This past week at Sports Illustrated, Luke Winn and I revealed our Top 100 Scorers for 2014-15, our Top Freshman Scorers, our Breakout Scorers and our Top D1 Transfers. Past college stats, coach effects, and recruiting rankings were used to predict player performance.

Every year I try to do something new with my projections. Last year I added a simulation. Rather than simply project a mean for each player, I looked at the variance in performance based on player type (freshman, senior, transfer), and I allowed individual performances to vary as some players outperform or underperform expectations. I simulated the season 10,000 times and used the median projections to rank all 351 D1 teams. This year’s team rankings will be revealed by SI on Nov 4th.

This year the biggest thing I wanted to do was to make player projections more accessible to readers by projecting PPG, RPG, APG for some of the nation’s top players. I chose to focus on PPG in large part because I think it is more accessible to most casual fans of college basketball. I certainly understand how PPG can be misleading in certain situations. There are certainly a large group of fans that value ORtg and usage over PPG and would prefer we not indulge in “paceism” thereby elevating players from North Carolina at the expense of players from Virginia.

But I don’t think we should completely trash PPG. As a single measure, it contains a lot of information. PPG incorporates both information on efficiency and usage. It also incorporates information on a player’s relative value to a team. Coaches are interested in playing their best players major minutes. In the preseason, it doesn’t make sense to project a bench player to be very efficient and under-utilized unless a team is extremely deep at a certain position. Certainly in most sensible preseason models, minutes will be correlated with player quality. And because PPG incorporates efficiency, usage, and player value (through minutes), it says a lot about who are the most important players in college basketball.

PPG is very sensitive to the minutes’ projection. And that’s why working with Luke Winn has been such a tremendous advantage. Luke has the contacts to help vet more of our lineup projections. But Luke also has a great statistical background as well. One of the first things Luke noticed when he saw the player projections was that we needed to adjust playing time based on coach-specific rotation patterns. For example, Notre Dame’s Mike Brey tends to give his best players major minutes, while Arkansas’ Mike Anderson tends to use a more balanced rotation. Thus this year we added coach-specific rotations to the model.

That said I am always looking for areas to improve the model. And that’s why I love Twitter questions. Sometimes readers innocently ask questions that shed a lot of nuance and light into the projection process:

Paraphrasing @DarenHill: Why is Branden Dawson projected to have fewer RPG than last year?

First, we think Dawson will be one of the best rebounders in college basketball this year. We project him to have the 26th most rebounds per game in the nation.

When I first saw this question, I panicked and wondered if the model was putting too much emphasis on Dawson’s height. I have a separate regression equation for freshman, transfers, and veteran players based on various characteristics, and height is an important predictor of rebounding. Dawson is an under-sized post-player and I was worried that the model might be weighting his height too heavily. But when I double-checked the numbers, Dawson’s height was not the key factor. For a senior like Dawson with three years of player stats, height is almost irrelevant in the model.

The second thing I was worried about was that we had Dawson’s minutes’ projection wrong. We project Dawson to play around 30 minutes per game. That could be low for a player many of us think will be Michigan St.’s best player this season. But keep in mind that Dawson is a forward and it is hard for forwards to get major minutes because they are more likely to get foul trouble. The current model will sometimes project a guard to play 35 minutes per game, but that’s a very dubious prediction for a post player. More importantly, Tom Izzo is not a coach who overuses his best players early in the season. Izzo really likes to give his bench a chance to play to evaluate his players. Dawson was playing more than 30 minutes per game in the post-season last year, but on the full year that was a fair representation of his playing time.

As it turns out, the reasons for Dawson’s slight decrease in rebounding is Dawson’s past college stats. But before I delve into Dawson’s situation, I want to talk about the problem of small sample sizes and bounce-back seasons. If I have a college three point shooter who shoots 40% on 30 shots from three as a freshman and 30% on 50 shots from three as a sophomore, I think we’d all acknowledge that we would expect him to bounce back and shoot better on his threes as a junior. And 10 years of historical data back that up. Last year’s performance is the best predictor, but we shouldn’t throw out the data on what happened two years ago.

The dilemma we often face when projecting players is what to make of players with a huge improvement in performance. For example, if a player shot 30% on 30 shots from three as a freshman and 40% on 50 shots from three as a sophomore, is that a breakout performance or a hot-streak? I can tell you based on the historic stats that when a player has this profile, on average he will make about 38.5% of his threes as a junior. And that slight drop in efficiency can actually lead to a lower ORtg prediction and a lower PPG prediction for what everyone perceived to be a breakout player.

Oklahoma’s Isaiah Cousins is a good current example of this. He improved his ORtg from 72.9 as a freshman and 112.8 last year. I now project him to have an ORtg of 110.0 this year. The reason last year gets so much weight is because college players are at the developmental stage of their career and breakouts are quite common. But it should also make some sense that the previous season should get some weight. The college season is short and we have a limited sample of games to ever have full confidence in a player’s ability. Cousins made 38 of 94 threes last year, but that’s not a large enough sample to really know that he is an elite three point shooter. (The model is also worried because Cousins was a 2.7 star recruit out of high school. Star ratings often provide information about a players potential and they suggest that Cousins may be close to his ceiling.) Regardless, when you see my projections for Oklahoma this is one of the reasons my model doesn’t have the Sooners as high as some other college basketball experts.

Jumping back to Branden Dawson, as a sophomore he grabbed 16% of the available defensive rebounds. As a junior he grabbed 21% of the available defensive rebounds. My model projects him to grab 20% of the available defensive rebounds this season. Thus his overall rebounding numbers are projected to be a little worse.

The historic stats say this is the most likely outcome for Cousins and we can debate whether last year’s improvement was real. But there’s an added wrinkle with Dawson and that’s the reason I wrote this longer column. Dawson essentially changed positions between his sophomore and junior seasons. As a sophomore, Dawson played a lot on the wing and spent a lot less time close to the basket. As a junior, particularly late in the season, Dawson was playing almost exclusively at the four-spot. And Dawson is expected to play major minutes at the four this year. Thus we should probably weight last season even more highly and discount his sophomore season when projecting Dawson’s numbers.

This is a hard adjustment to make systematically. In terms of Dawson’s position on the official Michigan St. roster, nothing has changed. But there is some data of this type available. Here at RealGM.com, we have a projection for player position based on the recorded stats. Ken Pomeroy also added this feature last season. Perhaps by incorporating this type of information, we can do an even better job projecting players in the future.

But in college basketball there is still a lot that the stats overlook and that we can only learn from watching the games. As much as I believe in the projection model Luke Winn and I have been working on, I can say emphatically that over the next month as teams begin to have exhibition games and host their first early season opponents, you will learn things that dispute what our numbers suggest. Basketball is still a sport where scouting and watching film is incredibly important. But to me, this is also the beauty of college basketball relative to MLB. In baseball, almost everything, including range on defense, can now be quantified to some degree. But in basketball, there is still a lot to be learned by watching the games.