With the portal season in our rearview, speculation season is now upon us. I’m not a big over/under season-win total guy, and I don’t even have a projected starting-5 to share at this time. What I do have are some more charts and graphs that hopefully help paint a picture of next season.
I pointed out last time how meaningless most projections are this time of year, but I wanted to see if there was anything we could glean from the ones we have access to, particularly in a historical context or when combined with other sets of data.
If the more old-school rating sites like On3 and 247 Sports make up one end of the projections world, on the other end of the spectrum you have the fancy formula guys like Bart Torvik and Ken Pomeroy. KenPom projections for next year aren’t out yet, he tends to wait for the game of musical chairs to conclude before updating his system, but Torvik’s system is capable of taking a stab at this stuff while the music is still playing.
Before looking at his projections, it’s important to first note the following, and this comes directly from Bart’s blog when he first started doing these:
As with all of my T-Rank stuff, this is not to be taken seriously…
With that in mind, it should be noted that I do not put a ton of weight behind these things either, but it is nice to have some real numbers to play around with this time of year and I’m thankful Bart provides some.
Looking specifically at the Big Ten, the model projects Penn State to finish dead last in the conference, and quite a ways out from the rest of the pack except for Minnesota. All 12 other teams project somewhere in the top 60 nationally, while the Gophers and Nittany Lions are sub-125, with Penn State coming in at 144 with a record of 11-20 (5-15). If that projection ends up anywhere close to the actual results of next season, it would be the 2nd lowest finish for Penn State in the last 15 years (that’s as far back as his model goes), only finishing ahead of Pat Chambers' 2016 team that finished 149th.
If we look at the individual team page to understand the data that goes into the team model, we see that Bart’s model tries to project the stats of the 10 biggest contributors on the team next year. In this case, that means Demetrius Lilley, Favour Aire, and Bragi Guðmundsson are expected to be bottom-three in terms of contributions next season, and they are left out of the projection formula.
This seems likely at the moment, so no major disagreements from me there. Additionally, the model predicting that Ace Baldwin will lead the team in minutes and offensive production seems very likely as well. Based on what we know about the team today, and considering much of this model is just the output of an automated formula, most of what’s here passes a sanity check.
Historical Performance
Keeping in mind those caveats around how serious to take these things, the natural next question to ask here is, how good are these ratings? I looked back at the last two seasons of Torvik’s numbers and compared his projected league standings to the media’s preseason picks and the actual standings at the end of each season.
Torvik and the media end up aligning closely, and in the places they do diverge slightly the trends still tend to match. Whether it’s because the media leans on metrics like his to make their determinations, or that the metrics tend to support the conclusions that the media comes to independently (or perhaps just coincidentally sometimes), there tends to be a lot of similarities.
There are big misses by both occasionally, neither the media nor the model saw Northwestern’s success coming last year, or Wisconsin the season prior. But even the misses tend to be shared by both systems, missing by roughly the same frequency and degree.
I offer up this comparison because regardless of how you feel about Torvik’s projections, there’s a good chance the media won’t produce much different predictions come the Big Ten Media Days this Fall. It’s tempting to push back on his numbers when they look bad for our favorite teams, but it would be surprising if Penn State isn’t picked to finish in the bottom 2 of the league by most pundits.
Of course, some teams will surprise even the best prognosticators each season, in both directions. There are also a lot of factors such as injuries and suspensions that can’t really be built into a predictive model. Perhaps it goes without saying, but these models and predictions are working with incomplete information about the future, they are limited to data from the past. For this very reason, some teams are easier to make predictions about than others.
Zooming Out
To take it a step further, I wanted to look at how the last 5 years of Bart’s preseason Big Ten power ratings compared to the actual numbers at the season’s end and visualize that on a scatterplot. If the model were to get every projection exactly correct, the dots would all land directly along the orange line. Any point below the orange line then is a team that had a lower actual rating than projected, and any point above it is a team that outperformed their projection.
The further away from the orange line the greater the disparity between projection and actual. I highlighted a couple of the recent examples of teams that outperformed pre-season expectations by a significant margin for some easy reference points, like Penn State and Northwestern last year.
The trend illustrated here is that we see a tighter grouping of the good teams than the bad. Put more simply, bad teams are more unpredictable than good teams. It’s a small data set, and I don’t know how meaningful that last point is, but it might provide a glimmer of hope for any PSU fans reading this after seeing how they are predicted to finish next season.
I don’t know if I have the detective skills or math chops to uncover the root cause behind the larger deltas for bad teams. I have a theory that coaching changes and massive roster turnover, which often go hand in hand, could potentially cause issues in a system like Torvik’s, but that probably doesn’t explain all of it and likely doesn’t explain why the media will end up agreeing with Bart.
Much of the historical data that informs a predictive model could be impacted by flipping a roster upside down and then sticking a new head coach in charge of that roster.
The Talent Factor
One thing that I expect most preseason media coverage to mention about Penn State is the fact that they’ve acquired several players who were highly-touted prospects coming out of high school. Guys who, for one reason or another, have not lived up to that billing just yet but now find themselves in a new situation. I don’t know how well Torvik’s model or the pundit predictions will be able to account for such scenarios, particularly a situation where there are going to be 3-4 guys like that all going to the same place.
The graphic above is a way of visualizing Torvik’s current projections for next year against the high school prospect ratings of each roster in the Big Ten. The diamond mark at the top of each vertical line is the average high-school rating (via 247’s composite system) for the top 10 players on each roster next year (as of 6/21/23). The circle mark on the bottom of each vertical line is the median of those ratings.
These vertical team lines are plotted along an X-Axis representing Torvik’s 2024 projections (his BARTHAG power metric rating). So teams further to the right are projected to be stronger than teams to the left, and higher vertical lines represent a greater talent level than lower lines (based on HS ratings). And lastly, the lengthier lines represent a greater disparity between the average and the median of talent on those teams than the shorter lines.
Keep in mind, these high school ratings are 5+ years old on some players, but I think this is a decent way of trying to visualize what sort of talent each team has going into next season. There are other studies out there about how well those high school ratings pan out, but that’s a separate discussion.
As far as Penn State is concerned on this breakout, you’ll note their far-left position on the chart reflects the low projection by Torvik’s system. However, the talent they have assembled would suggest they should probably be positioned quite a bit further to the right on the graphic. This level of talent would suggest they belong in that group of Rutgers, Nebraska, and Iowa, at least in terms of their fit on the trend line.
The Ceiling Is The Roof?
This supports the idea that some guys on this year’s roster have underperformed in their college careers to this point. The topic I want to touch on in my next and final “forecasting” piece this summer will look at some of those guys, both for PSU and elsewhere. Specifically, I want to focus on recent examples of highly-rated players who didn’t pan out at Blue Blood programs and transferred elsewhere.
What I’m driving at here is that while I think it’s logical to project Penn State finishing in the basement of the league next season, there may be tell-tale signs that this team’s ceiling is significantly higher. Now whether they can come anywhere near that ceiling is a completely different story.