Why are the computer rankings down on Marquette?

Photo by Ryan Messier/Paint Touches

With a 2-0 record in a week of carnage for the rest of the AP top 25, Marquette was able to jump from 21st to 15th in the weekly ratings, the best ranking under coach Steve Wojciechowski and the top mark since March 2013, almost 6 years ago.

AP ratings don’t mean anything, in that NCAA Tournament bids don’t take these rankings into account, but they do provide plenty of promotional value for fans and alumni not following the team super closely. Highlight shows and bottom screen tickers also use the AP poll as a referendum on what teams are worth devoting time to. So rankings don’t matter, but they aren’t useless either.

And yet, if you took a quick look at the computer rankings that Vegas and the more analytically inclined fans use, you’d see quite a different placement for Marquette.

Sagarin: 25
KenPom: 35
TRank: 38
Haslam: 43

The 20-spot discrepancy in the rankings between the AP poll and KenPom was the largest of any of the AP top-25 teams, with Ole Miss and Houston both coming in 12 spots ahead of KenPom.

So the question becomes, why are the computer metrics so down on Marquette compared to the humans that make up the AP poll.

1. Different Measures

This is one that’s difficult to internalize, but is the most crucial. The AP poll is made up of 65 journalist, and each brings their own set of criteria. But historically, AP voters are taking a snapshot in time based on preseason expectations and existing results. So what the AP poll is basically measuring is a particular team’s resume to date.

The computer algorithms that make up KenPom and TRank (and the others) do take the resume into account, but they are predictive measures meant to be able to tell you how good a particular team is (expected to play based on efficiency and other measures) going forward.

When the NCAA was bringing in stat heads like Ken Pomery and John Gasaway to get a working group together to build a metric to replace the RPI, that was one of the key discussion being had. Do we want the metric to be predictive or do we want it to be resume based?

The difference is bigger than you think and gets to the heart of what selection to the Tournament signifies. Is it a composite of all the games you played and what the results were, or is it a measure of how well you played and can expect to keep playing? When it comes to the Dukes of the world, these are often one in the same. But you don’t have to look back very far to see this discrepancy in action.

Take Xavier last season. They had a sparkling 28-5 regular season record, winning a historically good Big East regular season title over eventual national champion Villanova and providing a resume worthy of a No. 1 seed, which they eventually got in the NCAA Tournament.

And yet, the computer models didn’t have Xavier anywhere near the elite 1-seeds. Xavier finished ranked 15th on KenPom and 12th on Trank, which roughly translates to a 4 seed. All of this is just to reiterate the fact that defining what you are measuring is the most important part.

2. It’s all about efficiency

For most computer models, and all the ones listed above, efficiency is used as the barometer for quality. Efficiency in the basketball case means how many points you are scoring/giving up per possession. This is why you will often hear about tempo-free stats.

Back in the day (like 10 years ago) a team that led the league in points per game was anointed the best offense. But points per game doesn’t account for total possessions. Is it better to score 80 points in 80 possessions, or 65 points in 60 possessions?

The Jon Rothstein’s of the world still base a lot of analysis off the former, but a simple division would show us that it is in fact the latter. (80/80=1.0 while 65/60=1.083.) If that team that scored 65 points played an extra 20 possessions, they would be expected to score 81.6 points in the same 80 possessions.

KenPom and the other models do adjust this based on not only tempo, but the quality of your opponent, though. So if you are facing an elite offense and hold them to 70 points in 70 possessions, it is rewarded much more favorably that giving up 70 to a poor offensive team in those same 70 possessions.

3. Blowouts matter

When you look at a win/loss record, there is very little context provided. So a historically improbable win carries the same weight as a comfortable win. But to the computer models of the world, efficiency is key. They measure performance against expectations and then penalize or reward teams that over/uder performance.

Marquette came into Big East play ranked 28th on KenPom and proceeded to lay an egg, getting blown out by 20 at a St. John’s team ranked 51 at the time, when it was only expected to lose by a point or two. As such, Marquette’s efficiency numbers took a hit and dropped to No. 36 in the country.

So even though Marquette only has 3 total losses on the season, two of those were by at least 20 points, negating a lot of the efficiency cushion created in prior games.

4. Not blowing out teams matters

Remember how we said that these computer models adjust for team quality? That means your ranking can go down even with comfortable wins, if the opponent’s quality is poor.

Marquette mostly met expectations against its cupcakes this year, but only had 1 40+ point blowout, which means it didn’t build an efficiency cushion there. It also had a relatively close call against UTEP in early December. That 7-point win against a team ranked in the lower half of the 200s hurt MU from a computer ranking perspective just as much (6 spots) as a 23 point loss at Indiana.

In the computer’s eyes, there is such a thing as a bad win.

5. It’s all relative

Let’s face it, if Marquette has the same 2-0 week just a week prior, it probably only moves up a spot or two, if that. The big reason Marquette jumped so much was that nearly everyone ahead of them lost at the same time.

A well-timed win can have lasting effects in the AP poll.

For the computer models, there is also plenty of relativity. MU’s Adjusted Efficiency margin of +16.56 puts them at No. 34 right now, but would be good enough for 31st last season. The results could all be the same, but the rankings are still dependent on how efficient those around you are as well.

Performance vs. Potential

One last time, the polls and models are not measuring the same thing. Marquette’s two wins last week were not particularly strong, performance wise, each ranking below 90 in TRank’s GScore.

This meant that it didn’t get much credit from the models. But they will count as wins just the same, and that is the final point I want to drive home.

Here at Paint Touches we spend a ton of time going through all sorts of advanced metrics to put performances and games into context. But that only matters at a very micro level. Playing great and losing is much worse than playing poorly and winning.

Check that Tweet above once more. Marquette has played at least half of its games in the Big East under Wojo with a GScore under 90. (Here’s a better explanation of what GScore is.) That’s not particularly good. But it has already won two of the three games in this segment. That tells us Marquette isn’t playing exceedingly well, but posting victories nonetheless.

At the end of the day, either Marquette’s performance will improve and the computer ranking rise, or some of that “luck” will run out and the AP ranking will fall. I wouldn’t expect the 20-spot gap to persist.

And yet, Marquette is a favorite in its next four games against relatively weaker opponents. If it wins and underwhelms in each of those four, timed with some losses from Florida State, Auburn, UNC and Kentucky, we could be seeing a Ken Pom ranking under 40 coupled with a top-10 AP ranking.

Long story short it’s dumb to worry about either rankings, just sit back and enjoy the ride.