12.4. Example 2.5 describes a generalized linear model for data derived from testing an algorithm for human face recognition. The data are available from the book website.

The response variable is binary, with Yi = 1 if two images of the same person were correctly matched and Yi = 0 otherwise. There are three predictor variables. The first is the absolute difference in eye region mean pixel intensity between the two images of the ith person. The second is the absolute difference in nose–cheek region mean pixel intensity between the two images. The third predictor compares pixel intensity variability between the two images. For each image of the ith person, the median absolute deviation (a robust spread measure) of pixel intensity is computed in two image areas: the forehead and nose–cheek regions. The third predictor is the betweenimage ratio of these within-image ratios. Fit a generalized additive model to these data. Plot your results and interpret. Compare your results with the fit of an ordinary logistic regression model.