Player Variance by Run Environment
I’ll apologize in advance, because I don’t draw any stunning conclusions from this post. I am, however, going to present the data from my most recent toilings with baseball data. I was reading Wendy Thurm’s most recent article, and I noticed a commenter that pointed out the positions’ offensive output was more similar and closer to overall league average in the last few years compared to the past 20 years. My immediate reaction was to blame or credit the low run environment. My thoughts are that a higher run environment would produce more random variation for each player, which in turn would produce more variance in the entire league.
There are a multitude of factors that affect the talent distribution aside from the possibility of run environment such as training, performance-enhancing drugs, expansion, wars, and talent evaluation. With my quick exploration, I was not able to take these into account, so the following data visualizations serve as more exploratory analysis than any conclusive analysis.
I found the variance of a handful of offensive stats among players with more than 500 PA and plotted them against the run environment. I included the seasons from 1900 to 2014 for NL and 1901 to 2014 for AL, but I removed the 1981 and -94 season owing to the strikes in those particular years. You can change the stat and the league. I also included a histogram to show the shape of the distribution of the stat for a particular year.
The time series plots point to a correlation between a high run environment and high variance between players. This has been especially true over the last several years, where most stats’ variance decreases as the run environment has decreased.
The histograms are interesting as well if you want to see how the talented has been distributed. Overall, baseball talent is not normally distributed. It does become somewhat normal when you increase the plate appearance (PA) limit. I used only players over 500 PA in a given year, so most of these histograms (particularly BABIP) are normal. Before applying the PA filter, I saw a lot of right-skewed distributions, particularly for WAR.
I build things here.
Seems to me like there’s been an emphasis on outfield defense recently. A while back nobody cared what your LF or RF did, as long as they could hit dingers.
Along these lines, I’d be interested in a study of individual player variation, from year to year or even month to month. I.e., are some players more consistent than others? I know it’s complicated by injuries, especially the kind that are played through without anyone on the outside maybe even being aware of them (and this itself is of course a relevant info for player evaluation), but I think some players probably have lower SDs in their performance over time than others.
wRC+ is such a weird scale though. its range depends on the average, and it also doesnt account for the spread in talent (as noted in the article you link)
furthermore, its a nonlinear scale! below 100 is bounded by zero, and above 100 its unbounded. that totally messes up the frequency distribution because all the numbers (players) below average are compressed to 100 units of wRC+, but above average they are spread over an infinite range. i think its a poor choice for comparing seasons, and really for anything at all other than the very specific question of by what percent of the average was a player above or below average. and why is that question actually useful? players performance should be taken in context with the entire population, not just the average.
The way I understand wRC+ is as a percentage. It is not possible to be more than 100% worse than average, but it is possible to be 3 times (200%) better than average.
Although for pitchers they use ERA- and FIP-, which means you can be only 100% better than average, but you can be infinitely worse.
So I guess you are right, it is weird.
wRC+ is bounded both ways though: a single at-bat wRC+ will show that. An out is a -100 wRC+, and the highest possible wRC+ in 1 at bat is obviously a HR. I can’t remember the exact wRC+ of 1 AB, 1 HR player, but it’s definitely bounded.
This suggests that the benefits of higher-run environments don’t accrue equally across the board–Willie Bloomquist in 1999 looks a lot like 2014 Willie Bloomquist, but Bryce Harper in 1999 is a lot more productive than Bryce Harper in 2014. (!)
The variance is much less at the height of the steroid era in the AL than the NL. Perhaps this is because McGuire, Sosa and Bonds were all in the NL which might be best named as the steroid league in that period. None of the AL HR records were smashed in this period even though the R/G were higher in the AL due to the DH. Of course, steroids were prevalent in both league so its curious. Perhaps just random variance, or maybe the better pitching was in the AL, which may also have been random .
I do believe the reasons the variance is so low nowadays is not so much due to the low run environment, although it makes sense the SD would decline proportionately with a lower mean and vice versa, but its due to the fact the talent pool relative to the number of teams is much higher than it was due to wars, segregation and expansion.
PED use, assuming its still prevalent, has been more standardized in terms of drugs and dosage due to the need to pass the testing, which will reduce the variance due to PED’s. Dosages might so to low now that even players who use PED’s experience minimal benefits relative to peers who used DHEA (banned last year), creatine, HGH (testing started from 2 years ago), etc or nothing at all, unlike in the Bonds years (although the percentages of PED users may have been higher there was likely a vast spectrum of drug quality and dosages being used)