The cone-shaped scatterplot indicates that the model performs differently depending on the magnitude of the estimated value. In this case, the model performs better (the residuals are smaller) for tracts with fewer crimes than for tracts with many crimes.
