THE BOOK--Playing The Percentages In Baseball

As you can see, there are some brutal players in the OF who get paid WAY too much money or have no business playing major league baseball at all.

As I have said many times in the past, there is a HUGE inefficiency in the way some teams value corner outfielders and first basemen.

It is because they fail to fully understand positional adjustments, they do not know how to quantify fielding (and of course don’t know how to evaluate it in the first place), and they just LOVE offense (which relates to the other two things) whereas defense is king of an afterthought (for these positions).

king=kind

You just had to have your own main post on this, didn’t you?  Just kidding, but yeah this is great great stuff.

"hope he is using the non position-adjusted ones unless his projections are position neutral.”

Shouldn’t that be the opposite? 

I’m using position specific combinations of the fan report attributes, and the projections are position-specific.  I’m using these values, from a previous post:

1B: Instincts 44%, 1st step 44%, Speed 12%
2B: Instincts 22%, 1st step 22%, Speed 11% Hands 22% release 11% strength 6% accuracy 6%
3B: Instincts 15%, 1st step 15%, Speed 10% Hands 15% release 15% strength 15% accuracy 15%
SS: Instincts 18%, 1st step 18%, Speed 12% Hands 12% release 18% strength 12% accuracy 12%
OF: Instincts 17%, 1st step 33%, Speed 33% Hands 17%

So for outfielders, I’m leaving the arm ratings out and only evaluating their range.

I like the idea of speed scores, I used them last year in my projections and once I get them together I’ll probably re-run these, at least for the players who have no fan report.  As you can see, outfield speed is a big part of the modified fan rating used.

You may be right about using the fans as another datapoint and still regressing.  I’m not sure.  What I like about how I did this is the fan scouting report is more important for players with limited time, and less important for players with 5 full seasons.

My last post did not come up for some reason.  As long as you are using Fan ratings that are “position specific” you are OK of course.  I forgot if Tango’s ratings are as compared to players at the same position or all players.

If Pujols in an overall 40 (where 50 is average for all players, including SS, CF, etc.) then you can’t use that to regress Pujols towards unless the number you are starting with before the regression is a position-neutral numbers.

But if you have Pujols as +5 compared to the average first baseman, and he is an overall “40” where 50 is average for all fielders (at all positions), you would not downgrade that +5.  But if Pujols is a 40 as compared to all first basemen, where the average first basemen is a 50, then you would downgrade the +5.

I am sure you know what to do, I just wanted to make sure that you are reading and using Tango’s numbers correctly.  I forgot if he rates the players as compared to all fielders (so that the average first baseman has a bad rating) or as compared to all players at their position (so that the average first baseman has an average rating).

If you give a real example, it would be helpful.

I think all Rally does is
1. figure the Fan value using the above weightings for each position (say Pujols is a “70” using the 1B weights)
2. figure the average Fan value for each position (say 1B is “40")
3. Take the difference (30)
4. Convert that to runs (say by dividing by 2)

And so, Pujols would be, in this completely made up illustration, as +15 runs.

Yes, that illustration is pretty much the process.

Since I use position specific ratings, it is possible for Joe Crede to rate above Brian Anderson for 3B rating, but Anderson to rank above Crede for outfield rating.

Rally, given that every OF is rated at both CF and LF/RF, you must use the 10-run rule in addition to data at each position (if available).  Do you only use the 10-run rule if a player didn’t play one of the positions at all, or do you use a combination of the data and the 10-run rule for all players?

Second, why not do something similar with infield positions.  For example, you’ve got some players at, say, -5 runs at SS and -1 runs at 2B.  That’s mostly because of small sample sizes, but why not at least throw an adjustment in there because our best-guess, given limited data, is that the player will be better as a 2B than a shortstop.  It’s going to take a lot of data to convince anyone that’s not the case, right?

yeah, I was wondering something similar as Sky.  If a player played multiple positions, say, 2B and SS, how are you using the data to compute his projection at 2B and then again at SS.  I assume you are using the SS data to inform your 2B projection and vice versa.  One way would be to just combine the 2B and SS data adjusting one or the other and then split it up again for the SS and 2B projection.  Another way would be to weight the 2B data more heavily when doing the 2B projection and weight the SS data more heavily when doing the SS projection.  I think that is the better way to do it.

Yes, I use the 10 run rule for outfielders.  For infielders I’m looking at each position separately, as I’m not convince that the skills are as transferable as in the outfield.

But I like MGL’s suggestion #9, different weighting would be a good compromise.  I think I’ll try that.

I’ve reposted the infield (2b/3b/ss) ratings, based on MGL’s suggestion.  I’ve combined the data, weighting the position being projected at 1.0, and others at .5 (2b data counts 1/2 towards ss rating, etc.) SS treated as +5 runs better than 2B/3B, who are equals.

1B projections are unaffected.

Sorry if this sounds stupid, but what are the units for the CHONE defensive projections?  Is it runs saved/cost per 150 games?

Yes, runs per full season.

I’m #12 Anonymous.  Thanks.  I wasn’t sure if the data might be scaled to projected playing time.

I have a theoretical question which may be better addressed elsewhere.  First, I’m pretty new to looking at advanced stats (i.e. this offseason).  I was thinking about doing a defensive Marcel on UZR data (not sure if that’s already published), and I wanted to apply some kind of aging factor.  I’ve seen Tango’s article on SS aging curves, and as I haven’t been able to find any other data, I decided I could apply the data to other positions.  I understand the limitations therein, but I figure a rough estimate based on the demands of a different position would still be better than no regression.

I’m not sure if I’m even going through with that right now, since I came across these defensive projections (I wanted to add a defense element as I build team projections).  But, what would be a fair but simple method for converting to UZR?  If I take 50% of a +2 fielder, I would obviously have less than a +1 fielder.  I also realize that the data are based on the average at each position, meaning that a precise calculation would have to look at BIP data by position by season.  I’m basically asking if there’s a decent rule of thumb (like 10 runs=1 win) that would apply.  I’ll probably just use these projections, maybe mixed with a simpler Marcel.  I just can’t find the time to work out the more complicated stuff.

P.S.  Are aging curves for other positions coming, dropped for now, or published and I missed them?

Oh, and I bought The Book this winter, the first baseball book I ever bought!  I’m fascinated by all of this.

A rough aging factor for defense would show a player losing 1 run per year.  This is close enough for all positions except first base, since they aren’t involved in as many plays.  You could use 0.5 runs for 1B, though I’ve also heard that there is little observed aging effects for 1B defense so you could just ignore it.

For your second paragraph I don’t understand the question.

I didn’t really ask a question in there.  What I was thinking about doing would go like this: suppose a 32-year-old performs at 99% his 31-year-old level, then, I would weigh 99% of his 31-year-old defense as a projection.  I would then do the same for his 30-year-old defense based on the combined aging rates between those years, etc.  That’s what I meant by applying a Marcel method defensively.  With a rough aging factor, I’d have averaged out 5*.99*d31 + 4*.98*d30 + 3*.97*d29, for instance, if those were actual aging rates.  That’s when I realized how complicated it would be with UZR being a fluctuating league-average-based stat..I’d have no idea how to convert 99% of the previous season’s performance into an adjusted UZR value (although I imagine it would be practically identical anyway).

I didn’t really form my thoughts well yesterday, but you gave me a good rough estimate.  Now a dirty method would be to weigh the last few years and then subtract a run from the average.  But I’ll trust these projections over that.

Thank you!!

Don’t trust those projections too much.  They are based on averaged zone ratings from BIS and STATS data.  There are better sources out there, like UZR on fangraphs.  But nobody else, as far as I know, has combined them into projections and posted the results in public.

I’m not going to do it, already have too much on my plate.  So maybe my projections are the best available on the web, for now, but they shouldn’t be.  The opportunity is just sitting there waiting for someone to take it.