Looking Back at Some 2010 Marcel Projections
When projections begin to be released, everyone seems to have there own opinions on how the projections are wrong. With Tom Tango just releasing his 2011 Marcel Projections, here is a look back at how some various players performed compared to their 2010 projections.
First, here are the top ten projected hitters by wOBA for 2010 and how they actually performed:
The projections overestimated a on a few players (Mauer, Sandoval, Ramerez, Rodiguez), underestimated on some others (Votto and Holliday) and was fairly close with the rest. When looking at the average of the 10 hitters, the number of home runs projected and actual ended up exactly the same at 27. Yep, dead on.
The hitters did under performed their projected wOBA by 0.016. With this small sample of 2010 data, Marcels pin pointed the home run total and were a little optimistic with projecting wOBA.
Besides the hitters, here are the top 10 projected pitchers according to their ERA:
These pitchers did much better in 2010 as group compared to their Marcel projections when looking at their counting stats. They averaged 1.5 more wins across the board along with a almost 40 strikeouts. The key here was that the group was generally healthy throughout the season. Removing one pitcher pitcher from the group (subtracting 13.5 wins and 190 strikeouts from the total) numbers puts the win total almost exactly on projected value and the strikeout closer in line. This group of pitchers is relatively young and healthy and may not need to be regressed as much to the league average.
On the other hand, the ERA value was about identical (3.29 vs 3.27). The pitchers’ overall talent is farirly predictable with a system like Marcels, but the ability to predict the inning pitched is much harder.
As a whole Marcels held up pretty good when looking at 2010 projections and the actual results of the top hitters and pitchers. The key is for other projection systems to take projecting a step further and fill in the extra information that Marcels doesn’t include.
Jeff, one of the authors of the fantasy baseball guide,The Process, writes for RotoGraphs, The Hardball Times, Rotowire, Baseball America, and BaseballHQ. He has been nominated for two SABR Analytics Research Award for Contemporary Analysis and won it in 2013 in tandem with Bill Petti. He has won four FSWA Awards including on for his Mining the News series. He's won Tout Wars three times, LABR twice, and got his first NFBC Main Event win in 2021. Follow him on Twitter @jeffwzimmerman.
Kind of bizarre to see Jair Jurrjens on the list of top 10 projected pitchers but not Cliff Lee.
I am pretty sure Lee was #11, but yea, the list of pitchers was a little weird.
It’s because he started the season injured.
For analyzing projections (and making them, for that matter), wouldn’t it make sense to make every stat a rate stat except for playing time? So look at proj vs actual HR/PA instead of total HR, for example. And it would tell you that Jurrjens would have nailed his Marcel proj for K had he started ~30 games instead of only 20. I guess at some point you just get down to the boring “true skill” numbers, but separating out playing time for projection analysis seems to make sense.
Also, it would be nice if you could tidy up those Excel tables. They’re a little hard to read, the formatting isn’t consistent between the two, and you’ve still got the squiggly mispelling underlines (screencap?). Maybe Appelman could make a “Fangraphs” style for Excel tables and charts so that they are both well-formatted and easy for the writers to work with?
Yes. Separating out rate and playing time makes the most sense for a projection.
@ Jeff: All in all, the Marcels did pretty good. Check out the article here: http://www.fangraphs.com/community/index.php/comparing-2010-hitter-forecasts-part-2-creating-better-forecasts/
I know about both of them. I scheduled this article to be published and then he began cranking them out. It is some out them
Think i’ll stick with Bill James.
Jeff, you really should normalize the projections to their overall mean (weighted by actual playing time) and then compare them to the actuals normalized to their mean (also weighted by playing time).
There is no way for a forecaster to know whether the stats for the whole league will be low or high in any given year, due to the weather, fluctuations in the construction of the baseball, PED use, a good or bad crop of incoming rookies, etc.
The primary reason that the Marcels were low on wOBA for the top batters was NOT because they under-projected them – it was because Marcel uses the last 3 years of offense and pitching for its baseline and 2010 was an anomalously low-offense year. You cannot expect a projection system to know that, if it is due to something other than the talent and playing time of the players.
I want to know what the reasoning behind not having Miguel Cabrera in the top 10 was because I certainly can’t think of one. Coming into 2010 he was just 26 years old and in his previous 6 seasons he averaged a .394 wOBA and was coming off of a .402 wOBA. So despite entering what most people consider the prime era of a hitter he was going to drop off from his career numbers after showing zero evidence that it might be a possibility? I find that highly unlikely.
I forgot to add that I was going to say maybe the explanation would be that they had concerns about him holding up because of his weight but that argument would go out the window after seeing Fielder and Sandoval on the list.
No pitcher projected to win more than 14 games?
That is weird. I think it’s probably because injury can cause a pitcher to win zero games, whereas there is no opposite phenomenon that could see him win 40. I think that depresses all projected win numbers.
Interesting to look back at the top 10 players, but it would be more interesting to me to see how the Marcels did at projecting a good fantasy team. Were there players who were way over- or under- rated compared to other projection systems? If you used the Marcels to put a team together, what was a likely outcome paired against other projection systems?
Marcel did pretty well:
http://tangotiger.net/forecast/results2010.html
Being a somewhat statistics-based site, I was hoping for something done on a bit larger scale than taking a 10 player sample. Sure, 30 at the least, but with Excel and everything else, it would not have taken you more than another 5 minutes to give more quantitatively relevant date rather than just that of 10 position players.
The post made the point on a smaller scale that a big scale post couldn’t.
I’ve done it both ways for public consumption, on a grand scale, and on a small scale. Small scale sells better:
http://www.hardballtimes.com/main/article/forecasting-2006/
***
As an aside, it sounds like you are complaining that someone else didn’t do enough work. You can complain that someone did a bad job wasting your time (which in this case is not true). You can’t complain that someone didn’t do enough.