mexico had access to drinking water services, while only 57 percent of the rural population had access to that kind of services (INEGI, 1997).
A cross-sectional secondary analysis of the Second Mexican National Health Survey (ENSA II) will be conducted using logistic multivariate statistical models.
METHODOLOGY
Individual measures were developed from ENSA II conducted in 1994. ENSA II is a personal interview type of survey, using a stratified multistage probability sample of households. The primary sampling unit was a dwelling. For sampling purposes, the country was divided in five regions and a sample size of approximately 2,523 dwellings was drawn from each region. Table 1 presents how these regions are defined.
Table 1
ENSA II regions
Region States Dwellings (n) Individuals (n)
NORTH
Baja California, Baja California Sur, Sonora, Chihuahua, Sinaloa, Coahuila, Nuevo León, Tamaulipas, Durango, Zacatecas
2,570
4,905
DF Mexico City Metropolitan Zone
2,520 5,139
CENTER San Luis Potosí, Guanajuato, Querétaro, Michoacán, México, Tlaxcala, Nayarit, Aguascalientes, Jalisco, Colima
2,620 5,225
SOUTH Morelos, Puebla, Veracruz, Campeche, Tabasco, Yucatán, QROO
2,520 5,227
PASSPA Hidalgo, Oaxaca, Guerrero, Chiapas 2,520 4,887
PASSPA: Health Services Aid Program for the Uninsured Population
Within each region, the sample was proportionally distributed according to the population size of each of the 32 Mexican states, setting a minimum of 100 dwellings for the smallest states (Secretaría de Salud, 1994).
ENSA II is representative of the Mexican population at both the national and the regional level. It contains information on health status and health care utilization of individuals, as well as demographic and socioeconomic characteristics of households and individuals. The survey has two questionnaires. The household questionnaire was designed to collect information for every household member, while the individual one collected information for every user of health services. The overall response rate was approximately 96.7 percent for both the household and the individual questionnaire. Data were obtained for 12,615 households, including 61,524 individuals (Secretaría de Salud, 1994).
The marginality index, an aggregated measure, was obtained from the National Population Council (CONAPO, 1998).
This index was developed to measure the degree of marginality in each Mexican state and county. The index is an indicator of deprivation based on housing, income, and schooling information collected from the 1990 Mexican Census and the 1995 Population and Housing Count (CONAPO, 1998).
ENSA II is linked to the state level marginality data by assigning a level of marginality to the state of residence of each individual. ENSA II is not designed to support state level estimation, however. To determine whether each state is represented in ENSA II in proportion to its actual share of the Mexican population, the distribution of ENSA II was compared to the 1990 census and 1995 mid-count data. The ENSA II sample distribution is similar to the census data except for the PASSPA states, which were oversampled. The final data set for this analysis is composed of individual level data, but it also includes a contextual level measure, the marginality index at the state level.
HEALTH OUTCOMES
Two indicators of perceived health status are used as health outcomes in this study: self-assessed health and reported morbidity. Both the social class gradient literature and the income inequality approach have shown that these morbidity indicators are sensitive health indicators to the effects of social inequality (Marmot et al, 1991; Farmer and Ferraro, 1997; Kawachi et al, 1999a).
Self-assessed health is a measure that has been shown to have a strong predictive validity (Farmer and Ferraro; Idler and Benyamini, 1997).
This indicator measures the perception that each individual surveyed had of its overall health status. This measure was originally rated on a 5-point scale: very poor, poor, average, good and very good. For the purposes of this study, it was transformed to a dichotomous measure equal to 1 if the response was average, poor or very poor.
Reported morbidity, on the other hand, is a health indicator that measures specific health problems such as heart disease, diabetes, digestive and respiratory diseases. This indicator captured health problems reported in the past two weeks, including illnesses, discomforts or accidents. The main health problems reported where respiratory diseases, muscle and bone related problems, gastrointestinal diseases, headaches, hypertension, mouth diseases, fever, diabetes, and cough. These health problems accounted for almost seventy percent of the total health problems reported in the ENSA II survey. This health outcome was also measured as dichotomous variable equal to 1 if the respondent suffered any health problem and 0 otherwise.
SOCIAL CLASS
To measure social inequalities, two approaches have been followed in the public health literature (Krieger et al, 1997; Muntaner et al, 2000).
Social stratification, the first approach, is a hierarchical ordering of individuals according to the degree of education, type of occupation and income level (Krieger et al, 1997).
Most studies use more than one dimension of stratification to better capture the effects of social inequalities on health. The class perspective, on the other hand, classifies individuals according to ownership of the means of production, authority over others in the workplace, skills exercised in the job, and control over different types of assets (Wright, 1997).
This study follows the first approach but it also incorporates elements of the class perspective such as business and land ownership.
ENSA II does not include all of these criteria of social class membership. For example, it only includes information on land ownership, but not on the size of the land. The operationalization of social class was therefore constructed following a stratification approach according to the occupation and education indicators available from ENSA II. This class operationalization was chosen to better measure its social gradient effects on health. Individuals were first stratified according to two broad occupation categories: the urban sector which comprised the manufacturing and the services industry and the rural sector which included all agricultural related economic activities. The urban labor force consisted of six occupational categories: employer, independent professional, employed, non-salaried workers and salaried workers. These occupational categories identified some of the technical aspects of work associated with prestige, wealth, skills and specific working conditions. The first two theoretically correspond to high-income socioeconomic status from a social stratification perspective and to a high social class or bourgeoisie from a standard class perspective. Education, measured by years of schooling, was also added as a criterion to avoid misclassification. This category therefore only included individuals whose occupation was either an independent professional or an employee of the manufacturing or services sector, which had at least a college degree in addition to the employers, which had 9 or more years of schooling.[ The variable operationalization appendix shows how each social class category was constructed.] The employed and non-salaried categories were more difficult to classify because they did not indicate a specific social class position. Education was therefore added as an additional criterion to sort individuals into the corresponding social classes to account for their respective urban labor force skills. Several measures of socioeconomic status have been used simultaneously in many studies because using a single measure has not adequately captured the health effects of social inequality (Muntaner and Eaton, 2000; Berkman and Kawachi, 2000).
Thus, the employed, which had 15 or more years of schooling, were classified in the high-class category, while the employed and non-salaried workers with 9 to 14 years of schooling were classified in the middle-high class. The non-salaried and salaried workers categories were sorted out into middle-low and low-income working class by years of schooling as well. The middle-low class were those that had between 7-8 years of schooling and the low-income working class corresponded to those that had 6 years or less of schooling. The education dimension of social class was based on the Mexican occupational stratification estimated by the Economic Commission for Latin America and the Caribbean (CEPAL, 1997).
The CEPAL stratifies occupations in three categories. The high category includes employers who have on average 9 years of schooling, managers with 13 years, and professionals with 15. The middle strata includes technicians who have 13 years of schooling, employees with 11 years, and merchants with 8. The low occupational category includes industrial and service workers who have on average between 6 and 7 years of schooling, and the agricultural workers who have 4 years of schooling on average. Table 1 shows the four categories of social class defined for the urban labor force according to the criteria described above.
Table 2
Urban social class operationalization
The agricultural labor force consisted of five occupational categories: landowners including ejidatarios[ Ejidatarios are small landowners who benefited from the agrarian reform, which initiated a land redistribution process in the 1930s with the Mexican Revolution. Under the ejido form of tenure land was granted to a village rather than to individuals. Although the land was generally held in common, in most cases it has been worked individually (Hansen, 1971).], land tenants, self-employed, salaried workers and non-remunerated workers. Relations and size of land ownership generally determine the classification of social class in the rural areas. However, ENSA II did not provide any information on this regard. Education was again employed here to stratify the respective social classes. Additionally, a proxy for income was included as an additional criterion for agriculture class differentiation to reflect more accurately the material living conditions of the rural classes. This proxy for income was based on the number of durable goods each individual possessed: automobile, television, video cassette player, refrigerator, gas stove, and water heater. Four dichotomous variables were constructed: one, if an individual had all six items; two, if it had 4-5 (television, gas stove, refrigerator video cassette player and/or automobile); three, if it had between 2-3 (television, gas stove and/or refrigerator); and four, the reference category, if it had 0-1 (television).
Based on possible combinations of these categories, three categories shown in Table 2 were constructed for the agricultural labor force. Those that owned land, had 10 or more years of schooling, and owned all 6 goods were classified in the high-income rural class. Land owners, salaried workers and self-employed that had between 6 and 9 years of schooling and owned 3 to 5 goods were included in the middle-income rural class. Finally, all those that did not own land, had 5 or less years of schooling, and owned 2 or fewer goods were included in the low-income rural class.
Table 3
Agricultural social class operationalization
Although the choice of social class measures was limited by the availability of data, the selection criteria to assign each individual to a particular social class category were not arbitrary. They were both theoretically and empirically justified. From a theoretical point of view, they are based on the theoretical framework formulated for this study. Empirically, social class was measured following a stratification approach for data availability reasons. The reliability of these variables was further tested comparing it with two other data sources: the 1995 Mexican Census Mid-decade Count (INEGI, 1997) and the 1994 Income and Expenditure Household Survey (EIGH, 1996).
The first source provided the information on occupation and education, while the second provided income information as well. The social class categories from ENSA II and these two other sources were not substantially different.
INDIVIDUAL-LEVEL CONTROL VARIABLES
Other variables that have been shown to be associated with morbidity are included as control variables. Age was transformed to correct for its skewed distribution. Except for age, which is a continuous variable, all others are dichotomous variables. Age showed a highly positive skewed distribution. To reduce this skewness a power of -0.1 was used. Gender is equal to 1 if the individual is male. Housing sanitary conditions are defined as good and equal to 1 if the dwelling where the individual lives had both piped water and sanitary disposal facilities and 0 otherwise. Given the high non-response rate for income, a proxy for income was used based on the number of durable goods each individual possessed as defined earlier. The reliability of this variable has been described elsewhere (Figueroa, 1996).
Health insurance is equal to 1 if an individual had any health insurance and 0 otherwise.
AGGREGATED-LEVEL CONTROL VARIABLES
Dummy variables were created to identify the region where an individual lived. For survey purposes, as explained earlier, Mexico was divided in five regions: North, Center, Metropolitan Areas of Mexico City, South and PASSPA (Health Services Aid Program for the Uninsured Population).
PASSPA is particularly important from a policy point of view because it was explicitly defined by the Mexican government (CONAPO, 1998) as an underserved region in terms of health services. The North region was the reference category.
The area-based index of marginality is a policy-oriented indicator developed by the Mexican government that measures the lack of basic public infrastructure as well as education and material living conditions at the state and county levels. More specifically, this inequality indicator measures the percentage of people over 15 years who are illiterate; the percentage of people without complete basic education; the percentage of the employed labor force earning less than twice the minimum wage (6 US$ per day); the percentage of people living in households in a town with less than 2,500 inhabitants without running water, without electricity, without sewage facilities, without solid floor materials and overcrowding living conditions. Principal component statistical analysis was performed to construct this area-based indicator of relative deprivation (CONAPO, 1998).
The marginality index is a normalized Z-score ranging between -3 and 3 standard deviations that correspond respectively to very low and very high marginality (CONAPO, 1998).
This index, however, was transformed to an indicator variable for interpretation purposes. Three categories were created based on CONAPO definitions: high marginality, whose values ranged from 2.36 to 0.39; medium marginality ranged from 0.13 to -0.84; while low marginality ranged from -0.85 to -1.74 (CONAPO, 1998).
RESULTS OF INDIVIDUAL LEVEL ANALYSIS
Multiple logistic regressions models were employed to examine the gradient effects of urban and rural social class on self-assessed health and reported morbidity.[ All multiple logistic regressions models were analyzed using STATA Statistical Software: Release 6.0 (StataCorp, 1999).] Four models were specified for each morbidity indicator and for both urban and rural social classes. The full model examines the health effects of social class adjusting for age, gender, housing sanitary conditions, income, region and region marginality. The second model examined the relationship between social class and morbidity excluding income. The third model excluded marginality, while the fourth excluded both income and marginality. In sum, sixteen models were tested.
Table 5 shows the multivariate odds ratios for the effects of urban social class on self-assessed health. All models show a gradient effect of social class relative to the low urban class, the reference category. The odds ratio for high urban class indicates a decreased risk for self-assessed poor health relative to the middle urban class and especially relative to the low urban class in all models, ranging from 57 to 67 percent for the high class. The decreased risk for middle high class relative to low urban class ranges from 55 to 63 percent, while for the middle low class it ranges from 35 to 38 percent. All are statistically significant. However, some of the other variables specified in the models seem to modify this association, specially the number of goods variable.
Table 5
Characteristics of ENSA II by individuals reporting a specific health problem
Variable N (%) N (%) reporting a specific health problem
Male
Female
12304 (48.48)
13078 (51.52)
1198 (7.52)
1706 (10.29)
Age
13-15
16-30
31-59
60-90
2534 (9.98) 10568 (41.63)
9589 (37.78)
2692 (10.61)
135 (5.33)
657 (6.22)
1120 (11.68)
564 (20.95)
High urban class
Middle high
Middle low
Low urban class
631 (4.16)
2500 (16.47)
3667 (24.16)
8381 (55.21)
22 (5.49)
107 (7.17)
154 (6.92)
734 (15.49)
High rural class
Middle class
Low rural class
234 (4.86)
952 (19.76)
3633 (75.39)
2 (5.56)
21 (9.42)
88 (10.24)
High marginality
Medium
Low marginality
10259 (41.6)
9101 (36.9)
5303 (21.5)
1119 (8.88)
1024 (8.82)
570 (8.81)
Bad housing conditions
Good housing conditions
13242 (52.17)
12141 (47.83)
1644 (9.21)
1260 (8.59)
North
Center
DF
PASSPA
South
4905 (19.32)
5225 (20.58)
5139 (20.25)
4887 (19.25)
5227 (20.59)
629 (10.13)
668 (9.85)
437 (7.03)
582 (8.95)
588 (8.65)
Any health insurance
No insurance
10287 (40.53)
15096 (59.47)
2249 (9.48)
3339 (8.83)
Goods
0-1
2-3
4-5
6
7049 (31.53)
5934 (26.54)
5357 (23.96)
4018 (17.97)
889 (9.15)
811 (10.61)
575 (8.58)
312 (6.46)
This proxy for income seems to be attenuating the effects of class except for the middle low class category, whose odds ratios remain stable in all models. The odds ratios of having fewer goods indicate an increased risk relative to having all six goods, but without a gradient effect. Individuals with one or fewer goods present less increased risk of perceiving their health as poor than do individuals that own two or three goods. Additionally, age and gender seem to have an effect of their own. Age has a strong positive association with a poor-assessed health. The odds ratios of gender indicate that men have about a ten percent decreased risk of perceiving their health as poor compared to women.
Table 6
Urban social class odds ratios for self-assessed health
Model 1 Model 2 Model 3 Model 4
Variables Odds ratio
95 CI Odds ratio
95 CI Odds ratio
95 CI Odds ratio
95 CI
High class 0.43
(0.33-0.57) 0.43
(0.33-0.56) 0.33
(0.37-0.41) 0.33
(0.27-0.42)
Middle high 0.45
(0.38-0.53) 0.44
(0.38-0.52) 0.37
(0.32-0.42) 0.39
(0.34-0.44)
Middle low 0.65
(0.57-0.74) 0.62
(0.55-0.71) 0.59
(0.52-0.66) 0.61
(0.54-0.68)
Age 3.31
(2.84-3.9) 3.16
(2.71-3.7) 3.08
(2.67-3.55) 3.10
(2.70-3.57)
Gender 0.88
(0.81-0.95) 0.88
(0.81-0.96) 0.91
(0.85-0.98) 0.91
(0.84-0.98)
Housing 0.78
(0.70-0.87) 0.79
(0.71-0.88) 0.72
(0.66-0.79) 0.72
(0.66-0.79)
DF 1.65
(1.43-1.90) 1.65
(1.43-1.90) 1.52
(1.34-1.72)
Center 1.36
(1.19-1.57) 1.53
(1.35-1.74) 1.47
(1.31-1.66)
PASSPA 1.08
(0.88-1.32) 1.19
(1.05-1.35) 1.19
(1.06-1.35)
South 1.02
(0.85-1.23) 1.13
(0.99-1.35) 1.13
(1.004-1.27)
0-1 goods 1.60
(1.34-1.91) 1.69
(1.43-2.01)
2-3 goods 1.75
(1.49-2.06) 1.84
(1.57-2.15)
4-5 goods 1.56
(1.34-1.82) 1.65
(1.43-1.92)
Low marginality 0.83
(0.67-1.02)
Medium marginality 0.95
(0.81-1.14)
Reference categories: Low class, female, low marginality, North, six goods, bad housing
Model 1 = Full model
Model 2 = Full model minus marginality
Model 3 = Full model minus marginality and income
Model 4 = Full model minus marginality, income and region
The odds ratios of the lower marginality levels show a decreased risk compared to the high level of marginality, the reference category. The odds ratio of individuals living in a low marginality state indicate about a 20 percent decreased risk than those living in high marginality states. The effect of medium marginality level, however, was not statistically significant. Individuals living in the Metropolitan area of Mexico City (DF) and the Center region seem to have an increased risk of perceiving their health as poor relative to the North. Individuals living in the PASSPA and the South regions showed a slightly higher increased risk.
Table 6 shows the multivariate odds ratios for the effects of agricultural social class on self-assessed health. The social class indicator variables in the agricultural labor force do not seem to influence self-assessed health when income is adjusted for. When income is not included (Models 3 and 4), however, the odds ratios of low class indicate an approximately 50 percent-increased risk compared to the high class, the reference category. Both age and gender are statistically significant. The odds ratios of age indicate an increased risk as age progresses, while men seem to have an approximately thirty percent-decreased risk compared to women.
The level of marginality modifies the effect of social class in the agricultural sector. When marginality is not included (Models 2 and 3) social class seems to have a smaller effect on self-assessed health. This suggests a collinearity problem between these two variables. However, marginality seems to have an effect of its own even when income is adjusted for (Model 1).
The odds ratios of high marginality (Model 1) indicate that individuals living in high marginality states are two times more likely to perceive their health as poor compared to individuals living in low marginality states. Individuals living in the Center, the South and the PASSPA regions are more likely to perceive their health as poor compared to the North region. The influence of region, however, seems to be attenuated by the level of marginality. In fact, PASSPA is mostly a high marginality region. The odds ratios of living in the Center and in the South regions are higher when marginality is not included (Model 2 and 3).
Table 7
Agricultural social class odds ratios for self-assessed health
Model 1 Model 2 Model 3 Model 4
Variables Odds ratio
95 CI Odds ratio 95 CI Odds ratio
95 CI Odds ratio
95 CI
Low class 1.25
(0.74-2.11) 1.08
(0.65-1.81) 1.42
(0.97-2.08) 1.52
(1.04-2.22)
Middle class 1.26
(.80-1.99) 1.11
(0.71-1.74) 1.37
(0.92-2.04) 1.28
(0.86-1.92)
Age 3.27
(2.70-3.96) 3.13
(2.60-3.78) 3.17
(2.63-3.82) 3.22
(2.68-3.88)
Gender 0.64
(0.48-0.86) 0.67
(0.50-0.90) 0.69
(0.51-0.93) 0.69
(0.51-0.93)
DF 1.33
(0.58-3.06) 1.58
(0.69-3.62) 1.46
(0.69-3.09)
Center 1.44
(1.19-1.74) 2.07
(1.70-2.53) 1.96
(1.61-2.38)
South 1.01
(0.87-1.18) 1.61
(1.32-1.96) 1.56
(1.29-1.89)
PASSPA 1.88
(1.56-2.28) 1.85
(1.53-2.22)
0-1 goods 2.23
(1.28-3.91) 2.79
(1.62-4.78)
2-3 goods 2.40
(1.40-4.11) 3.07
(1.82-5.17)
4-5 goods 2.26
(1.33-3.82) 2.82
(1.70-4.71)
High marginality 2.04
(1.54-2.71)
Medium marginality 1.77
(1.32-2.4)
Reference categories: High class, Female, North, 6 goods, Low marginality
Model 1 = Full model
Model 2 = Full model – marginality
Model 3 = Full model – marginality and income
Model 4 = Full model – marginality, income and region
Table 7 shows the multivariate odds ratios for the effects of urban social class on reported morbidity. Social class seems to have a similar gradient effect on the reporting of more specific health problems as it did for self-assessed health. However, the odds ratios of social class lose statistical significance in some of the models, particularly for the middle-low indicator variable. The odds ratio of high-class individuals indicates about a 40 percent decreased risk compared to low class individuals in the fully adjusted model. However, when income is not included, individuals in the high class show a decreased risk of little less than 60 percent relative to the low class individuals. Although income seems to be attenuating the effects of class on reported morbidity, class remains statistically significant.
Table 8
Urban social class odds ratios for specific reported health problems
Model 1 Model 2 Model 3 Model 4
Variables Odds ratio
95 CI Odds ratio 95 CI Odds ratio
95 CI Odds ratio
95 CI
High class 0.56
(0.36-0.88) 0.58
(0.37-0.89) 0.43
(0.29-0.63) 0.43
(0.29-0.63)
Middle high 0.73
(0.56-0.94) 0.71
(0.55-0.92) 0.59
(0.48-0.72) 0.57
(0.47-0.70)
Middle low 0.82
(0.66-1.02) 0.81
(0.66-1.001) 0.69
(0.57-0.83) 0.68
(0.57-0.82)
Age 4.26
(3.31-5.47) 4.24
(3.32-5.41) 3.9
(3.11-4.9) 3.92
(3.12-4.91)
Gender 0.72
(0.64-0.81) 0.71
(0.63-0.80) 0.73
(0.65-0.81) 0.74
(0.66-0.82)
Health insurance 1.19
(1.03-1.37) 1.16
(1.01-1.33) 1.10
(0.97-1.25) 1.13
(0.99-1.28)
DF 0.76
(0.61-0.94) 0.77
(0.62-0.95) 0.73
(0.61-0.88)
Center 0.94
(0.77-1.16) 0.92
(0.77-1.10) 0.88
(0.75-1.04)
PASSPA 0.89
(0.66-1.20) 0.81
(0.68-0.97) 0.84
(0.71-0.98)
South 0.78
(0.59-1.03) 0.71
(0.60-0.85) 0.73
(0.61-0.86)
0-1 goods 1.73
(1.34-2.22) 1.63
(1.27-2.09)
2-3 goods 1.92
(1.51-2.45) 1.87
(1.48-2.38)
4-5 goods 1.41
(1.10-1.79) 1.41
(1.11-1.79)
Low marginality 1.21
(0.89-1.64)
Medium marginality 1.01
(0.79-1.29)
Model 1 = Full model
Model 2 = Full model minus marginality
Model 3 = Full model minus marginality and income
Model 4 = Full model minus marginality, income and region
Individuals living in Mexico City Metropolitan Area (DF) seem to have an approximately 20 percent-decreased risk compared to individuals living in the North region. Individuals with more goods seem to be less likely to report health problems than individuals with fewer goods. Individuals with health insurance seem to be more likely to report health problems than those without insurance, although the odds ratios are not statistically significant.
Table 9
Agricultural social class odds ratios for specific reported health problems
Model 1 Model 2 Model 3 Model 4
Variables Odds ratio
95 CI Odds ratio 95 CI Odds ratio
95 CI Odds ratio
95 CI
Low class 1.02
(0.38-2.74) 1.05
(0.40-2.72) 1.40
(0.65-2.98) 1.35
(0.64-2.84)
Middle class 1.14
(0.46-2.82) 1.26
(0.52-3.03) 1.58
(0.73-343) 1.60
(0.74-3.46)
Age 4.31
(3.00-6.2) 3.91
(2.74-5.59) 3.9
(2.74-5.55) 3.86
(2.73-5.45)
Gender 0.75
(0.46-1.23) 0.78
(0.48-1.27) 0.78
(0.48-1.25) 0.78
(0.49-1.26)
Health insurance 0.88
(0.63-1.25) 0.88
(0.63-1.23) 0.87
(0.63-1.21) 0.90
(0.65-1.23)
DF 1.35
(0.41-4.48) 1.33
(0.40-4.37) 1.33
(0.45-3.90)
Center 0.84
(0.60-1.19) 0.83
(0.59-1.16) 0.79
(0.57-1.10)
South 0.71
(0.54-0.94) 0.71
(0.51-1.01) 0.71
(0.51-0.99
PASSPA 1.005
(0.76-1.39) 1.004
(0.73-1.37) 1.001
(0.74-1.35)
0-1 goods 3.16
(0.90-11.01) 3.86
(1.15-12.96)
2-3 goods 3.43
(1.01-11.66) 3.99
(1.23-12.97)
4-5 goods 3.68
(1.13-11.97) 4.06
(1.30-12.69)
High marginality 0.89
(0.59-1.36)
Medium marginality 0.77
(0.49-1.21)
Reference categories: High class, female, no insurance, North, low marginality
Model 1: Full model
Model 2: Full model – marginality
Model 3: Full model – marginality and income
Model 4: Full model – marginality and income and region
Table 8 shows the multivariate odds ratios for the effects of agricultural social class on reported health problems. Income, assessed as numbers of goods individuals have, seems to be more strongly associated with this indicator of morbidity than social class here. However, the income gradient compared to the highest income category is not as expected. The odds ratios of individuals owning more 4-5 goods indicate a higher percentage risk of having a health problem than individuals owning one or less goods. Age again is strongly associated with this health indicator. The odds ratios of gender indicate that men are at a 20 percent decreased risk of reporting a health problem compared to women. Neither the level of marginality nor the region seems to have a statistically significant effect on whether individuals report a health problem.
INDIVIDUAL LEVEL DISCUSSION
Overall, the findings of this study provide empirical evidence that social inequality has a negative impact on health at the individual level. The lower the social class, the poorer the perception of health. The results also suggest, as hypothesized, that social inequality may be influencing health through both material and psychosocial mechanisms, as well as at different levels of aggregation. In this study, two self-reported health outcomes were chosen and a different social class specification was made for the urban and the rural labor force. As expected, there were some differences between the two health outcomes as well as the two social class specifications.
Self-assessed health showed a strong association with social class, especially in the urban labor force. This finding is consistent with previous studies that have shown that self-assessed health is highly sensitive to social class differentials (Marmot et al, 1991) and income inequality (Kawachi et al, 1999) in industrialized societies. The social gradient effect on health reflects a combination of negative exposures and lack of resources held by individuals in lower social positions (Lynch and Kaplan, 2000).
The other morbidity indicator, which indicated more specific and less subjective health problems such as diabetes and respiratory problems, however, showed a weaker association. This empirical difference suggests that social inequality may be operating through psychological mechanisms as well. Self-assessed health may be manifesting the psychosocial consequences of social inequality. In fact, poor self-assessed health has been found to be strongly associated with stress that is detrimental to health (Krause, 1987; Levkoff et al, 1987; Farmer and Ferraro, 1997).
Unfortunately, no direct measure of psychological status was available in the ENSA II survey. Thus, this pathway should be further explored with more direct empirical evidence.
The social gradient effects of social class in the agricultural sector for both self-reported health outcomes were not as clear as the urban sector. Furthermore, they were not statistically significant when income, assessed as the number of goods individuals owned was adjusted for. This in part may be due to a methodological issue that has to do with including the proxy for income in the operationalization of rural social class. However, these findings are more meaningful if they are interpreted in their Mexican context. In Mexico, most individuals that earn their living in agricultural-related economic activities or live under poverty conditions belong to a low social class. ENSA II data show that almost 80 percent of the rural sampled population belong to this category, while other studies figures range from 89 to 85 percent (Infante and Schlaepfer, 1994; Boltvinik, 1999).
On the other hand, high-class individuals are usually underrepresented in this kind of surveys. The fact that most people belong to a low social class in the agriculture sector and that the high class is underrepresented indicates that there is less social class variation that is reflected in a less evident social gradient effect on the two morbidity indicators. Furthermore, studies have shown that health status underreporting is higher in both rural areas and where poor people live (Infante and Schlaepfer, 1994).
This might be the case for the morbidity indicator that identifies more specific health problems such as diabetes and hypertension, illnesses that are more difficult to self-report if a medical diagnosis, which is more objective, is not made. A study using ENSA II reported that there could be some underestimation of reported morbidity (Figueroa, 1996).
This last indicator that included health problems and accidents did not show any statistical significance even for the model that only adjusted for the effects of age and gender.
Although the social class differentials show the gradient effects of inequality at the individual level, the results of the analysis suggest that inequality may be operating at aggregate levels as well. Regional differentials reproduced the urban and agricultural social class division. Living in mostly urban regions of the Metropolitan Area of Mexico City (DF) and the Center was statistically significant for the urban social class, while PASSPA and the South, which are mostly rural regions, were significant for the rural social class. PASSPA in particular is a region specifically defined to target the most underserved geographic areas in terms of public health infrastructure and medical care.
The influence of the level of marginality was significant in the agricultural sector. The results of this analysis show that individuals living in both high and medium marginality states were more likely to report their health as poor than individuals living in low marginality states in rural areas. This finding is consistent with many studies that have shown that health status is poorer in deprived areas (Reijneveld et al, 2000).
In Mexico, higher levels of marginality exist in rural areas. Furthermore, marginality is a policy-oriented indicator that measures the lack of basic public infrastructure including electricity, sewage, education and material living conditions (CONAPO, 1988).
Thus, ecological characteristics seem to have an influence on health beyond the effects of social class and income.
LIMITATIONS
The cross-sectional character of the data poses some limitations to the interpretation of the results. This type of data is inadequate for fully elucidating the inequality effects on health over the course of a lifetime. However, longitudinal studies have shown that self-assessed health has a considerable predictive validity of mortality (Farmer and Ferraro, 1997; Idler et al, 1996) thus providing some insights into how it is that health declines over the life course. Furthermore, there is empirical evidence suggesting that social class differences in morbidity persist through long periods of time. The findings of the second Whitehall study of British civil servants showed no reduction in the social gradient in morbidity with a new cohort of civil servants compared to the first Whitehall study conducted 20 years earlier (Marmot et al, 1991).
The nature of the data available from ENSA II constrained the operationalization of social class. Although a theoretical framework guided the construction of social class categories, to some extent theory had to be sacrificed in order to obtain empirically testable categories. For example, to better distinguish between middle and high-class categories required information on ownership of the means of production, authority over others in the workplace, and skill exercised in the job. However, none of these characteristics were available. Instead, level of education was used to sort individuals into the different social classes.
Income is another variable that limited the analysis because the non-response rate was very high. A proxy for income, whose validity was tested in another study using ENSA II, (Figueroa, 1996) was employed instead. Although this indicator measures relative deprivation rather than income itself, it proved useful to account for its mediating effects in the association between social class and perceived health.
PART II
POLICY EFFECTS OF SOCIAL INEQUALITY
The purpose of this second part of the study is to examine the association between social inequality and population health at the state level in Mexico from 1990 to 1997. It also explores the ecological pathways through which social inequality influences health at this level of aggregation. Studying the social, economic and political features of an area may help to better understand the relationship between social inequality and health at aggregated levels of analysis (Macintyre and Ellaway, 2000; Soobader and Leclere, 1999).
However, most ecological studies have failed to explicitly define the geopolitical dimension of inequality (Soobader and LeClere, 2001; Macintyre and Ellaway, 2000; Soobader and LeClere, 1999; Malmstrom et al, 1999).
Thus, this second part of the study sought to contribute in three ways to this less explored and theoretically underdeveloped research field in the social inequality and health literature.
General estimation equation (GEE) models were specified to account for the pooled cross-sectional type of data used for the analysis. Path analysis was also conducted to examine the political pathways through which social inequality affect health via distribution of health care resources.This second part of the study empirically examined the association between social inequality and population health using pooled cross-sectional data. Secondary analysis was conducted for the 32 Mexican states for 1990 and 1996 in three ways. First, partial correlation coefficients were calculated to explore the association between social inequality and three population health indicators as well as other influencing factors. Correlations between the social spending and its potential determinants, including political factors are estimated as part of this exploratory analysis. Second, pooled cross-section analysis were used to analyze the association between the three specified health indicators, social inequality, and health spending controlling for the effects of other relevant variables. Also, the association between three health spending indicators, social inequality and potential mediating variables was examined. Third, path analysis was performed to explore the electoral pathways through which social inequality negatively influences population health.
STATE LEVEL METHODOLOGY
Several sources provide the data for this state level analysis. The main source of the state health indicators was from FUNSALUD, a Mexican private non-profit research institution. The data, which provide information from 1990 to 1996, originally come from the Census records (INEGI, 1990 and 1997) and from the National Population Council (CONAPO, 1998).
These outcome variables are maternal and infant mortality rates as well as years of life lost for all causes. The mortality rates were classified according the 9th International Disease Classification (ICD-9).
These rates were standardized for age from the Census records. Corrections for under registration were made for adults over 20 and those over 60 (FUNSALUD, 1998).
Public health expenditures and primary care physicians come from the Ministry of Health (Secretaría de Salud, 1999).
Measures of social inequality come from the National Population Council (CONAPO, 1998) and the National Institute of Geography and Statistics (INEGI, 1997).
The marginality index was obtained from the National Population Council (CONAPO, 1998).
The income inequality indicators come from the National Surveys of Income and Expenditure (INEGI, 2000; 1998; 1994).
Finally, the Federal Electoral Institute (IFE, 1998) was the source of the electoral strength of the Mexican political parties in the 1991 and 1997 federal elections.
POPULATION HEALTH INDICATORS
The most common population health indicators used in previous studies are average life expectancy at birth (Wilkinson, 1992; Duleep, 1995; Judge et al, 1998; Wilson & Daly, 1997) and infant mortality (Rodgers, 1979; Flegg, 1982; Bronfman, 1992; Wennemo, 1993).
In addition to infant and maternal mortality rates, years of life lost for all causes (YLL) were employed as health indicators. Maternal mortality rates were expressed as deaths of women in reproductive age per ten thousand live births, while infant mortality rates were expressed as number of child deaths per thousand liver births. Preliminary data analysis suggested that these health indicators needed some transformation to specify the correct functional form, as all of them were positively skewed. Logarithmic transformations were used to address this, except for homicide rates, which required a higher power transformation (i.e. 0.3).
STATE LEVEL SOCIAL INEQUALITY
Social inequality was measured with the index of marginality. This index was developed in Mexico to measure the degree of marginality at the state and county level. The index is an indicator of deprivation based on housing, income, and education information collected from the 1990 Mexican Census and the 1995 Population and Housing Count (INEGI, 1995).
The housing component refers to the percentage of people living in households in a town of less than 2,500 inhabitants, lacking running water, electricity, solid floor materials and sewage facilities as well as overcrowded living conditions. Education measures the percentage of illiterate people older than 15 years and percentage of people who did not finished the six years of basic education. The income component refers to the percentage of the economic active labor force earning less than twice the minimum wage, which is equivalent to 6 US dollars per day. CONAPO performed principal component statistical analysis to assess its validity (CONAPO, 1998).
This index is a normalized Z-score ranging between -3 and 3 standard deviations that correspond to very low and very high marginality respectively. Interaction effects between the marginality index and several independent variables were expected to influence both health outcomes and the social policy design indicators, particularly with gross domestic product per capita, average years of schooling and the political strength indicators.
SOCIAL POLICY INDICATORS
Public health care was chosen in this study as the social policy arena to explore its mediating effects between health, social inequality, and electoral strength of the main Mexican opposition parties. In the health sector, both targeted and non-targeted spending indicators were specified. Non-targeted health care spending was measured as total state public health care expenditures per capita. Targeted health care spending was measured as state Solidarity health care expenditures per capita. These variables presented a positively skewed distribution, which was corrected with a logarithmic transformation. These indicators, however, only measure how much was spent on health. Increasing relative health expenditures does not necessarily imply a greater access to primary care, nor they indicate how resources are allocated. Empirical studies suggest complementing these social expenditure data with more direct measures of social benefits (Korpi, 1989; Wennemo, 1993).
Thus, primary care was additionally measured as number of non-specialist physicians per thousand. Both theoretical and empirical evidence exist to support the use of this indicator to examine its association with both population health and social inequality at an ecological level (Starfield, 1994; Shi et al, 1999; Shi and Starfield, 2000).
POLITICAL STRENGTH INDICATORS
Empirical attempts to measure to what extent political factors shape social policy are generally focused on the correlation between partisan government incumbency and policy output (Korpi, 1989).
Here the number of votes against the ruling party, PRI, was used as a measure of the relative strength of opposition parties. Although a better indicator would consider the distribution of important power resources among major interest groups or classes in society, the lack of relevant empirical data limited the analysis to electoral outcomes as indicators of the degree of discontent with government policies. Electoral strength of the political parties was measured as the percentage of votes received in the 1991 and the 1997 federal elections by the two main opposition political parties, PAN, a right-center party, and PRD, a left-wing party. The electoral strength of PAN and PRD in state elections held during this period will also be measured to examine if these elections influence social investment as well. These indicators also required logarithmic transformations to correct for their positively skewed distribution.
STATE LEVEL CONTROL VARIABLES
Finally, other relevant variables were included to control for their mediating effects. These control variables included the income per capita for each state measured as the total gross domestic product (GDP) per state in 1993 thousand pesos divided by its total population. This variable was included in the GEE models because it is an indicator of state economic development. This degree of development is an important factor in determining public resources allocation decisions in Mexico and has been used in empirical studies of public expenditures determinants (Molinar and Weldon, 1994; Gershberg, 1994; Díaz Cayeros, 1998; Morgenstern, 1997).
GDP presented a positively skewed distribution. A logarithmic transformation corrected this statistical problem. This transformation is usually used in quantitative analysis to address the nonlinear relationship between income-related indicators and health (Ecob and Davey Smith, 1999; McDonough et al, 1997; Kaplan et al, 1996; Flegg, 1982).
Maternal mortality rates were also included as an indicator of health care needs.
STATE LEVEL RESULTS
Pooled cross-section analyses were used to analyze two associations. First, the association between population health and social inequality is analyzed. Maternal and infant mortality as well as years of life lost (YLL) were specified to explore how marginality, GDP per capita, and public health care expenditures were associated with these health indicators. Three public health care expenditures were specified: solidarity expenditures, total public health expenditures, and primary care physicians per capita. All of these were included in the analysis to test for their differential sensitivity to social inequality and the electoral strength of opposition political parties. Second, to examine the association between health policy spending, marginality, and the electoral strength of PAN and PRD, the three same health spending indicators specified in the first set of GEE models were used here. GDP per capita and maternal mortality rates were also included in the models to account for its influence on health spending allocating decisions.
Pool data are often used in comparative quantitative studies of the political economy of democratic welfare countries (Hicks, 1994).
Less examined, though, is the association between social spending and political factors within countries (Díaz Cayeros, 1998; Morgenstern, 1997).
In this ecological study, two years, 1990 and 1996 (t=2) as well as thirty-two cross-sections (n=32 states) are pooled to allow for a larger sample analyses based on temporal and cross-state variation. However, if ordinary least square (OLS) equations are used, errors tend to be heterocedastic and autocorrelated within states, as well as autocorrelated across states (Hicks, 1996; Hicks and Swank, 1992).
To address these statistical problems analysis were performed using general estimation equation (GEE) models to analyze data for the 32 states and for two years, 1990 and 1996. Analytical weights were employed to correct for potential heteroskedasticity and autocorrelation problems as well as to account for the population size of each state. To control for the effects of changes over time a year dummy variable is used. This year dummy is equal to 1 if the year is 1996 and 0 if 1996.
MARGINALITY AND POPULATION HEALTH
Table 4 presents the GEE coefficients between marginality, total public health expenditures, and the three specified health indicators. Only the models with the best fit were shown.[ All the models specified for the state level analysis were included in the appendix section. ]
Table 4
GEE regression coefficients of marginality and total public health expenditures on health outcomes
VARIABLES MATERN
(Std dev)
z YLL
(Std dev)
z INFANT
(Std dev)
z
MARGINALITY
INDEX 0.94
(0.36)
2.63 0.52
(0.038)
1.34 0.36
(0.60)
0.604
PUBLIC HEALTH EXPENDITURES -0.49
(0.24)
-1.99 -0.26
(0.05)
-5.67 -1.95
(0.57)
-3.43
YEAR 0.37
(0.32)
1.16 0.14
(0.28)
2.34 -1.98
(0.69)
-2.85
CONSTANT
3.2
(1.3)
2.42 9.77
0.28
34.56 36.76
(2.9)
12.65
GDP 0.07
(0.42)
0.17 0.29
(0.95)
3.07
GDP*MARGINALITY -0.31
(0.18)
-1.78
WALD STAT 37.33 329.88 906.45
The marginality index coefficients show both a positive and statistically significant association only with maternal mortality, adjusting for the effects of gross domestic product (GDP), the public spending indicator, and the time variable. As marginality increases, population health gets worse across states. The coefficients were small, though. Furthermore, maternal mortality was the health indicator that showed the strongest association. Total public health expenditures per capita are on average negatively and significantly associated with years of potential life lost (YPLL) and infant mortality adjusting for the effects of the other variables. These coefficients suggest as well that infant mortality is the most sensitive indicator to total public health expenditures.
The GDP coefficient is only statistically significant for the years of life lost, showing a positive association. The interaction term between GDP and marginality indicates that this association is stronger for the low-marginality states. Finally, the year coefficients indicate whether the health indicators were on average lower or higher between the two years of analysis. Both years of potential life lost and maternal mortality show an increasing trend between 1990 and 1996. However, only YPLL was statistically significant. Infant mortality, on the other hand, presented a decreasing and statistically significant trend comparing these same years.
Table 5 shows the GEE coefficients of marginality, Solidarity health care expenditures per capita, and the three specified health indicators. The marginality index coefficients showed a positive and statistically significant association only for maternal mortality, adjusting for the effects of the other variables including GDP per capita. The coefficients of the Solidarity health care expenditures are both quite small and statistically not significant. The Solidarity program was not implemented in all the 32 Mexican states. It was limited to 19 states. This small number may have had accounted for these insignificant results.
Table 5
GEE regression coefficients of marginality and Solidarity expenditures on health outcomes
VARIABLES MATERN
(Std dev)
z YLL
(Std dev)
z INFANT
(Std dev)
z
MARGINALITY 1.31
(0.43)
3.04
1.02
(0.96)
1.07
SOLIDARITY -0.012
(0.07)
-1.17 0.001
(0.90)
0.12 0.10
(0.58)
0.17
YEAR -0.11
(0.12)
-0.89 -0.18
(0.01)
-13.81
CONSTANT
-0.97
1.19
-0.811 9.17
0.17
53.92
GDP PER CAPITA 0.74
(0.49)
1.52 0.01
(0.07)
0.16
GDP*MARG -0.28
(0.192)
-1.48
WALD STATISTIC 35.08 332.63 1.28
The GDP coefficient and its interactive term with marginality are not statistically significant for the maternal mortality and years of life lost models. Finally, the year coefficients indicated a small but statistically significant YLL decrease from 1990 to 1996.
Table 6 presents the GEE coefficients of marginality, primary care physicians per capita, and the three specified health indicators. The marginality index coefficients show both a positive and statistically significant association with all health indicators at the 0.05 level of confidence, except for the infant mortality rates, which is statistically significant only at the 0.10 level of confidence. These results indicate that the higher the state marginality, the higher the mortality rates and the years of life lost.
Table 6
GEE regression coefficients of marginality and general physicians on health outcomes
VARIABLES MATERN
(Std dev)
z YLL
(Std dev)
z INFANT
(Std dev)
z
MARGINALITY 0.99
(0.33)
3.03 0.09
(0.04)
2.21 0.69
(0.58)
1.195
PRIMARY CARE PHYSICIANS -1.06
(0.37)
-2.97 -0.31
(0.08)
-3.87 -3.24
(1.06)
-3.05
YEAR 0.03
(0.13)
0.23 -0.10
(0.03)
-3.83 -3.74
(0.23)
-15.8
CONSTANT -1.58
(1.39)
-1.13 7.81
0.33
23.7 23.03
1.54
14.97
GDP 0.49
(0.45)
1.09 0.39
(0.10)
3.8
GDP*MARG -0.28
(0.16)
-1.78
WALD STATISTIC 45.14 397.3 883.74
Primary care physician population ratios are negatively and statistically significantly associated with maternal and infant mortality as well as with years of life lost. The GEE coefficients indicated a stronger association with the maternal and infant mortality rates. The GDP coefficient is only statistically significant for the years of life lost. The interaction term between GDP and marginality indicates that this association is stronger for the low-marginality states. Finally, the year coefficients indicate that both years of potential life lost and infant mortality across states were on average lower in 1996 than in 1990.
POLITICAL STRENGTH AND HEALTH SPENDING
Table 7 shows the GEE coefficients of the strength of marginality, maternal mortality, GDP per capita as well as PAN and PRD, the Mexican political opposition parties on three indicators of health spending. In all three models, the coefficient of marginality presented a statistically significant association with the social spending indicators specified, controlling for the effects of GDP per capita, maternal mortality rate, year, and the political strength of PAN or PRD.
Table 7
GEE regression coefficients of political strength on health spending indicators
VARIABLES PHCE
(Std dev)
z PCPH
(Std dev)
z SOLIDARITY
(Std dev)
z
MARGINALITY 0.42
(0.14)
3.10 0.12
(0.04)
3.04 -0.98
(0.23)
-4.176
MATERNAL MORTALITY -0.10
(0.05)
-1.93 -0.08
(0.03)
-2.87 0.86
(0.45)
1.91
YEAR 1.19
(0.05)
25.05 0.25
(0.02)
13.63 -0.011
(0.39)
-0.09
GDP PER CAPITA 0.59
(0.15)
3.88 0.72
(0.09)
7.62
GDP*MARGINALITY -0.18
(0.06)
-2.88
PAN 0.13
(0.05)
2.77
PRD 0.75
(0.33)
2.27
WALD STATISTIC 1435.06 304.21 26.87
N 64 64 37
PHCE: Total public health care expenditures per capita
PCPH: Primary care physician population ratio
SOLIDARITY: Solidarity health care expenditures
Solidarity expenditures per capita, however, were negatively associated with marginality. This indicated that as marginality rose across states, these health care expenditures were on average lower. Furthermore, this type of health expenditures per capita seemed to be the most sensitive dependent variable to marginality variation across states.
Maternal mortality is negatively associated with total public health care expenditures and primary care physician population ratio, while Solidarity expenditures is positively associated to this health outcome. However, only the primary care indicator was statistically significant. The association with GDP per capita is also positive and statistically significant with total public health care expenditures and primary care population ratio. This indicated that on average, the higher the GDP per capita across states, the higher the health expenditures and the higher the primary care population ratio tended to be.
The political variables coefficients indicated a moderate yet significant influence on the health spending indicators. The PRD coefficient, on the one hand, shows a positive and statistically significant association with primary care physicians per capita adjusting for marginality, GDP per capita, maternal mortality rate, and year. Higher IMSS Solidarity expenditures seemed to be associated with higher PRD percentage of votes in across states. The electoral strength of PAN, on the other hand, was positively and statistically significant with total public health care expenditures per capita. Finally, except for the Solidarity expenditures, the year coefficient had a positive and statistically significant association with the other two social spending indicators. Both of them showed a decrease resources allocated from 1990 to 1996.
PATH ANALYSIS
Path analysis was performed to examine the electoral pathways through which social inequality influences population health. This type of analysis involves three steps. First, a theory driven model is defined based on the literature review and a theoretical framework (Shi, 1997).
The literature on the political determinants of public spending allocation (Korpi, 1989; Hicks and Swank, 1992) suggest that governments weigh electoral demands and support when making allocation decisions of expenditures and other public resources. Governments allocate public expenditures between geographical areas such as states taking into account electoral interests.
In the Mexican context, elections driven social spending has been empirically tested using both targeted spending indicators (Molinar and Weldon, 1994; Gershberg, 1994) and non-targeted spending (Díaz Cayeros, 1998; Morgenstern, 1997).
Their findings suggest that the electoral threat from the opposition, particularly from the PRD, influences social spending allocation decisions. Thus, it is hypothesized that targeted social spending will increase in states where PRD is stronger. In turn, states with higher targeted social spending will have worse health outcomes than states with lower social targeted spending.
Second, the path coefficients are estimated through regression equations (Frankfort-Nachmias and Nachmias, 1996).
These estimations assess the direct effects of the variables included in the model. GEE models were used as well to analyze data for the 32 states and for two years, 1990 and 1996. Two models were specified for each dependent variable: maternal and infant mortality (See Models below).
These health outcomes were chosen because they are the most sensitive indicators to the effects of health spending. To estimate the path coefficient between the percentage of votes received by the PRD and targeted spending, Solidarity health care expenditures are regressed on the votes received by PRD. Only the PRD is included in the analysis because the literature suggests that the electoral threat of left-wing parties such as PRD influence social spending (Korpi, 1989; Hicks and Swank, 1992).
The Mexican scholars suggest that targeted spending is the type of social spending that is more likely to be influenced by electoral considerations (Laurell, 199; Molinar and Weldon, 1994; Gershberg, 1994).
Thus, Solidarity health care spending is the social expenditure indicator used in the analysis. To measure more directly the health effects of expenditures, an indicator of primary care was added to the model. Both theoretical and empirical evidence exist to support the positive association between primary care and better health status at the ecological level (Starfield, 1994; Shi et al, 1999; Shi and Starfield, 2000).
Thus, it is hypothesized that states with a lower proportion of general practitioners will have worse health outcomes than states with a higher proportion of general practitioners. Each equation includes as many terms as there are arrows leading to the dependent variable. Thus, the following equations represent the political pathway models:
MODEL 1: MATERNAL MORTALITY
MATERNAL = P12*MARGINALITY + P13*PCPHY + P14*SOL
PCPHY = P32*MARGINALITY + P34*SOL
PRD = P52*MARGINALITY
SOL = P42*MARGINALITY + P45*PRD
PCPHY: Primary care physicians per capita in 1990 and 1996
PRD: Percentage of votes received by PRD in 1991 and 1997 federal elections
SOL: Solidarity health care expenditures per capita in 1992 and 1996
MATERNAL: State maternal mortality rate in 1990 and 1995
Pij are the path coefficients, i being the dependent variable and j the independent variable.
MODEL 2: INFANT MORTALITY
INFANT = P12*MARGINALITY + P13*PCPHY + P14*SOL
PCPHY = P32*MARGINALITY + P34*SOL
PRD = P52*MARGINALITY
SOL = P42*MARGINALITY + P45*PRD
PCPHY: Primary care physicians per capita in 1990 and 1996
PRD: Percentage of votes received by PRD in 1991 and 1997 federal elections
SOL: Solidarity health care expenditures per capita in 1992 and 1996
INFANT: State infant mortality rates in 1990 and 1995
Pij are the path coefficients, i being the dependent variable and j the independent variable.
To estimate the indirect effects of the social policy design and the electoral strength indicators, the values of the path coefficients that link two variables via intervening variables are multiplied (Frankfort-Nachmias and Nachmias, 1996).
Thus, to estimate the indirect effect of the PRD votes on health via primary health care, P52 is multiplied by P13. The same rationale is used for the second path model.[ The measures of both the dependent and independent variables that yield the more robust results in the previous multiple regression models will be used in the path models.]
Figure 1 shows the results of the path analysis examining how the effects of marginality on maternal mortality are mediated through the influence of the electoral strength of PRD on the distribution of Solidarity health care expenditures as well as the influence of marginality on infant mortality through its effects on primary care physicians per capita and the targeted health spending indicator. Both the marginality index and the primary care physician population ratio had a strong and statistically significant influence on maternal mortality, the former positively and the latter negatively. However, Solidarity health care expenditures did not show a statistically significant association with maternal mortality. The indirect effects of both Solidarity health care expenditures and the marginality index through their relationship with primary care physicians per capita were modest but statistically significant at the 0.5 level of confidence.
FIGURE 5
PATHWAYS TO MATERNAL MORTALITY
PRD had a positive indirect effect on maternal mortality of 0.06 through its influence on Solidarity health care expenditures, while the Solidarity expenditures had a stronger but negative indirect effect of -0.12 through its influence on primary care physician population ratio. The index of marginality presented a positive indirect effect through the primary care indicator of 0.36 and negative indirect effect through Solidarity expenditures of -0.07. The total influence of marginality (both direct and indirect) on maternal mortality was 0.86.
Figure 2 illustrates the extent that the electoral strength of PRD affects infant mortality through its impact on the distribution of IMSS Solidarity health care expenditures as well as the influence of marginality on infant mortality through its effects on primary care physicians per capita and the targeted health-spending indicator.
FIGURE 6
PATHWAYS TO INFANT MORTALITY
* (P