Figure 2

Measurement performances for all city pairs. Mean absolute errors (MAE) achieved when predicting worst-off and best-off deciles by (A) income and (B) educational attainment for all source (training) and target (prediction) city pairs. Diagonal entries show results using data from a single city as described in the text. Random allocation of census tracts to deciles in the target city would result in an MAE value close to 5 (dark red), whereas perfect prediction would correspond to a MAE value of 0 (dark blue). Shades of blue depict better prediction performances compared to shades of red