Sunday, December 23, 2007

Analysis: It's the defense, stupid

While I haven't read every post on HoyaTalk since the Memphis beat-down, I've read enough to know that there is a wide spectrum of opinion on which players performed well and which did not.

But it should be made perfectly clear that, as ugly as the 2nd half was for Georgetown, the failure was not on the offense, but entirely by the defense. I'll explain how I think I know this.

I've touched on this in a previous post, but I'll reiterate here that it is possible to evaluate team performance, using the information in's stats database. If you look at the Georgetown schedule page on his website, you'll see a prediction for the remaining games, as a result (W/L), score, pace and probability. However, for games completed, this information is removed (why?).

Fortunately, I've worked out enough of KenPom's system so that I can calculate the expected offensive and defensive efficiency for any team against any other team, using KenPom's adjusted OEff and DEff posted on his website. This system isn't perfect, or even close to that, since it doesn't know about injuries, suspensions, etc. on the date of any game, but it can give you a rough idea how well Georgetown should fair against any opponent, accounting for game site, if they played today.

The result is what I call performance. Here, I use KenPom's expected point spread, adjusted for the actual pace of the game, and compare it to the actual point spread of that game. That is, if the Hoyas are expected to beat Team A by 10 points, based on KenPom's stats, but actually win by 14 points, the performance for that game is +4 points. Instead of printing another table, here I'll display the results as a category plot:

What performance gives you is context - it lets you know whether a 32 point win at home against Jacksonville is impressive (it isn't), and even allows you to rank each game by how well the Hoyas played. From this plot, you can see that the season-opening William & Mary win at home by 15 points (Perf. = -12 pts.) was more of a disappointment than yesterday's loss at Memphis by 14 points (Perf. = -10 pts.).

Of course, these evaluations are exceptionally hazardous to do early in the season, since there will likely be a lot of movement by all teams in both quality of play, and the statistics ability to judge this. But, as far as crude tools go, it's not bad.

You can take this a step further, and break out performance into its two components, offense and defense. Again, I can work through the math and plot how well G'town's offense and defense performed in each game, accounting for the quality of the opposing defense or offense (respectively) and the actual game pace. That is to say, if the Hoyas are expected to score 70 points against Team B in a 63 possession game, but actually score 66 points, the offense performance for the game is -4 points; same method for defense performance. Note that offense and defense performance summed should equal overall performance, unless I've buggered up the math.

First, offense performance so far this year:

It turns out that against the vaunted Memphis defense, currently ranked #11 by KenPom, scoring 71 points in a 70 possession game was not a bad performance at all, but actually +5 points above expected.

The other, more general note, is that there appears to be a trend of improving offense as the season has progressed, although I'd be a bit hesitant to draw this conclusion until another few games have been played.

On to defense performance:

Yes, it really was that bad.

After playing the first 8 games with a defense performance in the range of -4 to +6 points, the Hoyas' defense dropped a -15 pt. performance yesterday (or, equally as valid, Memphis had a +15 point offense performance against Georgetown).

I'll leave it up to those who know more about basketball to figure out what went wrong on defense (cough, cough, rebounding, cough), and come up with suggestions on how to fix it.


I've updated the Georgetown Season Statistics - Team and Individual Splits page through the Memphis game. There are big differences now in home/road splits, thanks to the Memphis game.

On another note, I've read many posts on HoyaTalk complaining about the perceived and/or real lack of production from this blog's favorite underdog Hoya, namely Vernon Macklin. While defensive stats are hard to come by and even harder to evaluate, right now we can see that Big Ticket has an Offensive Rating of 105, ranking near the bottom of the 9 regular players (and Chris Wright is making up ground in a hurry). We also are all painfully aware of his free-throw shooting woes (3-18 [17%] after Memphis).

Out of curiosity, I changed his free-throws made to 11-18 [61%], a pedestrian rate by any scale other than the current Hoyas, and his ORating moves to 132, which would lead the team. Quite simply, other than the execrable free throw shooting, Macklin is playing as well as we could have expected. Of course, you can't ignore FT shooting.

Finally, I'm just now getting to a brief story that Ray Floriani sent in earlier this week, summarizing the UNC/Rutgers clash.

What a difference a game makes.

On Thursday, Rutgers defeated winless NJIT at the Prudential Center in Newark.

On Sunday it was a meeting with top ranked North Carolina at the RAC. A breakdown of both games follows.

Rutgers NJIT
Score 65 55

Possessions 69 69
Off. Eff. 94.1 79.6

Four factors

eFG % 37 39.8
TO % 18.8 21.7
OR % 34.9 25
FT Rate 63 38.9

The game was as close as the score indicated; Rutgers led by 6 with 8 minutes remaining. The Scarlet Knights went on a tear to increase the lead to 16 before NJIT made a few late baskets for the final margin. NJIT stayed around a good part of the contest before Rutgers used their three guard set to spread the floor and take the Highlanders off the dribble. Notice the free throw rate as indicating Rutgers ability to get in the lane and ultimately get to the charity stripe.

UNC Rutgers
Score 93 71

Possessions 80 80
Off. Eff. 116 88.6

Four factors

eFG % 46.3 41.5
TO % 15 24.9
OR % 49 35.2
FT Rate 26.3 40

It was a faster tempo game with UNC putting the pedal to the medal. Rutgers paid dearly for their 20 turnovers, as UNC outscored the Scarlet Knights 30-10 off those miscues. The OR Rate attests to fact that Rutgers not only saw guard pressure (Ty Lawson 26 points 4 steals and several deflections ) but a formidable frontcourt (Tyler Hansbrough 20 pts 11 boards and Deon Thompson 9 boards).


  1. Hmmm. I think you are correct in your conclusion, but your logic doesn't seem quite right to me. By looking at per game stats (rather than per possession stats) you are missing the major point of tempo-free stats, aren't you? So the overall upward offensive trend that you observed out of your analysis comes at least as much from our increased pace over the past 3 games (which is 12 possessions higher than the previous 5 games) than increased efficiency (which did increase 0.14 points per possession). So can you do the same analysis with expected points per possession rather than expected points per game?

    I fully admit that I may be missing something here as I haven't spent a lot of time thinking about this. If I'm wrong, please feel free to let me know where I've misunderstood.

  2. Performance is calculated in "units" of points, so you are right, it isn't purely tempo-free. That is by intent. I guess I've spent too much time staring at Baseball Prospectus' obtuse stats (WXRL, WXL, Leverage), and wanted something readily tangible to any reader.

    To get the stats the way you'd like, all I'd need to do is subtract predicted offensive efficiency from actual offensive efficiency for a game.

    For example, based upon the stats through yesterday's games, if Georgetown played at Memphis tomorrow, you'd expect the Hoyas to have a game OEff of 94.7 (this is based upon GU's Adj. OEff and Memphis' Adj. DEff, along with a weighting factor for game location). In the actual game, they had a 101.7, which works out to a +7.0 OEff.

    Well, I have a vague idea of what that means (points per 100 possessions), and you may too, but it's a bit abstract. You'd need to know what a good or bad OEff value is, and how important +7.0 in OEff is over the course of a typical game. But if I just multiply by actual game pace, we now know that the Hoyas scored +4.9 points more than expected, a number that just about anyone can understand.

    So what would the effect be in comparing performance from games of different paces? For example, the game at ODU has an offense performance of +5.3 points, but in only 57 possessions. If you translate that to OEff, you'd get +9.3.

    So, we can compare Memphis vs. ODU truly tempo-free, and get +7.0 vs. +9.3 pts/100 possessions, or using performance and get +4.9 vs. +5.3 points.

    That difference may seem large to you, but if you look at the scatter in any of these performance stats over the course of a season, you'd see that the variability game-to-game mostly swamps this effect. Here's the last look at performance I ran last season, as a preview for the OSU Final 4 game. The range in performance is -29 to +19 points, with a st. dev. of 11 points; the ranges and st. devs. are slightly less for offense or defense performance.

    With regards to the offensive trend, the tempo-bias is nearly meaningless. Looking at a linear fits of the purported offensive trend (vs. date), the R^2 is 0.51 with performance, 0.50 with the tempo-free stat.

    Having said all that, it is certainly trivial for me to transform performance stats into completely tempo-free values. Maybe I'll pursue it a bit further down the road (with last year's data?) and see how much change there is in the plots. But not tonight.