Predicting poverty and wealth from mobile phone metadata

See allHide authors and affiliations

Science  27 Nov 2015:
Vol. 350, Issue 6264, pp. 1073-1076
DOI: 10.1126/science.aac4420
  1. Fig. 1 Predicting survey responses with phone data.

    (A) Relation between actual wealth (as reported in a phone survey) and predicted wealth (as inferred from mobile phone data) for each of the 856 survey respondents. (B) Receiver operating characteristic (ROC) curve showing the model’s ability to predict whether the respondent owns several different assets. AUC values for electricity, motorcycle, television, and fridge, respectively, are as follows: 0.85, 0.67, 0.84, and 0.88. (C) ROC curve illustrates the model’s ability to correctly identify the poorest individuals. The poor are defined as those in the 5th percentile (AUC = 0.72) and the 25th percentile (AUC = 0.81) of the composite wealth index distribution.

  2. Fig. 2 Construction of high-resolution maps of poverty and wealth from call records.

    Information derived from the call records of 1.5 million subscribers is overlaid on a map of Rwanda. The northern and western provinces are divided into cells (the smallest administrative unit of the country), and the cell is shaded according to the average (predicted) wealth of all mobile subscribers in that cell. The southern province is overlaid with a Voronoi division that uses geographic identifiers in the call data to segment the region into several hundred thousand small partitions. (Bottom right inset) Enlargement of a 1-km2 region near Kiyonza, with Voronoi cells shaded by the predicted wealth of small groups (5 to 15 subscribers) who live in each region.

  3. Fig. 3 Comparison of wealth predictions to government survey data.

    (A) Predicted composite wealth index (district average), computed from 2009 call data and aggregated by administrative district. (B) Actual composite wealth index (district average), as computed from a 2010 government DHS of 12,792 households. (C) Comparison of actual and predicted district wealth, for each of the 30 districts, with dots sized by population. (D) Comparison of actual and predicted rates of electrification, for each of the 30 districts. (E) Comparison of actual and predicted cluster wealth, for each of the 492 DHS clusters. CDR, call detail records.

Stay Connected to Science