The Census Data Wealth Index: An Application To Predict Education Outcomes in Developing Countries
Rodrigo Lovaton Davila¹, Dorothy Gondwe³, Aine S. McCarthy¹, Phatta Kirdruang¹, Uttam Sharma¹
¹Department of Applied Economics, University of Minnesota, St Paul, MN, United States; ²Minnesota Population Center, University of Minnesota, Minneapolis, MN, United States; ³School of Public Health, University of Minnesota, Minneapolis, MN, United States

This research aims to develop a reliable and robust measure for socioeconomic status at the household level and to use this index to predict education outcomes. We develop the measure using census data available from the Integrated Public Use Microdata Series- International (IPUMS-I), the world's largest census database. Based on theoretical grounds, household income and expenditure levels are often used as measures of household wealth or household socioeconomic status. However, data on income and expenditures are subject to measurement biases and are not often collected in censuses. Therefore, we use non-monetary indicators including asset ownership, utilities, dwelling characteristics, appliances, and other amenities that are generally available in censuses to compute an asset-based wealth index.

In this paper, we examine how appropriate the asset index measures household socioeconomic status for census data and identify some conditions to produce a robust indicator, taking into account that the number and type of available variables differ considerably across censuses. Calculation of the asset index is performed through Principal Component Analysis (PCA), a data reduction technique that creates orthogonal linear combinations from a set of variables and orders them according to their contribution to the overall variability. We produce separate rural and urban asset indices and verify their agreement with the overall asset index. In order to assess the reliability of the asset index, we make comparisons with the Demographic and Health Survey (DHS) asset index and with predicted household expenditures. The DHS asset indices are widely accepted among researchers and have shown to be a reasonable proxy of living standards. Using census data for the same countries and year as DHS data, we show agreement between the two types of indices based on the distribution of each un-standardized asset index (mean, variance, skewness, and kurtosis) and the comparison of standardized indices graphically through kernel density estimation methods.

Since household wealth and education outcomes are highly correlated, measuring education outcomes requires a sensitive indicator of household wealth. We verify the predictive power of our asset index against selected educational outcomes. Even though the general recommendation has been to use the most variables possible, as long as they are related to unobserved wealth (McKenzie, 2005), it remains unclear which types of assets contribute more to the constructed measure and what the minimum number of necessary variables is. To check the type of assets that contribute more to the measurement of wealth, we calculate separate indices for asset durables, housing characteristics, and utilities. We run regressions for our selected educational outcomes and determine the predictive power of the asset index. We also verify the consistency of rankings produced by each subset of variables through rank correlations and correspondence indices and lastly, analyze possible clumping and truncation problems.

Biography: Rodrigo Lovaton Davila is a PhD candidate in Applied Economics at the University of Minnesota. He has a master's in Economics from Universidad del Pacifico in Peru and a BA from Pontificia Universidad Catolica del Peru. After his Masters degree, he worked in Washington, DC as a research assistant for the Inter-American Development Bank. Currently, he works as a research assistant for the Minnesota Population Center while pursuing his PhD with research focus international development and education.

The corresponding authors are all graduate research assistants at the Minnesota Population Center, where they work on the Integrated Public Use Microdata Series - International project, which disseminates free international census data. Aine McCarthy, Phatta Kirdruang and Uttam Sharma are also PhD candidates in the Department of Applied Economics, where they study international development. Dorothy Gondwe is a Masters candidate in the School of Public Health, where she is studying epidemiology.