Last night after doing some diagnostics I corrected or changed the following items in the College Basketball Model:
- I discovered I was improperly calculating offensive/defensive rebounding expectations in the INDY model so I fixed that.
- I made some changes as to how I calculate expectation for various stats in the INDY and TEAM models.
- In the INDY model I have switched from using rosters from a team’s last game to using the combined rosters from a team’s last 3 games and apportioning minutes played expectations based on the average of % of minutes played in the last 3 games. I feel like this adds more stability to the roster/floor lineup probabilities. At first I was going to make it the last 5 games, but reduced it to 3.
- For live betting purposes, I changed the way I calculate % of possessions remaining, having researched and found that on average, about 46.8% of a game’s possessions occur in the first half, and 53.2% in the second half/overtime. I had originally been doing a straight 50-50 split. This has helped live total projections become more accurate as previously they were too skewed toward the under.
- BLENDER, ATM, and SIMPLE models remain unchanged.
I feel like the changes to the INDY model are significant enough to ignore any keys and result records on the website prior to January 15, but the other models shouldn’t change too much. The January 15 run was updated to reflect the changes, so all numbers January 15 and onward should reflect the current state of the INDY model.