This bayesian approach is built using an a priori APR calculated as follows:
If a driver is in the same team as last season, his a priori APR is last season's average. If he has changed team, things are more tricky: it is averaged between his APR last season and this year car's APR last season (modified).
For example FA's is the average between his APR and McLaren APR last season. The "modified" APR of the car is using a funny result ("Broken Stick Theorem"), who, loosely used here, says that in a team the best driver gets about double points of the worst driver. So for Alonso I have used the "best driver" APR for McLaren last year. It sounds complicated, I cannot negate it
Kubica's and Hamilton's cases are special because they don't have complete driver data for last season, so they have been averaged with a different weighting scheme.