The Effect of College Experience on NBA Production

When comparing draft prospects, I've found it difficult to deal with the effect of age. For example, Jae Crowder put up much better statistics than Michael Kidd-Gilchrist, but Crowder is also a senior while MKG is a freshman, so it's not an apples-to-apples comparison. To try to get a handle on this problem, I used data from Draft Express (DX) to estimate how a wide range of statistics change as players get more experience in college.

Throughout the analysis, I used pace-adjusted, per-40 minute statistics. DX provides these data for the 2001-2002 through 2011-12 seasons. DX categorizes each player as a point guard, shooting guard, small forward, power forward, or center. Some players are included in more than one position (for example, my friend Will Barton is listed as both a SG and a SF). I analyzed each position separately and allowed the players to be included in the analysis for more than one position.

The analysis is fairly simple. I create an "experience" variable that equals one in the first year I observe the player and increments by an additional year in each subsequent year observed. (DX doesn't make it easy to use class or age.) For each of the variables in the table below, I estimate a fixed effects (FE) model of the performance variable (e.g., points / 40) on the experience variable and a constant (that, by virtue of the FE model, is player-specific). The FE approach helps control for the fact that more productive players leave the sample with less experience (i.e., they get drafted). It does assume that all players improve (or get worse) at the same rate, which could be the source of some debate.

The table below shows the results. Bold numbers indicate statistically significant results. The percentage variables (e.g., 2PT%) are included such that 50 percent = 50.0 (rather than 0.500). I've included all of the basic performance variables, plus some that Ed Weiland uses in his player evaluation criteria. Each number is interpreted as the change in that statistic per additional year the player stays in college.

Variable PG SG SF PF C
Points 1.430 1.483 1.354 1.293 1.277
2PT% 0.611 0.429 -0.021 0.186 0.325
3PT% 0.586 0.675 0.949 1.687 2.962
eFG% 0.666 0.582 0.253 0.278 0.328
TS% 0.862 0.855 0.544 0.596 0.673
Off. Reb -0.036 -0.074 -0.098 -0.051 0.011
Def Reb 0.132 0.135 0.167 0.324 0.400
Tot Reb 0.097 0.061 0.070 0.276 0.414
Blocks 0.008 0.001 -0.013 -0.026 -0.023
Steals 0.005 0.004 0.002 -0.002 -0.017
Assists 0.206 0.126 0.085 0.132 0.159
Turnovers -0.026 -0.020 -0.015 0.024 0.046
A/TO 0.072 0.051 0.037 0.044 0.049
Stl + Blk 0.013 0.005 -0.011 -0.027 -0.040
Ast + Stl + Blk 0.219 0.130 0.075 0.104 0.118
Reb + Stl + Blk 0.110 0.066 0.059 0.248 0.374
FTA/FGA 0.007 0.011 0.008 0.013 0.016

One odd finding is that offensive rebounding tends to get worse (for all but centers) as players get more experienced in college. I don't have a good story for that one. Steals, blocks, and (most surprisingly) turnovers don't tend to improve as players stay in school. Scoring improves quite a bit. For example, a senior SG can be expected to score about 4.4 points per 40 minutes more than a freshman SG.

An application to a couple of player comparisons might be instructive. First I'll compare Marcus Denmon (Missouri senior) to Austin Rivers (Duke freshman). As expected, Rivers scoring goes way up, but the other stats are not dramatically different. He gets enough of a bump to pass one more of Weiland's criteria, but Denmon still looks like the better prospect. (Note: these are the pace-adjusted per-40 numbers straight from DX. Therefore, they do not match the values that I produced for my analysis of performance against top-100 teams.)

Player P40 2PT% A/TO RSB40 STL40 # Criteria
Denmon 20.3 52.6 1.85 7.6 1.8 5
Rivers as Fr 18.0 47.7 0.89 5.0 1.1 1
Rivers as Sr 22.4 49.0 1.04 5.2 1.1 2

Next I'll compare Jae Crowder (Marquette senior) to Michael Kidd-Gilchrist (Kentucky freshman) and Quincy Miller (Baylor freshman). The statistics are shown in two different panels for clarity.

Player P40 eFG% REB40 ASB40 A/TO # Criteria
Crowder 20.0 57.0 9.6 6.4 1.71 5
MKG as Fr 15.3 51.0 9.5 4.8 0.86 3
Miller as Fr 17.1 48.1 7.9 4.4 0.79 2

Player P40 eFG% REB40 ASB40 A/TO # Criteria
Crowder 20.0 57.0 9.6 6.4 1.71 5
MKG as Sr 19.4 51.7 9.7 5.0 0.97 5
Miller as Sr 21.2 48.9 8.1 4.6 0.90 3

The added experience helps MKG pass the final two of Weiland's criteria, but he goes from barely missing to barely meeting the ASB40 criterion, without much change in the value. (The threshold for ASB40 is 5.0.) MKG and Miller catch up to Crowder on some of the criteria (points for Miller, rebounds for MKG), but Crowder still has the best statistics after the experience adjustment.

What does this all add up to? Nothing definitive. After all, players can improve at different rates and having better college statistics doesn't necessarily make a player a better NBA prospect. However, I have had the feeling that some of the age adjustments that draft geeks make are too large. These results indicate that, on average, players who stay in college improve, but not dramatically. It's not clear to me that better prospects improve more rapidly than average college players (I offer Harrison Barnes and PJ3 as anecdotal evidence), which is my main concern with the methodology presented here.

I'd be curious to hear what you think, both in terms of whether it's instructive to adjust college stats for experience in this way, and whether you think there are major problems with the methods I've used to do so.