THE BOOK--Playing The Percentages In Baseball

Thursday, December 17, 2009

Is the number of PA a player gets in a season or in several seasons a proxy for true talent?

I think so.  If you are doing a projection - say a basic Marcel - and you regress your players to the mean of a population of similar players by age, you will over-shoot your projections for players with few PA per season and undershoot for players with lots of PA per season.  If you use size (height and weight) for your population of similar players, you will still have a problem I think, but it might not be as severe (because smaller players who have less offensive true talent will tend to be backups with few PA and the bigger, stronger players will tend to be your full-time players with more PA per season).

How much difference does it make?  I did a basic Marcel, weighting the last 3 years, 3/4/5, with age adjustments, and regressing with 1200 (which is like 300 without the 3/4/5 weighting) PA of league average (for that age) offense.  For players with less than 300 PA in their last year, the projection overshoots by 3.7 runs per 150.  For less than 200 PA, 4.3 runs.  For more than 300, it undershoots by .8 runs.  For more than 400 PA, 1.2 runs, and for more than 500 PA, 1.7 runs.

The moral of the story is that when you do your projections, use very different means depending on a player’s PA per season.  Obviously if those PA are depressed because the player was injured, this might not apply.  And if a player only plays full time because he is a good defender, it might not apply either. But in general, using PA as a guide to your population mean is very important.

To put in another way, if we take all players who have less than 300 PA in any year, what is their average lwts in the next year, compared to players with more than 300 PA in any one year?  The reason I am looking at the next year is that if a player has a bad year, his PA might get depressed because he is benched, sent to the minors or retired.  So if we want to determine the true talent of players who have few PA in any one year or years, we have to look at offensive performance in out-of-sample years.

So, for example, from 1950 to 2008 players who had more than 300 PA in any one year, had an average lwts of 2.9 the next year if they played the next year (for more than 500 PA, it was 5.5 runs).  If they had less than 300 PA, it was -7.6 runs!  That is a huge difference.  Basically part-time players are bad offensive players and you cannot regress their career numbers toward that of an average MLB player (for that age or size or whatever) when you do a projection.  Same thing for full-time players.  Basically teams know who the good players and bad players are independent of their stats and they use that true talent estimate to determine playing time.  So even if a part-time player has a really good year in 200 PA, his team is generally not going to give him 500 PA the following year.  Or if a full-time player has a terrible year, he is still likely to be a full-time player the next year.  Your projections must reflect the fact that the team knows that the part-time player is likely not really a good player (true-talent-wise) and that the full-time player just had a fluky bad or injury-plagued year and he is still a good player.

Tango, I would guess that if you looked at everyone’s projections from last year, you would find that the average forecaster over-projects the part-time players and slightly under-projects the full-time ones.