Here are the ranking projections for IRI according to my event simulator. This will be a trial run for my new pre-schedule projections. I’ll be updating this thread after we get a schedule and once or twice a day during IRI.
Looks like a very deep field, 2056 and 2590 are the only teams I have at over even odds to rank in the top 8, due in large part to their proficiency at getting the face the boss RP. It’ll be interesting to see how the predictions change after the schedule is released.
Feel free to critique/criticize all you want. If you see any consistent biases please let me know, my model has certainly had biases before. Just understand that if you wait until after the event to point out a flaw, I’m not going to give your opinion much weight since you’ve likely been deceived by hindsight bias.
Yeah, I don’t presently have any kind of correction for this in my model, which probably causes the predictions to be more overconfident than they would otherwise be. Without knowing exactly which teams are swapping drivers, the best I can do would be to revert all skill metrics toward the mean some amount. I’ll probably look into this eventually.
From a cursory look at these projections, I assumed that the sub-components were already heavily regressed to the mean. I think that’s safe way of doing things in general and pretty much standard among many of the major sports projection systems. (DVOA projections, ZiPS, etc.)
No, I haven’t regressed anything toward the mean yet, which means that the Elo ratings used here are each team’s end of season rating and the calculated contributions used here are each team’s max calculated contribution in that category from the season. I regress Elos toward the mean 20% (in addition to factoring in the team’s Elo from two seasons ago) between seasons and calculated contributions toward the mean 10% between seasons though. I’m pretty sure my off-season predictions could be improved by regressing a bit toward the mean, but I don’t know off hand what would be an optimal amount, and the optimal value likely even varies across off-season events, depending on how close we are to the end of last season and how “seriously” teams take the event.
I’m surprised by how confident people seem to be that 2056 will seed first. They may well be the favorites, but I’d take the rest of the field over them to seed first in a heartbeat. Indeed, every team except 2468 managed to seed first in at least one of my simulations.
2363 is still attending IRI. Not sure why they fell off the TBA list.
Some of these predictions are spicy, to say the least. I find it interesting that 2056 and 2590 have near equal chances of seeding despite 2056 going 17 for 17 on 90 point endgames on Tesla, while 2590 went 12 for 16…
I’m expecting that the ranking point spread at the top will be very narrow such that one good or bad match by any one of a dozen teams will result in large swings in their position, even toward the end of quals. Certainly this was the case with Turing. I think there are a large number of teams that have the capability of getting both of the bonus RPs in every match so their RP average will be based on the W-L ranking points. That will depend to a large degree on the luck of the schedule and other intangibles.