BubbleGrid for Wednesday, March 10: In the eye of the beholder

After posting yesterday's BubbleGrid, I received a comment wondering whether I'd forgotten Illinois from the list. Unfortunately, I hadn't. The Illini were the 12th team out and below the threshold I used to generate the list. (I have since expanded the list.)

Illinois is a great example of a team that causes big problems for the CtD model. (Teams with significant injuries, like Purdue, Ohio State, and Wisconsin this year, are also trouble.) The Illini have some very good wins (four in the RPI Top 27 with two on the road), but an otherwise blah resume. A dichotomous profile like this means that each of the ten members will have to decide which parts of the profile they consider more important. Our model tries to analyze what selection commmittees of the past have done with teams of all kinds of profiles and predict how they will handle this year's crop.

Even with replicas of the team sheets, it's hard to look at the profiles or two teams and compare the quality of their wins and losses. How can we make it easier to do this?

One possibility is match up each team's best win, second best win, etc. and look at the RPI rank of those opponents. We can do the same with losses, but instead compare their worst loss, their second worse loss, and so on. For example, here's a look at Illinois compared with Washington, another bubble team.

First, the wins:

The Avg column represents the average opponent RPI rank over all games up to that point. For example, over the Illini's first four best wins (#18 Wisconsin, #23 Vanderbilt, #25 Michigan St., and #27 Clemson), their average opponent rank was #23.

After calculating the Avg column for each team, we can calculate the differences. The Diff column for each line is the difference between each team's avg opponent rank for that line. So again, looking at each team's four best wins, Illinois had an average opponent rank of #23 while Washington's opponents averaged a #42 rank. (A negative difference means the average opponent rank for the team on the left is better, while a positive difference means the opposite.) Advantage Illinois.

By comparing the quality of each teams best wins, best 2 wins, best 3 wins, and so on, across their whole profile (wins and losses), we can get a general idea of how their profiles compare without picking arbitrary boundaries (e.g., record vs. RPI Top 50 or their two best wins) that aren't meaningful.

Let's do the same for each team's losses, but instead we'll put their worst loss first.

So what do these tell us? Well, they don't by themselves answer the fundamental question of which team's profile is better. Each committee member has to make that decision on their own based on their subjective preferences. If we have our own idea about how to compare two teams (e.g., I might care more about how their very best wins compare), this helps us easily identify those differences much more easily than the team sheets.

This is not perfect, of course. The numbers don't explicitly separate home/road results, but the coloring helps visually identify trends. For example, it makes it easy to see that Washington's best eight wins were at home, which some committee members might not like. It also doesn't explicitly account for different numbers of wins or losses. However, it does visually emphasize when one team has more wins or losses than the other.

Here are similar tables comparing Illinois and Rhode Island:

Illinois has better wins over their eight best, but the balance switches to Rhode Island after that.

And losses:

The takeaway here is that Illinois has 5 more losses, and while Rhode Island's first three losses are worse, after the Rams' remaining losses are even if not slightly better.

Questions? Comments? Requests?

And here's today's grid. With Illinois.