Every major fantasy league hosting site (Yahoo, CBS, ESPN) allows you to look at recent history (e.g. the last 7 days or the last 14 days). It’s also very easy to see the year-to-date stats any player has accumulated to this point in the season.
And now that we’re nearly half way through the current season, how much do those current year stats mean? If you’re trying to add a free agent, should you be looking at the last 7 days? Is the last month OK to use? How much can we expect production from the first half of the season to continue into the second half?
Let’s Take A Quiz
Before we get to the answers to those question, let’s put you to the test with some very specific questions. I’ll lay out a series of “story problem” (remember middle school math?) questions for you . Place yourself in each situation and make what you think is the best fantasy baseball decision.
Question #1
Your team recently suffered an injury and you must go out to the free agent list and find a replacement. Which of these measures is the best method of identifying the player who will perform the best for the rest of the season?
- Looking at the statistics for free agents in the last 7 days
- Looking at the statistics for free agents in the last 14 days
- Looking at the statistics for free agents in the last 28 days
- Looking at the statistics the free agents have accumulated to this point in the season (season-to-date stats)
- Looking at the projected statistics for free agents for the remainder of the season (like Steamer or Zips rest-of season projections)
Question #2
Which model(s) above do you actually use to make decisions?
Question #3
Which player would you rather have the remainder of the season given these levels of production so far?
Current production (as of 6/22/2014):
Player | PA | R | HR | RBI | AVG |
---|---|---|---|---|---|
Nelson Cruz | 306 | 45 | 23 | 60 | .299 |
Chris Davis | 252 | 32 | 12 | 37 | .220 |
Question #4
Which player would you rather have the remainder of the season given these levels of production and the Steamer RoS projections below?
Current production (as of 6/22/2014):
Player | PA | R | HR | RBI | AVG |
---|---|---|---|---|---|
Nelson Cruz | 306 | 45 | 23 | 60 | .299 |
Chris Davis | 252 | 32 | 12 | 37 | .220 |
Steamer RoS Projections (as of 6/22/14):
Player | PA | R | HR | RBI | AVG |
---|---|---|---|---|---|
Nelson Cruz | 320 | 41 | 17 | 47 | .261 |
Chris Davis | 341 | 46 | 20 | 52 | .261 |
Question #5
Similar scenario to question four above… But now imagine that we’re five full months into the season instead of at roughly the half way point. Who would you rather have in the final month of the season?
- The player who was incredibly hot for the first five months but that projections say will cool off towards his career averages or
- The player that has struggled for the first five months but is projected to improve and perform closer to his higher level of career averages over the final month of the season?
Question #6
Which player would you rather have the remainder of the season given these levels of production and the Steamer RoS projections below?
Player | IP | K/9 | ERA | WHIP |
---|---|---|---|---|
Andrew Cashner | 76.1 | 6.96 | 2.36 | 1.19 |
Homer Bailey | 90.0 | 8.07 | 4.68 | 1.45 |
Steamer RoS Projections (as of 6/22/14):
Player | IP | K/9 | ERA | WHIP |
---|---|---|---|---|
Andrew Cashner | 103.0 | 7.29 | 3.85 | 1.27 |
Homer Bailey | 95.0 | 7.99 | 3.80 | 1.22 |
The Research
The information that follows is a summary of the findings by Mitchel Lichtman and posted on his blog MGL on Baseball. You should definitely take the time to read the full articles linked below because there is a great deal of additional information I can’t fully incorporate into this post (the first bullet below summarizes his findings for hitters, the second for pitchers).
- What can a player’s season-to-date performance tell us beyond his up-to-date projection? by Mitchel G. Lichtman
- Mid-season projections part II – Pitchers by Mitchel G. Licthman
The two blog posts will challenge a lot of the assumptions you probably have about in-season statistics, stabilization points, and how much we can “trust” that information.
If you’re not familiar with Lichtman, he is one of the authors behind one of the most influential books in the field of sabermetrics that’s ever been published. They aptly named it, “The Book“. While it’s not written with fantasy baseball in mind, there are a lot of concepts with which a strong fantasy player can take from the book (hot and cold players, platoon information, batter vs. pitcher matchups, and more). It’s not for the faint of heart, but I’d consider it a must read for those looking to get to an advanced level of knowledge about the game.
Enough Already, What Did He Find?
Lichtman conducted a study of “hot” (players hitting 40 points over their projected wOBA to start the season) and “cold” players (those hitting 40 points below their projected wOBA to start the season).
He then compared their RoS projections from that point forward to their actual results for the RoS.
Before you look at the results, think about this and how it relates to the scenarios above. He was essentially testing the Nelson Cruz vs. Chris Davis experiment, but on a much larger scale. Should we take the player that’s already hit 23 HR half way through the season? Or should we take the guy who’s only hit 12 HR but that some really advanced projection systems like more in the second half?
Here are Licthman’s results for hitters. Again, I urge you to visit his blog and read the full article or buy his book. Give the man some due.
Scenario | Actual wOBA Start of Season | Projected wOBA RoS | Actual wOBA RoS | Actual PA RoS |
---|---|---|---|---|
“Hot” hitters for first month of season | .412 | .341 | .346 | 494 |
“Cold” hitters for first month of season | .277 | .342 | .343 | 464 |
“Hot” hitters for first 3 months of season | .391 | .339 | .346 | * |
“Cold” hitters for first 3 months of season | .283 | .334 | .335 | * |
“Hot” hitters for first 5 months of season | .391 | .343 | .359 | 103 |
“Cold” hitters for first 5 months of season | .289 | .338 | .339 | 70 |
* Lichtman doesn’t disclose the exact number but does say the hot players got 54 more PA over the remainder of the season
Here’s what he found for pitchers. Please note that the scenarios are slightly different than for hitters. I’ll explain the difference later. And you should know that NERA stands for “normalized ERA”, a measure for which 4.00 represents league average.
Scenario | Actual NERA Start of Season | Projected NERA RoS | Actual NERA RoS |
---|---|---|---|
Good pitchers who were “Hot” for first month of season | 2.49 | 3.78 | 3.78 |
Good pitchers who were “Cold” for first month of season | 5.38 | 3.79 | 3.75 |
Bad pitchers who were “Hot” for first month of season | 3.06 | 4.39 | 4.47 |
Bad pitchers who were “Cold” for first month of season | 6.24 | 4.39 | 4.40 |
Good pitchers who were “Hot” for first 3 months of season | 2.74 | 3.64 | 3.46 |
Good pitchers who were “Cold” for first 3 months of season | 4.60 | 3.67 | 3.84 |
Bad pitchers who were “Hot” for first 3 months of season | 3.53 | 4.43 | 4.33 |
Bad pitchers who were “Cold” for first 3 months of season | 5.51 | 4.41 | 4.64 |
Lichtman conducted his analysis of both hitters and pitchers on two different bases. The first was simply on all “hot” and “cold” players. Then he further broke those two groups into “good players that were hot”, “good players that were cold”, “bad players that were hot”, and “bad players that were cold”. The detail on hitters was not as great for this further breakdown. But he did show the greater level of detail for pitchers, so I chose to include that above.
Conclusions
We Need To Be Using RoS PRojections
I think the biggest take away for us as fantasy players is that we really need to be using RoS projections much more. And we should be completely ignoring year-to-date stats. No matter how gaudy the first half, the numbers show us to trust the projections.
Taking this a step further, making a change like this can be our opportunity to “buy low-sell high”. I hate to use that fantasy cliche. But there is surely an opportunity here.
Many other fantasy owners will be drawing their conclusions on some form of season-to-date statistic, whether it’s 7, 14, 21, or year-to-date production. By switching your focus exclusively to projected RoS stats, you will be able to exploit this inefficiency.
But Don’t Ignore Decreases in Playing Time
Lichtman did find that players in the “cold” groups received less playing time than those in the “hot” groups. You may want to reread his original posts, but the decreases in playing time didn’t seem to be overly significant. But this is certainly something to consider on an individual player basis with the MLB team’s roster circumstances in mind.
The playing time is not presented in the pitching table above, but Lichtman does comment about a decrease in batters faced by “cold” pitchers.
The Projections Are Not Perfect. But Much Better than Season-To-Date Performance.
Take a look at the pitchers who had accumulated 3 months of statistics in the bottom half of the table above.
In each scenario you see that the projections were off a bit from the RoS projections. For both “hot” and “cold” pitchers the projections didn’t give enough credit to the recent performances. For “hot” pitchers it projected them to cool off more than they really did. And for “cold” pitchers it projected them to warm up more than they really did.
But in all scenarios the projection was more accurate than than using current season information would have been.
You may also want to read this summary of Licthman’s articles by Dave Cameron at Fangraphs. Cameron offers some nice additions with his commentary. I especially like the way he concludes:
The evidence suggests the conservative path, leaning almost entirely on forecasts and putting little weight on seasonal performance, is the one that is wrong the least. ~ Dave Cameron, Fangraphs
Thinking About The Questions Above
Revisit the questions I asked above. We don’t have to go through each scenario again, but you might find it beneficial to revisit them. After taking Lichtman’s findings into account, how silly does it seem to make any decision on a 7 or 14 day history for a given player? If 5 months of season data are no more beneficial than a RoS projection , to use a handful of weeks is ludicrous.
If I’m being honest, I often use the last 14 and 28 day stats as a means of reviewing recent playing time. Lichtman did find that “hot” players see greater playing time than “cold” players. But it would be unwise to use these as more than just that.
More About The Projections
Just any old projection system won’t do. We’re not saying to go back to the beginning of the season and pro-rate ESPN’s projected totals for a player.
Licthman seems to speak positively of Steamer projections in the post. He actually used his own projection method to do the above analysis. He doesn’t describe the approach in great detail, but he does allude to the fact that a good projection system utilizes multiple seasons of data and does incorporate current season events into the model. So the current year level of production is not “ignored” completely, it’s just incorporated into the model and weighted accordingly.
Disclaimers
While I think this research is highly applicable to fantasy baseball, there are some caveats to keep in mind.
- There will be outliers that you will miss by following the suggestions here. You will miss the Jose Bautista explosions and you might get burned by the B.J. Upton dumpster fires. But you’ll also benefit by not chasing said outliers. There’s a significant opportunity cost to every outlier chased that never makes the leap.
- Lichtman used wOBA for his analysis. Most of us don’t play in leagues that use wOBA as a category, but wOBA incorporates walks, singles, doubles, triples, and home runs in its calculation. And all of those are positive events that contribute to our standard rotisserie categories.
- This research did not incorporate stolen bases, wins, and saves. Hopefully you can recognize that those categories are a different beast.
What To Take Away From This
There are two main ideas that I will take away from this analysis:
- There is NO point in the season where using current season statistics to make your decisions is better than using rest of season projections.
- We need to stay abreast of ongoing and current Sabermetric research. It applies to our game and is a great opportunity to gain an edge over the competition.
Thanks For Reading.
If you enjoyed this article and would like to help me out, please pass this along to someone else. I realize that nobody wants to share information with the enemy. So E-mail or tweet it to someone not in your league. I can’t thank you enough for trying to help me grow this blog.
Stay smart.
Interesting article. I do wish Lichtman had used a (modified) ERA estimator rather than ERA for pitchers. I don’t think there is any advanced fantasy owner out there who would look at actual ERA (or NERA) to project RoS ERA. He/she would use YTD xFIP or something of the sort and then adjust it for park and defense. I.e., the study doesn’t really show that YTD stats aren’t very predictive; it just shows that using single stat to predict itself during the same season is not a great idea. Anyway, it is a nice starting point.
Thanks for the comment, Evo. I hope you enjoy the site. Yes, I agree with what you’ve said above. In Dave Cameron’s piece he says that NERA is based on components, but after a few minutes searching the web I couldn’t locate a precise definition. Maybe at some point I’ll be technically capable of studying in-season K-rates & BB-rates and running those against the projected RoS measures. It would be interesting to see those results too.