Statistical Analysis. Humor. Knicks.

Thursday, April 17, 2014

Reconstructing the Draft

In the last post I argued that the NBA draft is not broken, despite recent complaints about the manner in which it gives bad teams incentive to tank and the manner in which it does not always distribute talent evenly according to record. In fact, in the lottery system there must a tradeoff whereby these opposing flaws are balanced against eachother. Most would agree that it is not acceptable to let tanking run rampant by assigning draft order strictly by record, or to disregard team needs by assigning draft order completely randomly. The trick is to find the best happy medium by considering the severity of each flaw and weighting the system accordingly.

I do believe that the current balance of power between tanking and equitable distribution of talent enforced by the draft lottery is a pretty good one. The system does a good job of reducing the competitive advantage to be had by tanking while simultaneously distributing talent more or less according to need.

Still, that doesn?t mean that the system is perfect, and there has been no shortage of interesting discussion regarding various ways in which the draft ordering process could be improved. (My favorite move so far, suggested independently by John Hollinger and our own Mike K, is to determine draft order by team records around the All-Star game, rather than at the end of the season, in order to discourage tanking.)

Without further ado, allow me to throw my hat into the ring. My plan has 3 independent considerations. It all adds up to a somewhat complex plan that departs from the lottery process entirely, so something like this will not be implemented by the NBA any time soon. However, any one of the three components suggested here could be individually incorporated into the current lottery system, and I think they all make good sense.

I will use the term ?draft score? to refer to some number assigned to each team that determines draft pick ordering. A higher draft score amounts to a higher draft pick. One can think of the current system as assigning draft score purely as a function of team winning percentage, with some weighted randomization introduced by the lottery process. I follow the basic convention of computing draft score by team winning percentage, but suggest 3 modifications.

1. Take conference into account

One of the much-bemoaned consequences of the 2007 lottery is that it seems to have exacerbated the imbalance of power between the Eastern and Western conferences. The West has already been significantly better than the East for years now, and now they are adding not one, but two players who are unanimously held to be franchise talents. It seems as if, before you know it, the East?s top 3 playoff teams will be on a par with the West?s bottom 3. (Actually, sadly enough, it is arguably only the presence of the Pistons that definitively falsifies that statement this year.)

What is going on here? One contributing factor is that, because of the competitive imbalance that already exists, records across conferences are not directly comparable. A .500 record in the West indicates much better team quality than does a .500 record in the East, for example.

Draft seedings should take this disparity into account by adjusting each team?s draft score according to the average quality of its opponents. One method by which this might be accomplished would be to make draft score a function of RPI rather than overall winning percentage. In brief, RPI is a function of a team?s own record, the record of the team?s opponents, and the record of the team?s opponents? opponents. RPI was originally developed so that NCAA teams could more readily be compared, but it can serve our purposes here as well.

Observe the rankings for the NBA?s 10 worst teams by win % and RPI for the 2007 NBA season (t-N indicates that a team was tied for rank N):

win % RPI rank change
Minnesota (t-6)
Portland (t-6)
New York (t-7)
Charlotte (t-7)
Sacramento (t-7)
New York
Portland (t-8)
Philadelphia (t-8)

The effect of ranking by RPI is to shift a number of Eastern conference teams ahead in the line, while also knocking back teams in the Western conference like Seattle and Portland. (But, also note that God-awful Memphis still manages to stay near the top of the heap.)

I have to admit I am not intimately familiar with how RPI works, and it may be that there are superior formulations that control for quality of competition. Nonetheless, whatever the measure turns out to be, something along these lines should be implemented so that teams? rankings are not obscured by how good (or bad) their opponents are. The current system is biased towards perpetuating the pre-existing competitive imbalance between conferences.

2. Take within-season consistency of play into account

Creating a draft score by team record (adjusted for opponent quality) is good and dandy, but there are still some lingering issues. One is the specter of tanking. Another is the anomalous case where a team?s record takes a nose-dive due to injuries to key players in midseason.

For both tanking and freak injuries, we might want to deduct points from a team?s draft score because the team record is artificially driven down by foul play or mere circumstance, rather than purely reflecting the team?s overall quality. (Presumably we want to distribute talent in order to compensate for talent-deprived rosters rather than to compensate for bad luck.) And in both cases, the tell-tale sign that should manifest is that at some point in the season, the team in question begins to lose at a higher clip than it did during the rest of the season.

The basic idea, then, would be to counterbalance the draft score credit given for overall poor play with a draft score penalty for periods of time in which the team played significantly worse than their season average. There are a number of ways the spirit of such a provision might be implemented. For instance, we might split the season into 10 equal segments (throwing out the last 2 games) and calculate a team?s RPI in each segment. (Using RPI or a similar measure should control for variation in the quality of opposition faced in each segment.) For each 8 game segment in which the team?s RPI is below its season RPI, deduct points from the team?s overall draft score by an amount proportional to the difference between season RPI and segment RPI.

A provision like this would directly work against tanking. When a team tanks, there is a period of time in which it plays significantly worse than its usual pace. The effect of tanking on a team?s overall record would result in a gain of some extra draft score points. However, the sudden dip in quality of play would result in a loss of draft score points. Ultimately, the tanking game simply wouldn?t pay.

Likewise, teams would not benefit just because of bad luck with regard to injuries. For instance, the Knicks played at a .500 pace for much of the season, but fell off sharply at the end due to injuries to key players like Lee and Crawford. Thus, the draft ranking that they ultimately received was out of whack with the type of team they were for most of the season, before mere circumstance intervened. Had they actually been able to use it, the 9th pick in the draft arguably would have been a slightly unfair reward for the Knicks. Deducting points from the Knicks? draft score by an amount proportionate to the degree to which that injury plagued late season swoon was below season-long standards would have resulted in a more equitable draft positioning.

One legitimate worry about this idea is that it might discourage teams from improving as the season goes on, because (using the specific implementation described above) such improvement would raise the team?s season RPI and thus increase the penalty incurred for those portions of the season with RPI lower than season RPI. An easy solution would be to give teams bonus draft score points for improving over the course of the season. Alternatively, depending on how the idea is specifically implemented, the concern mentioned here might not be salient enough to give sufficient incentive against improvement in the first place.

3. Take between-season consistency of play into account

This idea suggests that draft considerations should not be given solely on the basis of one season?s outcome. Some teams are bad for a long, long time, and for whatever reason find it difficult to crawl back to respectability. Other teams are new additions to the non-playoff club, fresh off of some period of years of relative success in the league and now on the inevitable downward trend, e.g. due to losing players to retirement or free agency. Still other teams have been good and still are good, but are just coming off one unusually bad season because of key injuries or other unfortunate circumstances. (I?m looking at you, 1996-97 Spurs.)

Should each of these types of teams be given equal draft standing? My intuition says no. For the sake of parity and for the sake of the quality of life of all those long-suffering fans, I would say that a .300 team that has been bad for years should get better draft prospects than a .300 team that has fallen from the grace of a recent playoff run (all else being equal). And both of those teams should get better draft prospects than an elite team that just stumbles to .300 for one season due to unfortunate circumstances.

The basic way to implement this consideration is as follows. Calculate a team?s draft score for the most recent season. Then add draft scores from some set number of previous seasons, weighting draft scores from each previous season less heavily the more distant in time they are. The cumulative draft score for each team is what determines draft ordering. For instance, a team?s cumulative draft score for the 2007 season might be something like DraftScore(2007) + (0.4 * DraftScore(2006)) + (0.2 * DraftScore(2005)).

One obvious rejoinder to this idea is that some teams are just disasters and don?t deserve the extra help; they?ve been this bad for this long simply because of horrendous management. The latter day Clippers and current day Hawks come to mind. But these especially poorly run teams teams would not be differentially benefited so long as the number of previous seasons used in the calculation of draft score were kept reasonably low, e.g. 3 years. Even teams that are run well inevitably suffer periods of 2 or 3 years of pain before beginning to recover when times get tough. The main idea is to give these teams that are ?between eras,? so to speak, a little more consideration than teams that have just completed an era, or teams that are still in the middle of one.