Let's Talk About the Coaches Poll

Why are the polls the way they are? Is there another way? Is it found in this very post? Maybe.

T.J. Breshears

Jan 30, 2024

Buckle up, because this one is a little longer than usual.
I guess I feel strongly about this topic. Go figure!

When last week's AVCA poll came out, a new #1 was announced, and I had some things to get off my chest.

Battle Cats: Cougars & Tigers - Round 2

T.J. Breshears

January 23, 2024

Read full story

"Here is my problem with the polls: they have short-term memories. Early in the season, it is still difficult (impossible?) to know who the best team in the country is. The sample size is just too small. The fluctuation in rankings at this point in the season is based on one or two ‘impressive’ matches."

When yesterday's poll came, I was surprised, to say the least. I was surprised that last week's matches were weighted so unevenly. A handful of coaches vote in this poll, so I'm sure there are individual ballots that would feel more internally coherent. But somehow, this was the result.

The poll we got yesterday confuses me. And here is why:

Long Beach and Grand Canyon at the top shows that winning matters. These are two programs that went to the national tournament last year and haven't lost since.
Stanford, however, is also undefeated and according to VBelo has faced a significantly harder schedule than any other undefeated team. And yet, they are still back in 4th.
UCLA and Ohio State dropped significantly because they didn't win. But these upsets were not the same. The UCLA/UCSB match was the largest upset in VBelo history, and probably not indicative of which team is better overall.
Meanwhile, Ohio State was defeated twice by Penn State. It might be hard to beat a team once. But by the second match, VBelo had Penn State as the favorite. A reasonable loss. For some reason, Penn State is two spots lower than Ohio State. You have to squint hard to find a reason to make that happen.
UC Irvine defeated BYU twice in Provo and they are two spots ahead of the Cougars. This makes more sense but is so confusing next to Ohio State and Penn State.
The bottom half of the ranking feels a little bit like names out of a hat. I'm not sure why they are in the order they are. One thing is clear though. Being in a long-established conference goes a long way.

A Better Way?

How can we rank teams without being overly swayed by a single win or loss? Spoiler: my answer is the VBelo model.

First, let's look at how wibbly-wobbly the AVCA poll has been this season (4 weeks in).

The first thing I notice is the glass ceiling between 9 and 10. No team that started ranked 10 or higher has been able to get into the single digits. This already seems a little strange.

Then the chaos of the most recent poll is shown right where four lines intersect near the bottom right corner.

As I have mentioned before, I am waiting to release the VBelo ranking for this season. I still want a little more data. Don't worry, I will show the entire season. I'm not trying to hide this part of the model.

Instead of looking at this year, here is what last year’s VBelo rankings looked like over the season (similar graph to the one above).

Yes, this looks like chaos. This is also an entire season's worth of data for half of the league. You do see movement, but you get to see how rankings change slowly over time. UCLA is a good example. They started as a top 5 team and even when they lost they stayed in the top 5 until they eventually finished on top.

You can find spots in these lines where teams lost (their line moves downward) but they keep their position (no lines cross them). Here is a good example:

On March 9, 2023, Penn State lost to then #2 UCLA at the Outrigger Invitational. Despite losing that match, when VBelo has two closely matched teams and the match goes to 5 sets, the change in VBelo points is not that much. It was a great matchup between two great teams, but at the end of the match, both teams were still great. One loss does not mean Penn State suddenly dropped several spots on the ranking.

I hope this helps illustrate why I find value in a model like VBelo. Close matches between closely ranked teams (according to VBelo) can only move the needle so much. When a team begins to lose when VBelo projects them to be favorites, you start to see drops in ranking over time.

Instead of being swayed by one match or trying to explain results away (e.g. so-and-so was injured), VBelo looks at the results and gives that a value. Over time, a team's rating is about the journey it took them to get there and not about a two-hour match.

The downside here is that VBelo responds slower than the polls. If it is clear that there is a new #1 team in the country, it will take VBelo time to recognize that. That team has to earn it. And other teams have to show why they no longer belong in the top spot.

I think the "slowness" of VBelo is one of its strengths. We like hot takes and rapid reactions. It's fun to see meteoric rises in polls. But this isn't a video game and teams take time to develop. It also takes time for us, as observers, to have enough information to make clear judgments.

At the end of the day, polls are fun and give a gauge of who's hot and who's not. The VBelo model is less sexy when it comes to hot takes. It is all about the slow burn with repeatable results. Long live data.

Enough of me. What do you think of the polls? I genuinely am interested in hearing what you think.