Use the National Wealth Index But Beware the Sample Weights
Posted: 25 Jun 2007
Background: Wealth index based on household asset scores is widely used as a predictor of health outcomes, choice of providers, etc. Wealth scores and their weights can be derived at several levels - national, national-urban-rural, or state-wise-urban-rural. Previous studies raised concerns about using a national level wealth index but did not compare alternative levels of wealth indices as a predictor of outcomes or economic status.
Objective: Thus, this study tested what the wealth scores at different levels reflect about the economic status of the households as well as their predictive powers in the context of maternal health service utilization (i.e. ANC visits, skilled attendance at delivery).
Method: Data from the Indian National Family Health Survey 1998-99 were analyzed with a variety of statistical techniques i.e. principal components analysis, correlations, regressions, simulations.
Results: First, the study tested the reliability of the wealth index as a measure of economic status of households. Wealth quintiles created at different levels (i.e. national, national-urban-rural, state-wise-urban-rural) were compared with each other and with alternative measures of economic status (i.e. food security index, standard of living index) to determine which factor best explained the variability of economic status of households. Then the study tested how much the alternative measures of economic status together explained the variability in the different levels of wealth indices. Food security and standard of living together explained the variability in the national wealth scores the best.
Second, the study compared the predictive powers of the different measures of economic status. State-level generalization of wealth scores did not perform better than the national-level wealth scores in predicting the use of maternal health services. The national household wealth index scoring system is a more significant predictor of maternal health service utilization compared to that measured by national-urban-rural wealth index, state-wise-urban-rural wealth index, standard of living index, or food security index. State-wise-rural-urban wealth index explained the least in the variability of maternal health service utilization.
Caution: The national sample weights used to create the national level wealth indices may not be appropriate. Because when this weight was used to create the wealth index (national, national-urban-rural, DHS), the households were not equally distributed in each quintile (i.e. 20% households in each quintile). When the national sample weights were not used to create the quintiles, there was equal distribution of households in each quintile. The state sample weights seem appropriate; when they were used to create the state-wise-urban-rural wealth quintiles the households were equally distributed in each quintile (i.e. 20% households in each quintile). Furthermore, national sample weights gave a pro-rich household distribution which biased the study results. For example, inequalities between quintiles (e.g. richest/poorest) showed a much larger gap than when the quintiles were equally distributed for each group, both with simple crosstabs (covariates not controlled) and multivariate simulations.
Conclusion: National level wealth index is more highly correlated with other measures of economic status and the best predictor of maternal health service utilization outcomes. However, national sample weights used to create national level wealth indices may not be appropriate.
Keywords: wealth index, national, India, maternal health service utilization
Suggested Citation: Suggested Citation