Below is a snapshot of the Web page as it appeared on 4/3/2011 (the last time our crawler visited it). This is the version of the page that was used for ranking your search results. The page may have changed since we last cached it. To see what might have changed (without the highlights), go to the current page.
Bing is not responsible for the content of this page.
APBRmetrics :: View topic - Contract Incentive Study
APBRmetrics Forum Index APBRmetrics
The statistical revolution will not be televised.
 
 FAQFAQ   SearchSearch   MemberlistMemberlist   UsergroupsUsergroups   RegisterRegister 
 ProfileProfile   Log in to check your private messagesLog in to check your private messages   Log inLog in 

Contract Incentive Study

 
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion
View previous topic :: View next topic  
Author Message
acollard



Joined: 22 Sep 2010
Posts: 56
Location: MA

PostPosted: Fri Oct 22, 2010 2:35 pm    Post subject: Contract Incentive Study Reply with quote

This is to serve as a more thorough explanation and technical discussion of the regression I ran and the findings it gave from my post at DailyThunder . It takes some methodology from the article I asked about a couple weeks ago, Playing for Keeps by K. Stiroh.

From prosportstransactions.com I collected contract information for every player who played 500 minutes in at least two seasons from 1998-2010. I broke these down into 3 categories, unrestricted free agents, free agents, and post long contract. I ignored extensions for the pre-contract effects, because it is incomplete to include everyone who signed an extension, everyone who is eligible must be used instead. This gave 4106 player seasons with 1033 contract observations. (Thanks to DsMok1 for the spreadsheet he shared in the Advanced SPM thread).

The basic regression was dvar= ai*age+ai2*age^2+player+rfa+ufa+post

Where player is a dummy variable for each player, rfa is a dummy variable for if the player will enter restricted free agency at the end of the year, ufa is the same but for unrestricted free agents, and post is a dummy variable coding for the year after a contract is signed that is four years or more in length (If I had to guess, I probably didn't catch all of these). The results of these regressions for some dependent variables I picked are below. I didn't use dummy variables for team and year because although they helped improve the R-squared a little, they didn't change results besides WS much (because WS are tied a lot to team effects) and I worried that I might have been overfitting.


Sorry for the lack of table formatting, I don't have time to figure it out now. DT has it formatted nicely.

Code:
   Win Shares   Win Shares/ 48   Mins/game   Usage   TS%   AST%   ORtg   DRtg
RFA   .467 **   0.0132***   0.827   0.249   0.00345   -0.1015   1.123*   -1.078***
UFA   -0.018      0.00438**   -0.745***   -0.203*   0.00125   0.0397   0.348   -0.492***
Post   .638 **   0.00827***   1.095**   0.4402**   0.00143   0.6963**   0.3956   -0.6342*
Age   2.53 ***   0.0356***   7.612***   2.68***   0.02606***   2.518***   4.903***   0.2104
Age2   -0.0472***   -0.00066***   -0.1412***   -0.052***   -0.000451***   -0.0483***   -0.0838***   0.00331
R-squared   0.6418   0.6145   0.6474   0.8148   0.5588   0.9029   0.5471   0.6055


What pops out is that restricted free agents have a significantly significant contract incentive in terms of win share, but unrestricted free agents do not. It looks like this is not due to a change in efficiency (as ws48 is still positive), but probably due the statistically significant reduction in minutes. Going on the rough $2mil per win in free agency, the restricted free agency effect looks like it worth almost $1mil. That's kind of cool.

Another strange thing is that the Drtg is correlated well with both types of free agency (over -1 improvement for restricted free agents???) but not well for Ortg. It doesn't really make sense given that Drtg is a noisier number, but it would at least imply that dudes are hustling more in their contract years.

I'm aware there is a lot of potential for selection bias. One bias I could think of is the fact that restricted free agents are often the same age and the same place in their development. I tried to test that, by finding the mean age of an RFA (24.2) and coding all 24 year olds as a variable called twentyfour, replacing rfa. twentyfour was not statistically significant.

Also, the fact that rarely was there a negative post contract effect was surprising, but thinking about it now, even most signing flops take a couple years to snowball, and its often that they just seem to age faster than expected, not

I wanted to try to scale the factors for player talent, because I thought that maybe the free agency effect would be proportional to the initial value of each stat, but it didn't seem to work well, even though I tried it a couple different ways.

Please let me know what you think of my methods, and what you think of the results. I've got the spreadsheets here, but I had to do the regressions in Stata because of all the dummy variables, so there's not much to look at besides the raw data.[/url]
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370

PostPosted: Fri Oct 22, 2010 3:47 pm    Post subject: Reply with quote

A long time ago, using PER and much less sophisticated methods, I found similar results. However, I did find that average PER decreased in the year after signing a new contract. Here's a link:

http://basketball-statistics.com/playingforthemoney.html
Back to top
View user's profile Send private message AIM Address
acollard



Joined: 22 Sep 2010
Posts: 56
Location: MA

PostPosted: Fri Oct 22, 2010 5:08 pm    Post subject: Reply with quote

Thanks for the link, John.

Two things are different between our studies. I included one-year links, because the one year qualifying offer is considered a new contract and it includes some significant players, and i figure the minutes/season restriction would be good enough. I

i also only included those on longer length contracts (+3 years) because I don't think people often view 2 or 3 year contracts as albatrosses. Here's what my regression gives for PER:

Code:
 per |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         rfa |   .7828513   .2223881     3.52   0.000     .3468214    1.218881
   freeagent |   .1825745   .1020244     1.79   0.074    -.0174617    .3826106
        post |   .5006318   .1752594     2.86   0.004     .1570058    .8442578
         age |   3.054301   .1277551    23.91   0.000     2.803815    3.304787
        age2 |  -.0590878   .0022954   -25.74   0.000    -.0635883   -.0545874
Back to top
View user's profile Send private message
acollard



Joined: 22 Sep 2010
Posts: 56
Location: MA

PostPosted: Sat Oct 23, 2010 5:03 pm    Post subject: Reply with quote

perhaps this table is a little clearer
Code:
    Win Shares       Ws/48        Mpg         Usage       TS%      AST%       ORtg        DRtg
RFA    .467 **      0.0132***     0.827     0.249     0.00345    -0.1015     1.123*      -1.078***
UFA     -0.018     0.00438**     -0.745***  -0.203*   0.00125     0.0397     0.348      -0.492***
Post    .638 **     0.00827***    1.09**    0.440**    0.0014    0.6963**    0.3956    -0.6342*
Age     2.53 ***    0.0356***     7.61***   2.68***   0.0260***   2.518***    4.903***   0.2104
Age^2  -0.0472***   -0.00066***   -0.141***  -0.052*** -0.000451*** -0.0483***  -0.0838*** 0.00331
R^2      0.6418       0.614        0.647      0.814     0.558      0.902        0.547      0.605

about a 5-10% effect on ws/48, but no effect on TS% or AST%... DRtg effect being so significant is still really interesting to me...
Back to top
View user's profile Send private message
Ed Küpfer



Joined: 30 Dec 2004
Posts: 786
Location: Toronto

PostPosted: Sat Oct 23, 2010 6:31 pm    Post subject: Reply with quote

Code:
      Win Shares    Ws/48         Mpg        Usage      TS%           AST%         ORtg        DRtg
RFA    .467   **    0.0132  ***   0.827      0.249      0.00345      -0.1015       1.123  *   -1.078  ***
UFA   -0.018        0.00438 **   -0.745***  -0.203 *    0.00125       0.0397       0.348      -0.492  ***
Post    .638  **    0.00827 ***   1.09**     0.440 **   0.0014        0.6963 **    0.3956     -0.6342 *
Age    2.53   ***   0.0356  ***   7.61***    2.68  ***  0.0260   ***  2.518  ***   4.903  ***  0.2104
Age^2 -0.0472 ***  -0.00066 ***  -0.141***  -0.052 *** -0.000451 *** -0.0483 ***  -0.0838 ***  0.00331
R^2    0.6418       0.614         0.647      0.814      0.558         0.902        0.547       0.605


I have found TABS2spaces to be really useful.

Also, you said DRTG is noisier than ORTG -- is this true? I haven't looked, but I can think of several a priori reasons why this would not necessarily be the case.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
acollard



Joined: 22 Sep 2010
Posts: 56
Location: MA

PostPosted: Sun Oct 24, 2010 8:40 pm    Post subject: Reply with quote

Thanks for the link, I'm sure I'll find that extremely useful.

I'm sorry that I made that conjecture about individual offensive rating vs individual defensive rating without explanation. My reasoning is that individual offensive rating just being ortg= (pts scored by player)/ ( 100 possessions used by player) would show individual player improvement more than defensive rating, which is affected a lot by the teammates around the player in question.
Back to top
View user's profile Send private message
acollard



Joined: 22 Sep 2010
Posts: 56
Location: MA

PostPosted: Sun Mar 06, 2011 9:20 am    Post subject: Reply with quote

[quote="Ed Küpfer"]
Code:
      Win Shares    Ws/48         Mpg        Usage      TS%           AST%         ORtg        DRtg
RFA    .467   **    0.0132  ***   0.827      0.249      0.00345      -0.1015       1.123  *   -1.078  ***
UFA   -0.018        0.00438 **   -0.745***  -0.203 *    0.00125       0.0397       0.348      -0.492  ***
Post    .638  **    0.00827 ***   1.09**     0.440 **   0.0014        0.6963 **    0.3956     -0.6342 *
Age    2.53   ***   0.0356  ***   7.61***    2.68  ***  0.0260   ***  2.518  ***   4.9,.03  ***  0.2104
Age^2 -0.0472 ***  -0.00066 ***  -0.141***  -0.052 *** -0.000451 *** -0.0483 ***  -0.0838 ***  0.00331
R^2    0.6418       0.614         0.647      0.814      0.558         0.902        0.547       0.605


Man... Really wish I had submitted this. I'm not saying it was as good as Arup's or anything, he had a really good presentation, but its interesting that we got somewhat different results with a pretty similar methodology.
Back to top
View user's profile Send private message
Display posts from previous:   
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion All times are GMT - 5 Hours
Page 1 of 1

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © 2001, 2005 phpBB Group