Link to last week's Power Rankings.

As you can guess, the comments are fairly unrestrained.

Anyone want to shed some light on the backstory to this one? Was it really a result of last week's ranking of the Red Sox at #7 and the O's at #20 despite their records? That would seem a little...stupid, to me at least.

Anyone want to shed some light on the backstory to this one? Was it really a result of last week's ranking of the Red Sox at #7 and the O's at #20 despite their records? That would seem a little...stupid, to me at least.

I guess the editors at SI got tired of being mocked.

edit: The A's were 17th last week. This is a team with the 5th best record in baseball so despite the fact I get WHY fangraphs ranked them so low, it looks kind of ridiculous to a mainstream audience and, in fact, may be ridiculous...

It doesn't seem stupid to me. It seems stupid to have "power rankings" that are just a list of teams ordered by WAR Wins. I've never even heard of WAR Wins. Last week's rankings might be more predictive, maybe, but they should be able to fudge them for things like recent performance, new players, etc. At the least.

The Fangraphs Rankings also listed KC as the #7 team in baseball when they were 3-13 and had lost 11 straight. This goes back well before last week's list.

Even though I insinuated last week's rankings were the straw that broke the monkey's back, I really have no idea what happened.

The problem with the fangraphs WAR rankings is that they aren't what people want out of power rankings. Assuming all of the regressing of peripherals comes out in the wash on a team level, they are basically just sorting the teams by pythag records.

It doesn't tell us anything novel about the current composition of teams or their ability to win games.

This approach basically says that if all teams had rosters that mimicked the same composition of players as they've had year-to-date (injuries or trades or callups be damned), and the same performances theyve had so far (unsustainably good or bad performances be damned), here's how we'd rank teams by expected run differential.

I know we had a discussion on this when it was first introduced, and IIRC, the whole concept was panned pretty much across the board. So if we hated it, it's not surprising to me that the general baseball public hated it even more.

The comments in last week's piece are fantastic. The number of people who have hated the system all year but keep reading it amuses me to no end.

Obviously something was very wrong with the way they were calculating them. Counter-intuitive results should not necessarily be discarded but I think you should be able to justify them and if you can't then you probably have a bad result.

The comments in last week's piece are fantastic. The number of people who have hated the system all year but keep reading it amuses me to no end.

I enjoyed the rankings. I was hoping the A's would make the playoffs but still be ranked 25th or something.

The entry for the Red Sox at #7 calls them a "massive failure" but talks about how good position they're in with the recent mega-trade, despite missing Adrian Gonzalez. There's not a single word anywhere in the piece that justifies why they should be the #7 team in MLB right now.

Q: Does this incident align with my growing contempt for the Fangraphs 'mindless numbers' approach to baseball statistics?

A: Why yes, yes it does.

The problem with the fangraphs WAR rankings is that they aren't what people want out of power rankings. Assuming all of the regressing of peripherals comes out in the wash on a team level, they are basically just sorting the teams by pythag records.

The Fangraphs power rankings are significantly weirder than that, and they are not particularly similar to a Pyth record ranking. They're not even really all that similar to a component runs approach.

Team quality in run scoring is measured in batting WAR, which is basically a component runs approach. That's normal enough.

Team quality in run prevention, however, is measured by a sum of FIP-based pitching WAR and fielding UZR. This method is wildly different from even a component runs approach, as it effectively estimates hits allowed based on an assumption of a league average distribution of balls in play. This will never been a good estimate of actual hits allowed, and it produces some huge differences even between Fangraphs team WAR and component runs wins expected.

The comments in last week's piece are fantastic. The number of people who have hated the system all year but keep reading it amuses me to no end.

The SI rankings have been ridiculous all year which is why I gave up on them sometime back in May. I'm glad to see that they've dumped the approach and gone to a more conventional ranking system.

The only decent value of something like this is an adjustment made due to differing strength of schedules. You could calculate a pythag win% for each team and then make sure it's balanced (IE, that when the predicted win% are multiplied by games, the total is exactly a .500 win%) and then run that through KRACH or one of my systems or whatever.

Hell, if someone could point me to an easily downloadable data source with all the scores I could do it in less than half an hour.

In my mind, "power rankings" need to take into account current team composition. Otherwise, they're just rankings based on past stats that anyone can look up.

#13 Why reinvent? Just steal Arpad Elo's work.

JJ1986,

Well that's not easy to do using only objective data. In any event, a time degrader on the results (IE results in the last month count more than results in April) does help that to some extent. Remember SI isn't trying to invent Salk Vaccine here, it's a fun little thing to put out there for people to talk about. Trying to make it more complicated than that probably isn't worth it to them.

You can certainly make it more complicated than that, but you don't have to for it to have some value.

Well because I think ELO is inferior to iterative systems. In some settings it works okay, but in European Club football it simply does not work at all (the champions in Belarus consistently outrank half the Premiership).The biggest weakness in ELO is the inability to look backward. If a low ranked team beats a top ranked team, both teams ratings change a bunch, but if that same low ranked team beats another top ranked team the next week, that team's ratings don't change nearly as much as the first top ranked team despite an identical result against an identical team. Iterative systems can go back and readjust our new expectations for the first match.

Furthermore iterative systems can be continually modified to extract all sorts of additional data (such as in college baseball, extracting park factors from teams who play vastly different schedules in different places). It's possible (just possible mind you) that someone did that very thing while working for an MLB team on the draft.

And since I've been doing these iterative systems for close to a decade now, I wouldn't have to re-invent much at this point. :D

You could calculate a pythag win% for each team and then make sure it's balanced (IE, that when the predicted win% are multiplied by games, the total is exactly a .500 win%) and then run that through KRACH or one of my systems or whatever.

You mean SRS?

#11,

Thanks, I hadn’t thought about it, but you’re right that using FIP plus UZR for run prevention isn’t the same as looking component runs allowed.

The “right” way to do this is to have a projection system that’s updated daily and then project out each teams rest-of-season depth charts and then sim based on rest-of-season schedule or something like that.

But that’s obviously a lot of work, and I think a good, well-informed analyst will get most of the way there based on basic intuition.

You mean SRS?

I haven't actually looked at it too deeply, but IIRC SRS just does like one iteration as opposed to a thousand or so. For MLB purposes the schedules are probably close enough to be "close enough", but for something like College Football or Baseball one iteration wouldn't be sufficient.

But yeah, same sort of thing. It's not anything new, it's just not anything most people normally do. I'm not normal and so that explains quite a bit.

It's iterated, at least if I understand your use of the term correctly until the results don't change from once to the next.

The biggest weakness in ELO is the inability to look backward.

But I can't get it out of my head!

Does Fangraphs use a FIP-based park factor for its pitching WAR? Or do they just the same runs-based park factors for both offense and pitching?

If it's the latter, fWAR is going to overrate pitchers who play in hitters parks that inflate scoring due to a high runs-per-BIP and underrate pitchers who play in pitchers parks that suppress run scoring due to low runs-per-BIP. Let's say Park A has a park factor of 110 while Park B is a 90, and that both parks have the same effect on HR/BB/SO. The only difference is that Park A has a .320 expected BABIP and makes it easier to hit doubles, while Park B is suppresses BABIP to .280 and limits doubles. A 4.00 FIP is equally impressive in either park, even though a 4.00 ERA is much more impressive in Park A than in Park B.

For example, Fenway is a hitters park because it greatly increases doubles. That makes it tougher for a Boston pitcher to post a good ERA, but it doesn't make it any tougher to post a good FIP. If Fangraphs is using regular park factors, fWAR is going to overrate a Boston pitcher with a league average FIP as above average.

Fenway is a hitters park because it greatly increases doubles.

no kidding. The Red Sox are having one of their most disappointing seasons, and they are (barely) within striking distance of setting a new MLB record for most doubles hit by a team. While actually being mediocre at hitting them on the road.
Sox doubles at home 189 (yes, they have played 4 more HG than RG, but still...)
AL avg 2B s at home 116
Sox doubles on road 110
AL avg 2B s on road 111

In some settings it works okay, but in European Club football it simply does not work at all (the champions in Belarus consistently outrank half the Premiership).

For the benefit of people who don't follow soccer: Why is that dumb?

The champions of Belarus outrank half the Premiership because half the Premiership never plays in European tournaments and therefore has no rating at all.

For the benefit of people who don't follow soccer: Why is that dumb?

Well it's roughly the same as rating the South Atlantic League Champions above the Marlins because of their superior record. Elo adjusts fro strength of schedule, just not nearly well enough when the schedules differ so vastly.

As for Sean, than yeah SRS is more or less the same thing, though I probably wouldn't use +/- runs myself. I'd probably convert into a pythag win% and go from there. Theoretically +/- system could underrate teams with very good pitching staffs or defenses.

The champions of Belarus outrank half the Premiership because half the Premiership never plays in European tournaments and therefore has no rating at all.

Actually they'd have a rating based on their matches in the premiership. There actually is some strength of schedule adjustment, but when the schedules differ so vastly, Elo falls apart in ways that iterative systems do not.

But I can't get it out of my head!

Sorry. For some reason I always want to capitalize that even though it's just a dude's name and not the name of a really dodgy prog rock band.

One neat thing about these sorts of systems for baseball is that you can easily fold in playoff results into the system.

Sorry. For some reason I always want to capitalize that even though it's just a dude's name and not the name of a really dodgy prog rock band.

One of the secret shames of my media library.

Does Fangraphs use a FIP-based park factor for its pitching WAR? Or do they just the same runs-based park factors for both offense and pitching?

Same one, or at least they did last time I checked their WAR explanation; it's been a while. And yes, the issue you point out is at least a medium-sized problem with the FIP-based approach.

I'm glad to see that they've dumped the approach and gone to a more conventional ranking system.

Just curious but why? What do people use baseball power rankings for? I can see some point in college sports where teams are playing vastly different schedules and you can have fun speculating about Alabama vs. Boise State. But heck we've now even got interleague and "powerful" or not, we've got a set way of determining the overall champ.

But also ... what is the advantage of the "conventional" system over the fangraphs one? Has the accuracy of the conventional system ever been tested or was it just first on the block? Is the conventional system any good at predicting future performance? Where have they ranked the Cardinals this year? Did the conventional rankings at the end of July suggest that the O's were about to go 20-9 and the A's go 20-10? That Cleveland was about to go 7-24 and the Cubs 8-23?

Does Fangraphs use a FIP-based park factor for its pitching WAR? Or do they just the same runs-based park factors for both offense and pitching?

Same one, or at least they did last time I checked their WAR explanation; it's been a while. And yes, the issue you point out is at least a medium-sized problem with the FIP-based approach.

This suggests that it's the same one for offense and pitching.

Park Adjustments. Replacement level FIP varies depending on the park a pitcher plays in. If a 5.63 FIP was the replacement level for an AL starter, and one ballpark depressed offense by 2%, then the replacement level FIP for that park would be 5.52 FIP — 2% lower than the AL replacement level. Pitchers only play half their games at home, though, so they only need to have their FIPs adjusted by half their home park factor.

Actually they'd have a rating based on their matches in the premiership. There actually is some strength of schedule adjustment, but when the schedules differ so vastly, Elo falls apart in ways that iterative systems do not.

Does the IFFHS use a system with issues similar to Elo? The recent headlines were about how Monterrey ranked above ManU and Arsenal, but even more astounding is that Club Universidad de Chile is #2 Boca Jrs is #4 and, Club Libertad Asunción of Paraguay is #9.