In the continuing saga of exploring MSU’s struggles with turnovers, I put together a simple regression model to see if I could determine how significant the effect of pace is on MSU’s offensive turnover percentage. I looked at four potential independent variables–factors that may effect how often MSU turns the ball over:
- AWAY: Proxy for away games. Playing in hostile environments could cause a team to turn the ball over more. (Note: Texas coded as home game; Missouri/BYU coded as away; 0.5 for UCLA game.)
- DefTO%: Opponent’s defensive turnover % (for the season). The more turnovers an opponent tends to create, the more turnovers MSU is likely to make.
- DefEff: Opponent’s adjusted defensive efficiency rating. It could be that playing a team with a good defense overall–as opposed to a team good at creating turnovers but bad at other forcing tough shots and/or rebounding the ball–could cause MSU to struggle and turn the ball over.
- PACE: As asserted previously, my theory is that the slower the pace of a game, the more MSU bogs down in the half-court offense and turns the ball over.
The dependent variable is MSU’s offensive turnover percentage (MSUto%) for a particular game.
Data was pulled prior to this weekend’s games. Here’s the equation resulting from the linear regression I ran in Excel:
MSUto% = 33.05 + (-.064 x AWAY) + (1.14 x DefTO%) + (.07 x DefEff) + (-0.63 x PACE)
The R-squared for the equation is 0.44, so it explains a decent amount of the variation in MSU’s turnover percentage from game to game.
The results for AWAY and DefEff are (1) small coefficients, (2) not statistically significant, and (3) counterintuitive. Playing on the road appears to reduce MSU’s TO%, and they seem to turn the ball over more against defenses that give up more points per possessions. But the fact that neither of those two variables is statistically significant means we can’t really be sure of the sign of those coefficients. In short: let’s ignore them. (I ran the model without those two variables and it makes very little difference; I kept them in the model in the interests of full statistical disclosure.)
The other two variables are statistically significant:
- For every additional percentage point of the opponent’s defensive turnover %, MSU’s offensive turnover % goes up 1.14 percentage points. This coefficient is statistically significant at the .02 level. In other words, there’s a 98% chance the variable is having a positive impact on MSU’s turnover % (positive numerically, negative in terms of basketball performance).
- For every additional possession per game (pace), MSU’s turnover % goes down by 0.63 percentage points.This coefficient is statistically significant at the .05 level. In other words, there’s a 95% chance the variable is having a negative impact on MSU’s turnover % (positive in terms of basketball performance).
The first result is pretty intuitive for any basketball team: The more your opponent tends to create turnovers, the more you’re going to turn the ball over. This model indicates it’s basically a one-for-one relationship.
The second result is less intuitive. As discussed previously, generally pace and turnovers are correlated the other way (higher pace = more turnovers), since turning the ball over shortens possessions.
These results imply that if MSU plays a 70-possession game, rather than a 60-possession game, it will turn the ball over two fewer times. So MSU picks up 12 possessions of scoring opportunities, while its opponent only picks up about 8 possessions (assuming an average TO% of about 20% and a neutral relationship between pace and turnovers for the opponent). That’s 4 points per game, baby. (Update: I should have pointed out that it’s important to remember that pace is a combined function of both team’s playing styles in a particular game. So, even with a concerted effort, MSU may not be able to increase the pace of a game by 10 possessions.)
Now it’s much easier to create a statistical model than it is to get a basketball team to play at a faster pace in a competent manner. But any time statistical inquiry is consistent with intuitive judgement, I think it’s worth paying attention to.
So I’ll stick with my original “Eureka”-inducing assertion: MSU should push the ball more on offense and look to play a little more aggressively on defense to force the action. This may result in some errors on both ends of the floor, but it appears that playing more conservatively can result in errors, too–turnovers resulting from poor half-court execution.
For those of you with some statistical background, I’m interested in your comments/critiques.
For those you without statistical background, I hope I haven’t completely wasted two minutes of your life. And feel free to comment/critique, as well–but in a less geeky manner.
As far as your regression analysis goes, the only concern I would have is the likelihood of multicollinearity between DefTO% and DefEff. According to Ken Pomeroy, efficiency is calculated as follows:
FGA-OR+TO+0.475xFTA
In a sense, you are capturing TO twice when looking at those two factors.
That being said, it was an interesting read. One has to wonder about the deliberateness of the Big Ten and its effect on MSU’s play. In the nonconference portion of the schedule, MSU was averaging nearly 67 possessions per game (66.8). In the Big Ten, the number of possessions has nearly decreased by 3 to 63.9. Given your analysis, perhaps it is not surprising that the average turnovers per game has increased from 13.7 in the nonconference to 15.5 in the Big Ten.
One note that may be interesting to watch is how tempo plays into the NCAA tournament. According to Pete Tiernan, who does a lot of bracket analysis for ESPN, slower paced teams do overachieve in the tourney:
“Considering this, it’s worth evaluating whether a team’s playing tempo has any intrinsic value in terms of tourney overachievement. After all, higher scoring teams do tend of have more offensive possessions than their lower scoring counterparts. Since points scored leads to overperformance, does it necessarily follow that playing tempo follows suit?
The answer, in short, is no. In fact, if anything, slower-paced teams are better overachievers than racehorse squads. Between 2004 and 2007, the average tourney team came to the dance with 67.1 possessions per game. Teams that had this many possessions or more actually underachieved against seed expectations (-.090 PASE). It’s true that three of the four champions in this era have been up-tempo teams, but only seven of the 16 Final Four contenders were faster-paced squads. Meanwhile, teams that had fewer possessions than average were +.102 PASE overachievers—and they accounted for one championship and nine of 16 Final Four slots.”
Currently MSU is 65.6 possessions, which might bode well for the tournament.
I like the statistical analysis a lot.
Couldn’t agree more with the gut-check time for the Purdue game.
Is it just me that wants to see more of Lucas penetrating more? He hits those floaters, creates some open three-pointers, and still dishes the ball well in the paint.
Bravo on the analysis. I haven’t had time lately to do much more than simply eyeball the box scores, so my analyses and thoughts have been more short and subjective than anything else. Having seen MSU play some of these defensively-minded teams at a slower pace, they always look to be the most vulnerable and have the least amount of confidence. Even Saturday they weren’t looking the sharpest, although turnovers and fouls weren’t out of hand.
All year, State has dangerously played to the level of their competition with few exceptions. Games against NC State, San Jose State, and a couple others were the only contests where we didn’t get sloppy with our play and didn’t lose our focus. So far, when the occasion has demanded it, we’ve played to the potential needed to win (Bradley (back then, a different team than now), BYU, Texas), or come valiantly close (UCLA). This week requires every bit of effort and potential (whether it’s been realized yet or not) that this team has.
I watched the whole IU game yesterday, and I’m now convinced that the primary reason they’ve been so successful this season is DJ White. Gordon gets all the attention because he can shoot from anywhere and is killer athletic. Ellis and Bassett get similar, though less sensationalized accolades as well. But having watched this team maybe half a dozen times this year (all against worthy opponents — not the non-coference fluff), Gordon is not a complete player. He can shoot the lights out and create something out of nothing better than anyone in the conference. I don’t believe there’s credible debate there. However, he makes a painful amount of poor decisions throughout each game, typically. Charges, ill-advised shots (he makes enough of these to continue taking them, but when he’s off, the team suffers big time), unforced turnovers, etc. Purdue, on the other hand, I haven’t seen play much this year but they are defying all reason and playing the best and gutsiest ball in the league right now. If we can find a way to come away with a single win this weak, it’ll be in Bloomington.
Wow, multivariate regression on a college b-ball blog! I’m psyched! what next? some post modern textual analysis?
I agree with the reader who pointed to likely problems of multicollinearity with your model specification. You might have added variables to test two common hypotheses out there is the blogosphere; first, a dummy for big ten teams, to test the hypothesis that teams that play MSU twice a year know all of its plays a head of time and this increases turn overs. Second — though this would demand a much greater level of number crunching than even you would probably be willing to undertake — that certain players (ie Gray) are responsible for the lion’s share of the turnovers.
Finally — some pedantry: your entry confused dependent and independent variables.
Good pedantry. I’ve corrected it.
Previous point on multicollinearity is well-taken. Dummy variable for Big Ten teams is a good idea.
Looking at individual players is tough, I think, because it’s so hard to control for the role each plays in the offense.
Great stats, anyone know why Summers was benched for the 1st half against NW?
I agree on pushing the ball and finding ways to take good shots earlier in the possession. My sense, in a non-statistical way, is that in the half-court plays develop slowly, and many times we are left with not enough time on the clock to set up for a good shot. That’s when things get forced and lead to turnovers. It isn’t so much the half-court as it is the kind of half-court offense we employ. Lots of set plays with quite a bit of standing around before the plays develop. I wonder what we could do with more motion and more penetration.
Donaldo poses an interesting scenario that (again) is not easily captured by the box score. Not only are we interested in the type of turnover MSU experiences, but WHEN does the turnover occur? Are the majority of the turnovers with 10 seconds or less?
[…] now to the numbers. After my stab at using linear regression to examine MSU’s turnover woes, there were a couple of good suggestions for revisions to the […]