Saturday, January 16, 2010

Once more unto the breach

I can't speak for the rest of Hoya-dom, but the coincidence of the next stretch of games with the date on the calendar brings to mind the dreadful memory of last season's implosion:  on January 16th, 2009, the Hoyas sat at 12-3 (3-2) heading into a road game against stiff opposition (Duke).

We all know how the rest of the season went:  4-12 (4-9).

In the next 5 games, Georgetown plays four of's top-25 (through Friday's games), with three of those games on the road.  Ken's predictions:
.                                    KenPom
Date          Opponent    Rank     Exp. Score
17-Jan      @ Villanova   [22]      L: 75-78
20-Jan      @ Pittsburgh  [25]      L: 60-63
23-Jan        Rutgers     [168]     W: 77-57
25-Jan      @ Syracuse    [6]       L: 68-77
30-Jan        Duke        [1]       L: 67-72

Regardless of what you think about Ken's methodology, the fact is that the Hoyas are likely underdogs in four games the rest of this month, and will likely lose multiple times.  The question, to me, is not how the Hoyas do in this next stretch of five games, but rather how the team responds after this death-march.

To be clear, the remaining schedule is no cake-walk with two tough road games (Louisville and West Virginia) and several other toss-ups.  Indeed, right now Georgetown's RPI strength-of-schedule projects to be #2 at the end of the regular season.

This all got me to wondering how this year is stacking up against last season, at nominally the same point.

My favorite comparison for this sort of thing is performance.  Simply put, 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 net performance for that game is +4 points.

I usually update the performance charts (linked at upper-right) after each game, but I won't hold it against you if you've never looked.  Here's team performance for last season and so far this year (click on any figure to enlarge):

What's immediately obvious to me is that there's a lot less variability game-to-game.  The 2008-09 team could be tremendous (+20 points above expected against Maryland, Savannah St. and UConn) or lousy (-15 points below expected against West Virginia, Louisville and St. John's (II).  And remember, these expected point spreads are worked out using end-of-season stats (i.e. after the great collapse).

Meanwhile, this season the Hoyas have played every game within 10 points of expectation - their best game was +6 points vs. DePaul, the worst was -9 points vs. Old Dominion.

Allow me to quantify this variability, going even further back for context:
Season        Games          SDev. of Perf.
2007-08       First 17       ±  8.6 points
2007-08       Last 17        ±  9.4
2008-09       First 15       ± 14.6
2008-09       Last 16        ±  9.3
2009-10       First 15       ±  4.3
It looks like typical variability for a ~16 game stretch is around ±9 points; that's to say that most games (about 2/3) should end up within 9 points of KenPom's predicted outcome, adjusting for actual game pace.

As you can see, the early season last year was all over the place.  I'm not sure that was some sort of harbinger of impending doom, but it does show that you never knew what kind of performance Georgetown would bring to each game (despite going 12-3).  Also interesting to me is that the Hoyas only played 2 games all year slightly better than expected (here defined at 0-10 points above expected).  Essentially, either the Hoyas were very good - mostly early in the season - or they just weren't good at all.

So far, this season is another matter entirely.  Most games the Hoyas are performing within ~4 points of what KenPom tells us to expect - good news for gamblers, I suppose.  As the great Dennis Green would say, they are who we think they are.

But why?

As my loyal reader knows, I can never leave well enough alone.  Instead of just looking at net performance, I can also look at offensive performance (points scored - expected points scored) and defensive performance (expected points allowed - points allowed).
.                            SDev. of       SDev. of
Season        Games          Off. Perf      Def. Perf
2007-08       First 17        ±  6.9         ±  5.8
2007-08       Last 17         ±  7.7         ±  6.8
2008-09       First 15        ± 10.2         ±  8.3
2008-09       Last 16         ± 10.5         ± 10.6
2009-10       First 15        ±  7.0         ±  5.7
What we see early (the first three lines) is that the variability of the offensive or defensive performance is less than the overall performance.  That's what we'd expect if the two are uncorrelated or vary together (i.e. when the team is playing well on offense, it's also playing well on defense).

Also, for four of the five cohorts the variability of the defense is smaller than the variability of the offense.  This also makes intuitive sense, along the lines of the baseball adage that "speed never slumps." A several game hot- or cold-streak from outside is much more likely to manifest itself in these stats than running into several hot or cold teams consecutively.

However, starting in the 2nd half of last season, and much more strongly now, we see that the variability of the offense and defense are larger than the overall variability of team performance.  What's happening now is that when the offense is playing poorly, the defense plays well (and vice versa).

Excuse me while I whip this (graph) out:

as compared to previous times:

If I were a clever fellow, I'd be able to tell you why this is happening.  Maybe it's just small sample size.  I don't think it's related to rotations, especially this year (what rotations?).

But if the trend continues, it means that, at the end of the game, Georgetown is performing about what KenPom's stats predict.  And he's predicting a whole lot of pain over the next two weeks.


  1. So then do kenpom's game predictions use the same standard deviation for every team?

  2. You'd have to ask Ken that, but I suspect he does work it out for each team (i.e. his consistency stat)

  3. Brian,

    I appreciate these posts of yours, and I appreciate you doing your best to set expectations appropriately. If I can look at the glass half full for a moment, I'd say that the good news is that if we really can set our expectations for a 1-4 stretch over the next 5, we're likely to be pleasantly surprised. Because Ken projects a nice cushion for Rutgers and 4 very close losses, the total projection for these games is much more likely to be 2-3 than 1-4.

    I'm a bit rusty on discrete probability distributions, but by my math, our odds look like this:
    5 wins - 0.8%
    4 wins - 7.7%
    3 wins - 27.3%
    2 wins - 40.9%
    1 win - 22.7%
    0 wins - 0.7%

    Now, those positive numbers are likely inflated by the fact that Ken gives us a 97% chance to win at Rutgers, but still, I think 2 and 3 isn't too much to hope for.

  4. Hey Canis,

    I totally agree with your sentiment.

    In essence, the Hoyas have 1 "easy" win and four games as the underdog to try to steal win(s). I'd honestly be disappointed (and a bit worried) if they don't win at least 2 games in the next 5 - as you point out, the odds of them winning at least 2 games is better than 3 out of 4.

    But, as I said at the top, the interesting part comes after this 5 game stretch, once everyone starts screaming that they Hoyas are fading once again.

    If they can close out the year 8-1 or 7-2, it'll shut up a lot of the naysayers.

  5. Agreed--I don't think lightning strikes twice though. None of our players are going to take anything for granted after last year.

    Updated percentages for remaining 4 games after the Nova game:

    Win all 4 - 2.1%
    Win 3 - 17.3%
    Win 2 - 44.2%
    Win 1 - 35.4%
    Win 0 - 1.1%

  6. Pls continue to update these probabilities as we march through this stretch. Thanx.

  7. Jack,

    I'm almost dead certain his predictions use a common distribution for all teams.

    I know he does consistency, but I think that's an end stat, not an assumption in his predictions (which it probably should be).