ABSTRACT
Consider our bike-sharing data (e.g., Section 1.1), which spans a time period of several years. On the assumption that ridership trends are seasonal, and that there is no other time trend (e.g., no long-term growth in the program), then there would be a periodic relation between ridership R and G, the day in our data; here G would take the values 1, 2, 3, ..., with the top value being, say, 3 × 365 = 1095 for three consecutive years of data. 1 Assuming that we have no other predictors, we might try fitting the model with a sine term:
