2022 Projection Accuracy: Hitter Rate Stats

It’s time for another examination of projection accuracies while I finish up hitters. I will focus on rate stats along with a different look at the RBI and Runs totals. The results are just slightly different than last year and now the aggregators take the top

From last season, here was my conclusion on the rate stats.

Hitter Rate/Counting Stats

For the stand-alone projections, THE BAT’s stand out (I’d use THE BAT X). Just below THE BAT are all four aggregators.

As for the background on how I’m comparing the projections, it can be found in this article. Additionally, I examined the counting stat accuracy in another article.

Besides the batting average, I turned adjusted the projected counting stat totals to be in line with the actual paying time. For example, if a hitter was projected for 80 RBI in 500 AB but only had 250 AB, the projection got dropped to 40 RBI.

When comparing the raw counting stats, the projections that hit the playing time estimates the best, dominated the other counting categories. With this test, the playing time estimate is removed. And finally, here are the results:

Note 1: Both Mr. Cheatsheet and Rotoholic start with a heavy dose of aggregated projections.

Note 2: CBS and ZEILE got the exact same RMSE values on every comparision.

Sigh … again the Wisdom of the Crowds rules as the aggregators just continue to crush the stand-alone systems. The “perfected” aggregators (ATC and Aggregator #1) didn’t perform as well as simple averages. I believe the reason behind this failure is that stand-alone projections are constantly making changes. The perfect mix from the previous season isn’t the perfect mix this season. Just knowing a few projection creators are trying their best should be enough to include them and create a competitive aggregate.

There are a few examples of when the aggregates don’t (e.g. Steamer/Razzball with home runs) come out on top, but there is almost no reason to not use some three-system combination for the rate stats. Time could then be focused on better playing time estimates. I’m not sure which systems to recommend so I ran one more test on Runs and RBI.

I wanted to focus on hitters with similar playing time projections to the actual results. In this case, 88 players had their average playing time within 50 AB of the actual results. I wanted to see if any standalone systems perform better at guessing these two stats. Or is projecting Runs and RBI a fool’s errand?

There will be some bias with these results because those who performed better in the playing time estimates will have an edge. That said THE BATs are the best standalone systems with ZiPS, Steamer/Razzball, and Davenport performing better than the paid options.

Going forward, the three best options I’d recommend are:

1. Use one of the publicly available aggregators like Mr. Cheatsheet, ZEILE, or ATC. The one item I would insure is that the playing time is an aggregate and not a single user input.
2. Use THE BAT X with an aggregated playing time. Currently, THE BAT X is using ATC’s playing time estimates here at FanGraphs.
3. Create your own aggregate using at least three stand-alone projections. An average of the three of the better stand-alone projections might interest some people since they could know and understand the inputs. I know a guy who can help people aggregate projections. With the median value holding up, I wonder if the median of four values might be the right call. There is no way in hell I’m going back to find out this year but I think I might test it next season.

I’m done with hitters for this year and will start with pitchers.