Deep Proxy Means Tests: Using Deep Learning to Assess Poverty Status

This project includes using deep learning to improve asssesments of  poverty for development organizations, such as the World Bank.

Project Lead:

Jonathan Hersh, Assistant Professor
Economics and Management Science, Chapman University

Detailed Project Description:

In development, a household’s poverty status is most commonly determined through the use of a Proxy Means Tests (PMT). This is, at its heart, a probability model of poverty status on the basis of easily identifiable household characteristics such as the number of children, age of household members, their education, work status, and some easily identifiable assets.

Most PMT models are estimated using either linear or binary models, and attain low levels of accuracy due to lack of sufficiently rich data and relatively simplistic models that are employed. Deep learning holds promise for PMT through the ability to discover and represent more complex relationships. This will allow for better and more efficient targeting of aid, and more accurate information on the distribution of the poor within a country.

If successful, this will help inform poverty models built by development organizations such as the World Bank, as well as determine the effectiveness of deep learning models for applications in development.

Technical Components:

I need assistance building deep learning models in Python, and can train them using a limited set of observations (20k at most). 

Skill/Expertise Requirement(s):

1) Capable of building deep learning models in Python and can train them using a limited set of observations (20k at most).

Data Set(s):

Household surveys that identify household poverty status.