Analytics

[skyblade]

It's crazy. The individual +/- for games was as large as 20 or 30 points between certain players. But the overall yearly average is what I was interested in.

How confident are you that EvanMiya is wrong? They list it as +/- and don't indicate they make any adjustments. Gobearcats.com doesn't have a gamebook for every game (@Wichita State doesn't have one for example) so it's pretty much impossible to check at least without being during the season.

Game by game +/- has a large variance and isn't a particularly reliable stat.
But I do think +/- is good at picking up players who do more (or less) than the stats indicate once you get to a large enough sample size. I do think it's debate-able if a season is a large enough sample size, but I do tend to trust a seasons worth of sample. Over multiple seasons there's generally enough variation (a player improves, coaching change, teammate change etc..) that any stat becomes less reliable.

 

[TheRealMick]

In that Temple game I was referring to, DDJ had a +/- of -17 in the second half despite being the leading scorer.

The shot charts are great too.

That has nothing to do with +/-. DDJ could be our leader in every category. It all depends on who you're sharing court with. Realistically it's worthless unless you want grade top 2-3 usage players on team. I can not score or let my man score, if other guys around get beat more than they win my +/- go be same.

 

[sedziobs]

That has nothing to do with +/-. DDJ could be our leader in every category. It all depends on who you're sharing court with. Realistically it's worthless unless you want grade top 2-3 usage players on team. I can not score or let my man score, if other guys around get beat more than they win my +/- go be same.

I agree. I think +/- is only useful to supplement individual performances that you observed. In this case Temple was attacking DDJ on almost every play down the stretch and I think it showed in his +/-. Even then it was more of a problem with our coaching than DDJ. But you can't just look at +/- and assume anything because each player is only 20% of the action on the court. There is way too much noise over the season. Using it to grade entire 5-man lineups might be more useful.

 

[skyblade]

I agree that +/- is best as a supplement, but I would argue that is true for pretty much all stats. I don't think it is worse than any other stat over the course of the season. Per game variance is high in almost every stat.

Over the course of the season luck should pretty much even out, so if a guys +/- is higher or lower than you'd expect it's a reason to look deeper into their play. Even over the course of the season it could be luck, but statistically it probably isn't. It's worth breaking down if it's just one or two outlier games or a trend, but that's hard to do with +/- because it is not very well tracked.

I think DDJ's +/- was low for two reasons on the season. One is that he was not very efficient on offense, he scored a lot but he also stalled the ball and took bad shots. The second is he is undersized on defense and teams took advantage of that. But as others have said it is team related, DDJ being undersized was part of the problem, but the bigger part was we were forced to put DDJ on bigger guards because we didn't have a better option.

 

[sedziobs]

All stats are tools, and you have to understand what they are telling you. Some conclusions can be drawn from stats, but others can't. For instance, you say "he was not very efficient on offense, he scored a lot but he also stalled the ball and took bad shots." There are some testable claims in there that we can look to stats to draw conclusions. We can look to true shooting percentage to determine if he was "efficient on offense". We can look at pts/40 to determine if "he scored a lot". And we can look at his long 2 rate as one measure of "bad shots".

The problem with +/- is we don't have a specific question to answer. We're looking at broad qualities with many variables and trying to draw conclusions that aren't robust.

With any statistical analysis, not just in sports, predictive value is what is important. If a stat can't help make accurate predictions, it's not worth relying on. It might help create a narrative, but it won't help make rational decisions. I prefer to stick to stats that are predictive. OReb, Free Throw, and Block+Steal rates are all generally consistent from year to year, perhaps after adjusting for usage and competition. Players generally see improvements in TO and Foul rates as they progress.

TS%, DReb, and Assist rates fluctuate a lot and can go up or down. Those are much less predictive, but at least they answer a specific question about the current/previous year. I just don't see much value in +/- over the course of a season. It's an interesting stat, but I don't think it's very useful.

 

[waterhead]

How confident are you that EvanMiya is wrong? They list it as +/- and don't indicate they make any adjustments. Gobearcats.com doesn't have a gamebook for every game (@Wichita State doesn't have one for example) so it's pretty much impossible to check at least without being during the season.

Game by game +/- has a large variance and isn't a particularly reliable stat.
But I do think +/- is good at picking up players who do more (or less) than the stats indicate once you get to a large enough sample size. I do think it's debate-able if a season is a large enough sample size, but I do tend to trust a seasons worth of sample. Over multiple seasons there's generally enough variation (a player improves, coaching change, teammate change etc..) that any stat becomes less reliable.

Go Bearcats does have all the gamebooks. I was calculating up until about 6-8 games left in season. I am confident in what I calculated for actual.

I explained to Sedz how to find some of the missing game books but I was able to find everyone until I stopped doing it. I will let someone else confirm if my numbers were incorrect. I am not under the impression I don’t make mistakes.

I am very confident Miya’s calculating is not actual plus minus. He describes how they do it on his site. It’s an algorithm. If it were not an algorithm it would just be simple addition and averages based on actual game plus minus numbers. Add plus minus and divide by number of games. Its not fancy

 

[waterhead]

Go Bearcats does have all the gamebooks. I was calculating up until about 6-8 games left in season. I am confident in what I calculated for actual.

I explained to Sedz how to find some of the missing game books but I was able to find everyone until I stopped doing it. I will let someone else confirm if my numbers were incorrect. I am not under the impression I don’t make mistakes.

I am very confident Miya’s calculating is not actual plus minus. He describes how they do it on his site. It’s an algorithm. If it were not an algorithm it would just be simple addition and averages based on actual game plus minus numbers. Add plus minus and divide by number of games. Its not fancy

He describes his process as stats related. An adjusted BPM which is also an algorithm. Plus minus IS already THE stat if you are calculating it you divide by number of games.

 

[waterhead]

He describes his process as stats related. An adjusted BPM which is also an algorithm. Plus minus IS already THE stat if you are calculating it you divide by number of games.

By the way I don’t place ANY emphasis on individual games or even short stretches. I only like plus minus over long term.

 

[sedziobs]

He describes his process as stats related. An adjusted BPM which is also an algorithm. Plus minus IS already THE stat if you are calculating it you divide by number of games.

He does have a "Box BPR" which is adjusted, but the "+/-" appears to be raw:

+/-: Number of points scored for the player's team with him on the court, minus the number of points scored by the opponent with him on the court.

If you didn't do the whole season, I'm not sure how you could be confident his numbers are wrong. The only reason I can see that they might not be raw is he discounts possessions when the game is a blowout. It doesn't say if that's only for BPR or if it also affects the +/-.

 

[waterhead]

He does have a "Box BPR" which is adjusted, but the "+/-" appears to be raw:

If you didn't do the whole season, I'm not sure how you could be confident his numbers are wrong. The only reason I can see that they might not be raw is he discounts possessions when the game is a blowout. It doesn't say if that's only for BPR or if it also affects the +/-.

His numbers are going to be correct for whatever measures he is using. I don’t think they are wrong. I am just saying it’s not an actual plus minus number. It’s weighted somehow. Maybe it’s weighed better idk. It’s just not actual

 

[skyblade]

No promises that I made no mistakes, I'm more confident that I got the starters right as they were almost always in the same order (unless someone was out due to injury). I agree with Waterhead that EvanMiya's +/- does not match the reality, not sure what adjustment they are making. I calculated +/- excluding Ashland (D2) and conference +/- and yes the Davenport conference +/- number is correct (I did a double take when I saw it). Make of it whatever you will.

Name: +/-, +/- (conference)
Ado: 20, -3
DDJ: 48, -4
Newman: 108, 14
MAW: 95, 19
Davenport: 108, 83
Ody: 27, -47
Saunders: 33, -52
Lakhin: 0, -47
Mason: 42, 6
Koval: 55, -28
Hensley: -12, -38

 

[sedziobs]

I calculated +/- excluding Alabama A&M (D2)

Do you mean Ashland? Alabama A&M is D1. Either way it looks like there is some kind of adjustment going on since just adding/swapping those games isn't enough to make up the difference. I think it's just that blowout possessions are being excluded.

Off Poss: Number of meaningful offensive possessions played.
Def Poss: Number of meaningful defensive possessions played.

One key step that we take to gain the best predictions from our data is to only look at possessions in a game that “mattered”. Analyzing possessions when the game is already well out of hand isn't as valuable to us as possessions when the winner hasn't been decided yet. We use the in-game naive win probability (which assumes that teams are equally matched) in order to assess when a game was out of hand. Once a team has a win probability of at least 99%, we start down-weighting the possessions until the win probability is greater than 99.99%, at which point we discard all possessions entirely.

 

[skyblade]

Do you mean Ashland? Alabama A&M is D1. Either way it looks like there is some kind of adjustment going on since just adding/swapping those games isn't enough to make up the difference. I think it's just that blowout possessions are being excluded.

Yeah, I even wrote D2 next to Ashland so I'd remember, then I subtracted out the Alabama A&M numbers and wrote their name here. I corrected it now.

Good finding the adjustment. Excluding blowout possessions would seem to be the more accurate measure.

 

[sedziobs]

Name: +/-, +/- (conference)
Ado: 20, -3
DDJ: 48, -4
Newman: 108, 14
MAW: 95, 19
Davenport: 108, 83
Ody: 27, -47
Saunders: 33, -52
Lakhin: 0, -47
Mason: 42, 6
Koval: 55, -28
Hensley: -12, -38

Everyone is within 25 of EvanMiya, except for Ado who you have 35 pts lower. There were quite a few games where Torvik's win probability reached 99% with a lot of time left (17 mins vs SMU, 15 mins vs X, 18 mins vs Florida A&M, in the first half against Presbyterian).

 

[skyblade]

Everyone is within 25 of EvanMiya, except for Ado who you have 35 pts lower. There were quite a few games where Torvik's win probability reached 99% with a lot of time left (17 mins vs SMU, 15 mins vs X, 18 mins vs Florida A&M, in the first half against Presbyterian).

Yeah, they don't match up exactly, but they are similar enough that it helps to confirm they are using the same data set (only EvanMiya's are adjusted to exclude meaningless possessions).

 

[waterhead]

No promises that I made no mistakes, I'm more confident that I got the starters right as they were almost always in the same order (unless someone was out due to injury). I agree with Waterhead that EvanMiya's +/- does not match the reality, not sure what adjustment they are making. I calculated +/- excluding Ashland (D2) and conference +/- and yes the Davenport conference +/- number is correct (I did a double take when I saw it). Make of it whatever you will.

Name: +/-, +/- (conference)
Ado: 20, -3
DDJ: 48, -4
Newman: 108, 14
MAW: 95, 19
Davenport: 108, 83
Ody: 27, -47
Saunders: 33, -52
Lakhin: 0, -47
Mason: 42, 6
Koval: 55, -28
Hensley: -12, -38

JD also missed a game as well as Newman. And Newman was on a pretty good streak right before injury and then didn't do a lot for a 6-7 game stretch. He did finish out pretty strong after that.

If you wanted to get real fancy with plus minus it would make sense to average the totals for games played. And even fancier would be to adjust for minutes played. Obviously DDJ, MAW, Newman and JD played the most minutes on the year. Ado played around 10 less minutes per game than these guys.

That is what I was doing. But I didn't adjust for the 4 guys above that were all getting around the same amount of minutes. But if Ado got 60% of what the others were getting I made an adjustment for that. So I had them all averaged by games played and then weighted for minutes.

I can't find my info that I worked on either which sucks because it took a while to do

 

[waterhead]

Do you mean Ashland? Alabama A&M is D1. Either way it looks like there is some kind of adjustment going on since just adding/swapping those games isn't enough to make up the difference. I think it's just that blowout possessions are being excluded.

I guess an easy way to minimize meaningless possessions would be to use just the first half plus minus numbers from Gamebook. And maybe add 2nd half numbers for conference games that were competitive til the end.

I like the idea of using the conference numbers by Skyblade because that is after other teams figured out how to play us. The team we were in the beginning was not who we were after Illinois and ARK etc.

 

[skyblade]

JD also missed a game as well as Newman. And Newman was on a pretty good streak right before injury and then didn't do a lot for a 6-7 game stretch. He did finish out pretty strong after that.

If you wanted to get real fancy with plus minus it would make sense to average the totals for games played. And even fancier would be to adjust for minutes played. Obviously DDJ, MAW, Newman and JD played the most minutes on the year. Ado played around 10 less minutes per game than these guys.

That is what I was doing. But I didn't adjust for the 4 guys above that were all getting around the same amount of minutes. But if Ado got 60% of what the others were getting I made an adjustment for that. So I had them all averaged by games played and then weighted for minutes.

I can't find my info that I worked on either which sucks because it took a while to do

Here you go, have fun. https://docs.google.com/spreadsheets/d/1kMjZr8J5duJfhm7Ym-xBVTZcR7dPvPEwavGYoxgn1b8/edit?usp=sharing

 

[TheRealMick]

DDJ can get beat 10x straight down court. If we hit 10 3s his +/- would be +10. I don't know what you'll use that stat for.

 

[waterhead]

DDJ can get beat 10x straight down court. If we hit 10 3s his +/- would be +10. I don't know what you'll use that stat for.

It's not a stat I would find useful in single games or even just a few games. But how the team is playing over the long term when certain players are on the floor is something I would want to know