1) Rank the lineups in terms of +/- per possession, counting only the lineups that surpass a certain minimum total minutes played barrier.
2) Scale the +/- per possession numbers all to zero. For instance, if the lowest lineup is -3.5 and the highest lineup is +3.5, these change to 0.0 and 7.0, respectively.
3) Sum the scaled +/- per possession numbers.
4) Compute the (scaled +/- per possession)/(sum scaled +/- per possession) fractions for all qualifying lineups.
5) Compute the (total minutes played)/(total minutes available) for all qualifying lineups.
6) Compute the differences in the values found in step 4 and step 5 for each qualifying lineup.
7) Sum the absolute values of the differences found in step 6.
8) Rank all teams based on the values found in step 7. The biggest values represent the least efficient lineup efficiency ratings, whilst the values closest to zero represent the most efficient lineup efficiency ratings.
Obviously one could argue that mathematically, the ideal situation would be to play the most efficient lineup all the available minutes, but this is obviously not realistic. I think the steps I have outlined help to mitigate this problem somewhat. One other major problem I see already, however, is the problem introduced by players moving between multiple teams during one season. For instance, if the best lineup for a team includes a certain player that gets traded halfway through the season, it will be more difficult for that team to earn as high of a lineup efficiency rating based on the steps I have outlined.
Thoughts?
Has anyone computed a coach-based lineup efficiency rating?
Re: Has anyone computed a coach-based lineup efficiency rati
Somebody did something similar within the last year. I don't immediately recall where. Might have been at WOW or maybe it was Ian Levy at his site.
Re: Has anyone computed a coach-based lineup efficiency rati
I think it would be really interesting to see how coaches rank at maximizing lineups in this type of manner.
Re: Has anyone computed a coach-based lineup efficiency rati
Kathoro, have you actually tried that method or were you just suggesting the idea? My method had some interesting results but was ultimately pretty crude. The results were not very stable from season to season and were greatly skewed by both injuries and mid-season trades. I haven't really been back to look at that work again recently but I never really arrived at a way of controlling for those factors.
Re: Has anyone computed a coach-based lineup efficiency rati
I have not actually tried the method, but I think the results could be really insightful. I think I lack the experience with and access to lineup and injury data that would be required to carry out the necessary calculations outlined in the method.
I think I may have created somewhat of a solution to the injuries and trades problem. I will re-outline the method below including the changes made to account for injuries and trades. If you see any fundamental flaws in the reasoning I would be happy to listen and try to make the method more refined.
1) Set the minimum minute fraction necessary for lineups to qualify.
2) Identify all qualifying lineups.
3) Calculate the +/- per possession for all qualifying lineups.
4) Scale the values obtained in step 3 to zero.
5) Sum all the values obtained in step 4.
6) Divide each value obtained in step 4 by the value obtained in step 5.
7) Accounting for injuries and trades, calculate the maximum minutes available to each qualifying lineup.
8) For each qualifying lineup, divide the total minutes played of the lineup by the corresponding value obtained in step 7.
9) For each qualifying lineup, find the absolute value of the difference between the value obtained in step 8 and the value obtained in step 6.
10) Sum all the values obtained in step 9.
11) Rank each team based on the corresponding value obtained in step 10.
I think I may have created somewhat of a solution to the injuries and trades problem. I will re-outline the method below including the changes made to account for injuries and trades. If you see any fundamental flaws in the reasoning I would be happy to listen and try to make the method more refined.
1) Set the minimum minute fraction necessary for lineups to qualify.
2) Identify all qualifying lineups.
3) Calculate the +/- per possession for all qualifying lineups.
4) Scale the values obtained in step 3 to zero.
5) Sum all the values obtained in step 4.
6) Divide each value obtained in step 4 by the value obtained in step 5.
7) Accounting for injuries and trades, calculate the maximum minutes available to each qualifying lineup.
8) For each qualifying lineup, divide the total minutes played of the lineup by the corresponding value obtained in step 7.
9) For each qualifying lineup, find the absolute value of the difference between the value obtained in step 8 and the value obtained in step 6.
10) Sum all the values obtained in step 9.
11) Rank each team based on the corresponding value obtained in step 10.
Re: Has anyone computed a coach-based lineup efficiency rati
I'll give it a shot Kathoro. It may take may a little while to get to it though, I have a few other projects stacking up. I'll certainly bring the results here though when I'm done.
Re: Has anyone computed a coach-based lineup efficiency rati
I'm glad to see you are interested at giving this thing a go. However, I think I would really need to start thinking hard about this for a few weeks in order to work out all the kinks in the method. I already noticed one major flaw. With the way the method is currently set up, the +/- per possession of each lineup is divided by the total +/- per possession for all lineups in order to get a +/- fraction. Suppose a potential midseason trade opens up the possibility of an excellent lineup. Suppose that before the trade, the top lineup was operating at a +/- of +5 per 100 possessions. Suppose that after the trade, a lineup becomes available that operates at +12 per 100 possessions, and the original top lineup becomes unavailable for the rest of the season due to the trade. The +/- fraction of the +5 lineup will be far too low to give a reasonable representation of how many minutes it should have played, because during the time it was available, the +12 lineup was not yet available, yet the +5 lineup is still getting divided by a sum that includes the +12 lineup.
I have thought of somewhat of a solution to this problem. Each lineup needs to be somewhat isolated in the calculations. The maximum minutes possible of a qualifying lineup could be determined. Then, each other qualifying lineup that overlapped to any extent with the qualifying lineup of interest needs to be identified. The +/- per possession values for each qualifying lineup can then be scaled to zero. The scaled values can then be multiplied by the corresponding fractional overlap factors. The fractional overlap factor for the lineup of interest would be 1.00, and the fractional overlap factor for the other qualifying lineups would range from >0 to 1.00 depending on the extent of overlap with the lineup of interest. After multiplying each scaled value of the qualifying lineups by the corresponding fractional overlap factors, the resulting values could be divided by the sum of the values to determine an estimate of the fraction of available minutes that each lineup should have played.
I am sorry if this explanation seems really confusing. If I have enough time within the next month I could really think harder about any more of the potential kinks in the method, really try to come up with something that might give some worthy results, and also type up an intricate explanation that explanes everything better.
I just think it's a good a idea to hold off on this for a while so you don't risk doing lots of work doing calculations that aren't thought through thoroughly enough.
I have thought of somewhat of a solution to this problem. Each lineup needs to be somewhat isolated in the calculations. The maximum minutes possible of a qualifying lineup could be determined. Then, each other qualifying lineup that overlapped to any extent with the qualifying lineup of interest needs to be identified. The +/- per possession values for each qualifying lineup can then be scaled to zero. The scaled values can then be multiplied by the corresponding fractional overlap factors. The fractional overlap factor for the lineup of interest would be 1.00, and the fractional overlap factor for the other qualifying lineups would range from >0 to 1.00 depending on the extent of overlap with the lineup of interest. After multiplying each scaled value of the qualifying lineups by the corresponding fractional overlap factors, the resulting values could be divided by the sum of the values to determine an estimate of the fraction of available minutes that each lineup should have played.
I am sorry if this explanation seems really confusing. If I have enough time within the next month I could really think harder about any more of the potential kinks in the method, really try to come up with something that might give some worthy results, and also type up an intricate explanation that explanes everything better.
I just think it's a good a idea to hold off on this for a while so you don't risk doing lots of work doing calculations that aren't thought through thoroughly enough.