This study is part of a collaborative research program on inequalities in health and health care by the Centre for Development Studies and the University of Montreal in a single Gram Panchayat (the lowest territorial administrative unit) located in the northern district of Wayanad in Kerala [23
]. Although Scheduled Tribe members account for only 1% of the population in Kerala State, there are clusters of tribal groups in some areas. The Panchayat in this study has a population of 16,110 of whom 28% have Scheduled Tribe affiliations. Three out of five Scheduled Tribe members in the Panchayat are Paniyas, who make up one-fifth of Kerala’s Scheduled Tribe population [24
]. Historically, Paniyas were bonded labourers. Today they are predominantly landless agricultural labourers who tend to migrate for employment opportunities. Paniyas also face social marginalisation, high levels of impoverishment, and poor access to health care [25
]. Small numbers of other tribal groups who have living standards similar to the general population are also found in the Panchayat. The Kuruchian constitutes the large majority of “other Scheduled Tribes” in this area (94% of non-Paniyas).
We draw on data from a health survey approved by the ethics committee of University of Montreal. Households were selected from a sampling frame with the complete list of households in the Gram Panchayat, developed by conducting a census. The list was stratified by location in 10 Panchayat wards and 17 percent of households were selected in each ward. A total of 543 households were selected through a circular systematic random sampling process; house numbers were arranged in a circular manner and every sixth household was selected. Over-sampling of the tribal population (n
226) was done by surveying all households in a tribal colony where one household had been randomly sampled. Each individual belonging to the household was interviewed, with the exception of minors (<18
y). Members were asked to respond to a standard questionnaire and underwent a physical examination by trained health personnel. Four health conditions common in populations prior to an epidemiologic transition (underweight, anaemia, goitre and tuberculosis) and one health condition common in the post-transition period (hypertension) were assessed. These measures enabled us to explore various patterns of morbidity associated with nutritional deficiencies as well as the epidemiologic transition.
Weight and height were measured using standard methods [26
]; underweight was defined as body mass index (BMI)
]. Pallor of the conjunctivae, tongue and nails was used to assess anaemia status [28
]. Goitre was assessed by palpation of the thyroid gland [29
]. Blood pressure was measured twice and the mean of the two readings was used for analyses. Hypertension was defined as the mean ≥140 and/or 90
mm Hg [30
]. Suspected tuberculosis cases (cases eligible for sputum examination) were defined as either coughing for at least one week during the last four weeks and having had moderate fever at night in the last four weeks, or having had a positive tuberculosis test in the last 12
To explore the health divide in adults, we compared individual age- and sex-standardised prevalence for morbidity measures. Standardised rates were computed using a direct method with the total study population as the standard population. Weights were assigned to generate representative estimates. Household caste or tribal affiliations were assigned to the three social groups conventionally used by the Government of India: Forward Castes (the most privileged), Other Backward Classes (a residual group of low castes), and the lowest ranked group, Scheduled Castes/Scheduled Tribes. The lowest group was further disaggregated by contrasting Paniyas with other Scheduled Tribe populations. To assess trends across age groups, analysis was done using three groups (18–30 vs.
y). The socioeconomic status of households was defined in three ways. The local administration classifies households as above (APL) or below (BPL) the poverty line using a three-step approach: households with land or asset holdings and a certain minimum income are considered APL; the poorest (destitute households, Maha Dalits, single women, households with disabled persons as breadwinner, households headed by minors) are considered BPL; and the remaining households are ranked based on a quality of life score (based on 13 parameters such as type of house, availability of clothes, sanitation, literacy, means of livelihood and indebtedness). In addition, we used household landholdings of more than 50 cents of land (100 cents is equal to 1 acre) and crowding (ratio of individuals to rooms ≥3) as indicators of socioeconomic status [32
Multi-level logistic regression models were used to estimate the predicted prevalence of morbidity for each age and social group, controlling for other covariates. Estimates were based on the current distribution of poverty and education across caste-age groups and the local population sex ratio. Random effect models were used to adjust for clustering at the household level and provide robust standard error estimates.
To further explore the sources of the health gap between tribal and non-tribal groups, we used a Blinder-Oaxaca decomposition [33
]. This technique enables us to quantify the part of the health gap due to group differences in the distribution of health determinants and the part due to differences in the effects of these determinants. The first component reflects differences in observable characteristics (endowments) between groups. The latter measures the part of the gap that remains unexplained after these characteristics are taken into account and is often seen as an indication of a discriminatory effect. In reality, the unexplained gap might reflect discrimination [35
] or unequal treatment [36
] of the groups, and also differences in omitted determinants of health [37
]. The larger this remaining gap, the more likely discrimination and other unobservable factors play a role in explaining the health gap. We applied the method to decompose the gap in the prevalence of underweight, anaemia and goitre between tribal and non-tribal groups first, and then between Paniyas and Other Scheduled Tribes. Observable characteristics included in the models were: age, sex, education, poverty status, land ownership, occupation (wage labourer), housing condition (crowding) and access to safe drinking water. Estimates were obtained with statistical routines designed for non-linear outcomes (fairlie.ado and ndecompose.ado) [38
]. We used Stata version 11 software for all analyses.