In a recent blog post I chronicled, in detail, the creation of a series of models designed to estimate the remaining career length of men’s AFL players and their likelihood of achieving game milestones in the future as at their completion of 1, 10, 25, 50, or 100 games.

Since then I have been investigating the inclusion of a number of additional variables:

Brownlow votes up to and including the previous season
Coaches votes up to and including the previous game
Coach changes
Height
Draft Position and Source
Team overall and recent win rate
Selection streaks and gaps

As the table below indicates, only the last three of those variable types provided sufficient additonal predictive power to be included in one or more models.