HokieEngineer22

I’m somewhat reluctant to even engage further, but I can’t really help myself. You do not build robust statistical models by cherry picking a few things like record and subjectively weigh them together as has been done with the RPI. You pick a loss/evaluation function to evaluate performance, then iterate on your chosen model to try to improve performance by seeing how it performs on unseen data. In essence, you want the model to learn what should be weighted from the data rather than just randomly guessing. In this case, the mean squared error (MSE) of a particular rating system’s predicted scoring differential between two teams would be the obvious evaluation function. A simple linear regression model with an indicator variable signifying if the game was played at home/away/neutral, an indicator variable for the opponent, and an indicator variable for the team themselves for each game could be the predictor variables that could be selected. The target variable would be the scoring differential between the two teams. This allows the model to learn weights to put each team’s performance in each game in context of the opponent they played. Simple linear regression utilizes a closed form solution to minimize MSE, exactly our goal here. The coefficients for each team would then give you an estimate of team strength properly controlled for location and schedule. You would then evaluate how well your model does on unseen data by holding out the 20% or so of a given season’s games and evaluating the MSE of your models’ predictions. This is the sort of methodology things like the NET, KenPom, Massey, etc. take when creating and evaluating predictive models. It sounds simple, but this approach will give a far more accurate representation of team strength when it comes to predicting which team is better than what the RPI does. ChatGPT could probably do a good job explaining this further if you were interested in learning more. Of course you are correct that selection is not being done based purely on metrics. Instead, we are allowing people to make arbitrary decisions that give them the flexility to advance their own agendas. Just take a look at the college football playoff committee. I would claim that VT (and essentially all non-SEC teams) would be better off in a system where decisions were made based of a fair, objective evaluation criterion like a rating system with publicly audible code than some committee.

It is absolutely not grounded in any real statistics, it just averages a few things together that sound like they might be useful (record, opponents record, opponents opponents record). If you tried to present that as a legitimate model you would be laughed out of the room at every job I’ve ever had. Even a simple linear regression based would be more defensible. On your point about the strong conference schedule being what helps the SEC in the RPI rankings, you are exactly right and that’s why a more sophisticated model that can go beyond simple record comparison is needed. Any regression based analysis could better control for unequal scheduling. The NET in basketball is a good example of the NCAA realizing this and switching to a more robust, albeit still imperfect, metric. There is no excuse for this kind of antiquated “metric” having so much control over college softball/baseball. A volunteer graduate student could cook up something better in 30 minutes.

They seemingly feel free to ignore it when the committee feels like it doesn’t actually reflect the best 16 teams. In the case of Virginia Tech, any legitimate metric you look at will show them among the best 16 teams. If there was a rigorous and data-driven process for softball like basketball gets, I feel confident we would have been in the 12-16 range because that’s what the facts reflect.

…all that closely in the seeding. If so, Arkansas would have been the #1 overall team. I do agree that it’s unlikely that the reliance on RPI will change because it gives SEC teams an inherit advantage because of the Midmajor teams they are able to schedule and their strong conference schedule. So there is a strong incentive for them to support the current system.

At risk of throwing fuel on the fire, I just can’t see a strong, data-driven argument for VT not being one of the top 16 teams. RPI is a joke and not at all a legitimate metric. Even so, the committee seemingly ignored it in their seeding in many of teams anyway (Arkansas was #1 in RPI and got the #4 seed). Massey has the best well-known public ratings and has VT at #14. I also developed my own rating system earlier this year where VT finished at #12, ironically right above LSU. Not here to whine, but in my view claiming VT was not among the best 16 teams is not supported by the facts. At risk of being conspiratorial, I can’t help but suspect things might have turned out differently if we had SEC patches on the jersey in both the seeding and the hotel situations.

…of the NET rating that was RPI-adjacent after the analytics community pointed out it just made the ratings worse for no reason. It’s now very close to what you get from KenPom or T-Rank.

There is nothing even close to resembling RPI in the current formulation of NET.

I don’t know how many people will actually show up but it’s going around VT student circles.

grassroots campaign to raise money for this 100k fence before we can even get to the rest of stadium 😂

There’s a pretty hardcore base of support for the women's team now.

Looks like the push to retain some of our current players has begun.

…was probably even worse IMO.

..with the bad loss to Syracuse

Some bad defense by us but some bad luck.

compared to what they get.