Statistical evaluation of a crowd-sourced market information system in Liberia

Many critical policy decisions depend upon reliable and up-to-date information on market prices. We evaluate Premise Data, a new technology for measuring price information using crowd-sourced data contributed by local citizens. Our evaluation focuses on Liberia, a country with a history of economic and political instability. Our results indicate that the crowd-sourced price data is strongly correlated with traditional price indices, but that statistically and economically significant deviations exist that require deeper investigation.

Datalab Faculty

Joshua Blumenstock

Collaborators

Niall Keleher (UW)

Project Description

Many critical policy decisions depend upon reliable and up-to-date information on market prices. Such data are used to construct consumer price indices, measure inflation, detect food insecurity, and otherwise influence macroeconomic policy. In developing countries, where many of these problems are most acute, reliable market price information can be hard to come by. Here, we evaluate Premise Data, a new technology for measuring price information using crowd-sourced data contributed by local citizens. Our evaluation focuses on Liberia, a country with a history of economic and political instability. Using Premise Data, which recently began data collection in Liberia, we analyze tens of thousands of individual price observations collected at hundreds of different locations in Monrovia. We illustrate how these data can be used to construct composite market price indices, and compare these constructed indices to several sources of "ground truth" data from the Central Bank of Liberia and the United Nations World Food Programme. Our results indicate that the crowd-sourced price data is strongly correlated with traditional price indices, but that statistically and economically significant deviations exist that require deeper investigation. We conclude by discussing how indices based on Premise data can be further improved with simple supervised learning methods that use traditional low-frequency data to calibrate and cross-validate the high-frequency Premise-based indices.

Publications

Blumenstock, J.E. & Keleher, N. (2015). The Price Is Right? Statistical evaluation of a crowd-sourced market information system in Liberia. The 5th Annual ACM Symposium on Computing for Development (DEV '15), London, U.K.

Status

Completed