Efficiency by Game and Player

[OFID]

Some time ago, somebody posted for a game or two that generated some discussion, so I thought I post some more

A visualization of efficiency by game is attached.

is a very crude measure:

• As per the formula given below, everything is weighted evenly. However, missing a free throw costs your team one point, but missing a field goal costs your team at least two points. Also, most blocks are close to the rim, so they're probably saving more than one point on average.
• There are a myriad of things that aren't recorded and/or made public that increase the probability of success on offense, and another myriad of things for defense. Since they're not recorded/reported, they're not in
• is affected by playing time. You can't do good things (or bad things) if you're on the bench.
• is affected by pace. Individual EFFs will be higher against, say DePaul, than against, say UCF.

Observation:

• For all of the talk early in the season of having no contributions beyond the Core 4, seven out of ten games before Jan 1 had 5 or more players with EFFs of at least 10.

Notes

• I used the WNBA definition: ((Points + Rebounds + Assists + Steals + Blocks) - ((Field Goals Att. - Field Goals Made) + (Free Throws Att. - Free Throws Made) + Turnovers))
• The data are from uconnhuskies.com
• The x-axis and y-axis are the same to facilitate comparisons across games

 

[OFID]

A visualization of efficiency by player is attached.

• Green bars indicate wins; red bars indicate losses
• An asterisk following the opponent's team name indicates a home game

If there is sufficient interest, I could update these files as the season goes on.

Observations:

• Megan's struggles against ranked teams (other than DePaul) are evident in the figure (page 1). Four of her five worst EFFs were Baylor, Tennessee, Oregon and South Carolina.
• I expected that Crystal's plot (page 2) would be pretty consistent, but she's also had her ups and downs. Aside from Tennessee, her EFFs for the ranked teams are pretty good.
• Molly is a good example of playing time effecting (page 3). Her score for Dayton and Seton Hall, when Crystal didn't play, and for Memphis, when Liv had foul trouble, stand out on her plot.
• Christyn's second half struggles are evident in the plot on page 4. She was pretty consistent through the Memphis game, but has often struggled after that.
• There's no bar for Olivia for the Baylor game because her was zero (page 6). On the other hand, her for the Oklahoma game (27 points on 13 of 18 from the field with 15 rebounds and 7 blocks) was the highest for any player in any game this season. Aside from the Baylor game, she's been one of our most consistent players.
• Anna's plot looks like what one would expect a freshman's plot would look line - lots of ups and down (page 7). She's produced both some very strong games and some real clunkers.
• Kyla is another example of the effect of playing time on EFFs (page 8). Her playing time has decreased after the SMU game, leading to lower EFFs.
• Like Anna, Aubrey's plot is what one would expect for a freshman (page 10), but even more so.

Notes

• I used the WNBA definition: ((Points + Rebounds + Assists + Steals + Blocks) - ((Field Goals Att. - Field Goals Made) + (Free Throws Att. - Free Throws Made) + Turnovers))
• The data are from uconnhuskies.com
• The x-axis and y-axis are the same to facilitate comparisons across players
• All data acquisition, processing and plotting was done in R

 

[eebmg]

Interesting. I think there would be a lot of interest by others (at least myself) if the mean value of the efficiency metric of each player was compared on the same histogram plot (maybe normalized by mpg?) . And probably a break down by 1) The Top 3 games, 2) The OOC games and 3) The AAC games. would be of further interest

 

[OFID]

I've added some observations to my first two posts in the hope that it would stimulate discussion.

I think there would be a lot of interest by others (at least myself) if the mean value of the efficiency metric of each player was compared on the same histogram plot (maybe normalized by mpg?)

As is often the case with statistics that are a ratio, strange things happen when the denominator is small.

• Against Cincinnati on 1/30, Evelyn played 5 minutes, had 5 points on perfect shooting, 3 rebounds, and nothing else. Her PER/40 minutes for that game was 64.
• Against Memphis on 2/7, Batoully played 3 minutes, missed her only two FG attempts, and had a turnover. The resulting PER/40 minutes was -40. She had another -40 against Baylor (one minute, one turnover and nothing else).

The distribution is going to be very long-tailed. Summarizing using the mean will not be appropriate, so I won't do those.

As mentioned above, depends on pace, so it's hard to say what happens when you take an average. With that caveat in mind, I'll have the other stuff later; interpret with caution.

 

[eebmg]

I've added some observations to my first two posts in the hope that it would stimulate discussion.



As is often the case with statistics that are a ratio, strange things happen when the denominator is small.

• Against Cincinnati on 1/30, Evelyn played 5 minutes, had 5 points on perfect shooting, 3 rebounds, and nothing else. Her PER/40 minutes for that game was 64.
• Against Memphis on 2/7, Batoully played 3 minutes, missed her only two FG attempts, and had a turnover. The resulting PER/40 minutes was -40. She had another -40 against Baylor (one minute, one turnover and nothing else).

The distribution is going to be very long-tailed. Summarizing using the mean will not be appropriate, so I won't do those.

As mentioned above, depends on pace, so it's hard to say what happens when you take an average. With that caveat in mind, I'll have the other stuff later; interpret with caution.

We appreciate any analysis you may find informative.

 

[OFID]

I think there would be a lot of interest by others (at least myself) if the mean value of the efficiency metric of each player was compared on the same histogram plot (maybe normalized by mpg?) . And probably a break down by 1) The Top 3 games, 2) The OOC games and 3) The AAC games. would be of further interest

I'm not sure what was meant by 'Top 3' games; I'm assuming they're the games against Baylor, Oregon and South Carolina (i.e., our losses).

Observations:

• The mean for each player looks pretty much as expected (page 1 of the attached file). I was a little surprised that Olivia looked better than Crystal, but Crystal had some real clunkers (page 2 of the efficiency by players file).
• Crystal actually did a little better in the losses than she did in the wins (page 2 of the attached file), probably due to the aforementioned clunkers in the wins. Everybody else had very steep drop-offs in the losses; the losses were a true team effort. Megan had the biggest drop-off.
• There wasn't a whole lot of differences between the conference and non-conference games (page 3 of the attached file). Anna had the biggest drop-off conference to non-conference, probably because most of the non-conference games were before she figured it out. Megan had the next biggest, probably strongly influenced by the three losses, all of which were non-conference games.

 

[eebmg]

I'm not sure what was meant by 'Top 3' games; I'm assuming they're the games against Baylor, Oregon and South Carolina (i.e., our losses).

Observations:

• The mean for each player looks pretty much as expected (page 1 of the attached file). I was a little surprised that Olivia looked better than Crystal, but Crystal had some real clunkers (page 2 of the efficiency by players file).
• Crystal actually did a little better in the losses than she did in the wins (page 2 of the attached file), probably due to the aforementioned clunkers in the wins. Everybody else had very steep drop-offs in the losses; the losses were a true team effort. Megan had the biggest drop-off.
• There wasn't a whole lot of differences between the conference and non-conference games (page 3 of the attached file). Anna had the biggest drop-off conference to non-conference, probably because most of the non-conference games were before she figured it out. Megan had the next biggest, probably strongly influenced by the three losses, all of which were non-conference games.

Great stuff. Thanks.

 

[JRRRJ]

I've added some observations to my first two posts in the hope that it would stimulate discussion.



As is often the case with statistics that are a ratio, strange things happen when the denominator is small.

• Against Cincinnati on 1/30, Evelyn played 5 minutes, had 5 points on perfect shooting, 3 rebounds, and nothing else. Her PER/40 minutes for that game was 64.
• Against Memphis on 2/7, Batoully played 3 minutes, missed her only two FG attempts, and had a turnover. The resulting PER/40 minutes was -40. She had another -40 against Baylor (one minute, one turnover and nothing else).

The distribution is going to be very long-tailed. Summarizing using the mean will not be appropriate, so I won't do those.

As mentioned above, depends on pace, so it's hard to say what happens when you take an average. With that caveat in mind, I'll have the other stuff later; interpret with caution.

The size of the denominator does not generate those strange results if you don't normalize to some arbitrary number of minutes -- just use the actual minutes played in each game.

Thanks for doing this!

 

[OFID]

JRRRJ: Sorry, I don't think I understand. The ratios of every pair of observations will be the same whether or not all observations are multiplied by 40. Evelyn's 64 will be the same number of standard deviations from the mean as will Batoully's -40's. The distribution will be just as long-tailed whether or not all observations are multiplied by 40.

 

[JRRRJ]

JRRRJ: Sorry, I don't think I understand. The ratios of every pair of observations will be the same whether or not all observations are multiplied by 40. Evelyn's 64 will be the same number of standard deviations from the mean as will Batoully's -40's. The distribution will be just as long-tailed whether or not all observations are multiplied by 40.

If you have someone who plays for 1 minute in a game and gets a basket, the pts/min will be 2.

If you display it as pts/40, it becomes 80, which I think is the kind of number you find strange. This is a result of using an interval so much larger than the actual interval. It even blows up the spreadsheet if the box score gives a player zero minutes!

I have frequently considered changing the per-40 spreadsheet to a per-minute sheet for this reason. Originally, I only had the summary-to-date sheet and this was only a consideration very infrequently and only for players who didn't play much. But since I created the versions where individual games can be called out this artifact has occasionally occurred.

 

[OFID]

I totally get that multiplying by 40 produces eye-popping numbers. What I'm trying to say is that 2/min is rare enough that it should also be eye-popping, but we just don't have the experience looking at points/min to realize that it is. I don't know if that made my point any clearer ...

 

[JRRRJ]

I totally get that multiplying by 40 produces eye-popping numbers. What I'm trying to say is that 2/min is rare enough that it should also be eye-popping, but we just don't have the experience looking at points/min to realize that it is. I don't know if that made my point any clearer ...

I get the point. The ratios aren't changing.

I'd posit that when a player comes in for a brief time and provides some fireworks it's good to call that out, and that 2 pts/min looks less cartoonish and worthy of respect than 80 pts/40.

But, at the end of the day it's only a change of perception, and only occurs occasionally. That's why I've left the spreadsheet as per-40.

 

[Dokey]

I suppose that while looking at the whole season is of some interest, the point of the season at UConn was to prepare, let’s face it, for NCAAT. So perhaps the more important figures are to be found in past ten or even past five games. Based on that which players can advance us in tourney?

 

[JRRRJ]

I suppose that while looking at the whole season is of some interest, the point of the season at UConn was to prepare, let’s face it, for NCAAT. So perhaps the more important figures are to be found in past ten or even past five games. Based on that which players can advance us in tourney?

That's why I switched to the pivot table -- so you can use the drop-downs at the top of the page to select which games you want to aggregate much more easily than my previous design...

 

[OFID]

Here's the end-of-season update to Efficiency by Game. Three games (Cincinnati, Houston, South Florida) were added.

 

[OFID]

And the end-of-season update to Efficiency by Player ...

 

[OFID]

... and the update to @eebmg (pages 1-3) along with a response to @Dokey.

For Efficacy by Player for All Games, I replaced the bar chart with box-and-whisker plots. For those who are unfamiliar with these plots

• The means are still in there. They're the large red plus signs.
• The medians are the heavy black lines in the middle of the boxes.
  • Means that are above the medians (Molly, Evelyn, Anna, Kyla, Aubrey) indicate that there are some very good games influencing the mean (late season success for the freshman and mostly minutes-related for the others). The reverse (Megan, Christyn) indicates that there are some bad ones influencing the mean (losses for Megan and second half slump for Christyn).
• The boxes contain the middle 50% of the EFFs for each player.
  • Smaller boxes indicate more consistency (Crystal) and larger boxes indicate more inconsistency (the freshmen and Christyn).
  • If the middle 50% of the EFFs were symmetrically distributed around the median, the median line would be in the middle of the box (Crystal, Kyla). For the freshmen, the line is near the bottom, indicating similar bad games, but good games that vary a lot. When the line is near the top (Megan, Crystal), there are consistent good games and variable bad games.
• The whiskers extend out to the highest and lowest values, unless they're too far out, in which case the values are plotted with circles.
  • Liv's performances against Oklahoma (good) and Baylor (bad) are pretty far out, as are Molly's games in which Crystal didn't play.

The Wins vs Losses and Conference vs Non-Conference plots are similar.

For @Dokey's request, I used the last 6 games, i.e., the games after the last loss to South Carolina.

• Megan has been at her best in the last 6 games (see also page 1 of the by player pdf).
• Crystal's mean in the last 6 games is dragged down by her in the Tulane and UCF games (page 2 of the by player pdf).
• Christyn had an of zero against UCF, but has seems to have broken out of her slump in the last 3 games (page 4 of the by player pdf). The zero pulled her the mean for her last 6 games below the mean for the first 23.
• In the game in which she was ill against Cincinnati, Liv had an of 3 (page 6 of the by player pdf). The other games among the last six were pretty good, but, like Christyn, that one bad game pull the mean for the last 6 games below the mean for the first 23.
• For Anna, the mean for the last 6 games looks great compared to the first 23. However, a closer examination (page 7 of the by player pdf) shows a great deal of inconsistency. We'd love to have the stat-stuffer version of Anna for the NCAA tournament, but she hasn't yet shown she can do that consistently.
  • I wonder why Anna struggles against USF?
• Aubrey's mean for the last 6 games is also much better than that for her first 23. She's been a little more consistent over those six games (page 10 of the by player pdf). Her highs haven't been as high as Anna's, but her lows also haven't been as low.
  • Has anyone else noticed that Aubrey, who is shooting less than 60% on free throws for the season, has hit her last 9? She looks much more confident up there, and seem to have altered her stroke.