Woo Hoo

[PseudoSABR]

 

That happens when a bad road team unexpectedly beats a good home team three times in a row.

After all, a projection model can only predict expected events... The Twins besting the Orioles three times in Baltimore was certainly not expected, particularly after the New York implosion that immediately preceded this series.

Beating a mediocre O's team on the road a few times shouldn't multiply the odds of a playoff contender so late in the season.    This isn't April and any team within five games of a playoff spot shouldn't see their probability swing so wildly if that probability is an accurate forecaster.   

 

The playoff chance thing is a glib analysis of how many wins do they have now and how many wins do my favorite statistics suggest they should have.  The fact that the odds change so much indicates that its dependent on wins and losses--which doesn't tantalize my higher brain functions all that much. I can already look at their win and loss record in comparison to other competitive teams--they call that the Wild Card Standings. 

[PseudoSABR]

 

The thing with these models is that unless they include the variance of the values (e.g. 7.5% +- 1.5%) the exact numbers are not worth a whole lot. They can order the teams but a quick look at the standings can do that too without any false impression of numerical accuracy.

This. 

[TheLeviathan]

http://2.bp.blogspot.com/-laPHjy0704U/T0qJc6NMzzI/AAAAAAAAA2o/SyN4FXNu03I/s1600/take-it-with-a-grain-of-salt.jpg

[Otto von Ballpark]

The thing with these models is that unless they include the variance of the values (e.g. 7.5% +- 1.5%) the exact numbers are not worth a whole lot. They can order the teams but a quick look at the standings can do that too without any false impression of numerical accuracy.

I don't think anyone is claiming accuracy, beyond KC being a mortal lock. It is a projection.

 

If you can just glance at the standings and come up with better percentage chances for each of the pile of teams still in contention, and each of the ways they could make the postseason, more power to you. But I kinda doubt that is actually true.

[TheLeviathan]

I think you're missing the point.  Eyeballing the standings would probably lead one to rank teams similarly in terms of chances, without any of the silly pretense of "accuracy".

[Brock Beauchamp]

 

Beating a mediocre O's team on the road a few times shouldn't multiply the odds of a playoff contender so late in the season.

Yeah, it really should.

 

The Orioles were one of the teams ahead of the Twins and fighting for the same (lone) playoff spot.

 

The Orioles are a very good home team.

 

The Twins are a very bad road team.

 

It was a four game set.

 

I think you're really underrating just how unexpected it was for the Twins to head to Baltimore and win four games in a row. As "mediocre" as the Orioles may be, one would expect them to beat the Twins in a neutral setting (the Orioles have been the better team this season) and one would expect them to trounce the Twins at home, most likely taking three games and at worst settling for a split.

 

Instead, the Twins entered the series 3.5 games behind the Orioles and are now 0.5 games ahead of them. That leaves just two teams to leap to reach the postseason instead of three. Sweeping a head-to-head matchup against a team well in front of you is going to greatly impact the chances of reaching the postseason.

[USAFChief]

I think if people could predict 40 games of baseball with any accuracy, they'd be too rich to publish their predictions.

 

 

[Brock Beauchamp]

 

I think you're missing the point.  Eyeballing the standings would probably lead one to rank teams similarly in terms of chances, without any of the silly pretense of "accuracy".

If anyone thinks this list is precise in any way, that's kinda silly... But it's almost certainly more accurate than eye-balling the standings.

 

Just because the Twins went out and did something completely unexpected it doesn't invalidate the numbers. Teams do unexpected things all the time and while the numbers are kinda fun to watch, one has to keep their expectations in check because, after all, "that's why they play the games".

[Craig Arko]

I think if people could predict 40 games of baseball with any accuracy, they'd be too rich to publish their predictions.

That's what Pete Rose said.

[USAFChief]

That's what Pete Rose said.

Eventually.

[Otto von Ballpark]

I think you're missing the point. Eyeballing the standings would probably lead one to rank teams similarly in terms of chances, without any of the silly pretense of "accuracy".

Where is there a pretense of accuracy? This reminds me of ZiPS critics getting hung up on Player A being projected for 22 HR and Player B 20. Those are just the projection results, it is not a weakness of the system.

 

And an ordinal ranking by glance at the standings might work for 1 race, but there are ~10 teams across 5 races here. Unless you are a savant, a glance at the standings is not going to produce a more meaningful picture than these projections.

[Otto von Ballpark]

Put another way, the Twins doubled (more than that?) their chance of the playoffs in mere days.

The Twins had won 3 in a row. Two teams previously in the lead for the 2nd wild card (LA, BAL) lost 3 in a row. Another team with a previously identical record to the Twins (DET) also lost 3 in a row.

 

Why shouldn't that move the Twins chances from 4 to 7.5%?

[spinowner]

 

What about overcoming a 12-run deficit-----after the 7th inning stretch??

http://www.baseball-reference.com/boxes/CLE/CLE200108050.shtml?mobile=false

I remember that game. Lou Piniella was managing the Mariners then, and there was a shot of him in the dugout in about the 8th inning. His team still had a good lead but he knew what was happening and he knew what was going to continue to happen and he knew that it wouldn't matter what he did because in a game like that you'll always make the wrong move. I have never seen a more sour look on a human face in my life.

[Otto von Ballpark]

Funny thing, that was the Mariners insanely awesome season, and that comeback helped Cleveland topple the Twins in the division that year too...

[TheLeviathan]

If anyone thinks this list is precise in any way, that's kinda silly... But it's almost certainly more accurate than eye-balling the standings.

 

Just because the Twins went out and did something completely unexpected it doesn't invalidate the numbers. Teams do unexpected things all the time and while the numbers are kinda fun to watch, one has to keep their expectations in check because, after all, "that's why they play the games".

More accurate is a pretty odd claim when we're talking about simulated possible futures. It's not just a lack of precision, it's that the whole concept is dressed up with accuracy at all. Or taken seriously in that regard.

 

Any better accuracy than eyeballing it is nebulous at best.

[Otto von Ballpark]

Eyeballing the standings will give you a vague idea, but there is a meaningful difference between being, say, 2 games out with 3 teams to pass and 40 games left, and 3 games out with 1 team to pass and 20 games left, that you can't quantify by eyeballing. Especially when you can be different numbers of games behind different numbers of teams in different races all at the same time.

 

Nobody is claiming that a team with 8.5% odds has a definitively better shot than a team with a 7% chance or anything crazy like that. These numbers are just an interesting way to quantify the implications of the standings and games remaining that is often difficult to do just by eyeballing.

[PseudoSABR]

 

Eyeballing the standings will give you a vague idea...

Here's the assumption of accuracy.  If eyeballing is merely vague--this forecast must be less-vague, hence more accurate.  (And of course that they call it Playoff Probability certainly suggests that it's forecasting with accuracy; it's why they choice to represent their calculation as a percentage instead of a whole number.) 

 

I actually think looking at the standings provides me with better information about the credibility of the Twins' playoff hopes than does a well-conceived* metric that reduces those odds to a percentage.  

 

As Chief suggests, if they're not meaningful enough to bet money on, they're probably not very meaningful. 

[PseudoSABR]

 

Yeah, it really should.

 

The Orioles were one of the teams ahead of the Twins and fighting for the same (lone) playoff spot.

 

The Orioles are a very good home team.

 

The Twins are a very bad road team.

 

It was a four game set.

 

I think you're really underrating just how unexpected it was for the Twins to head to Baltimore and win four games in a row. As "mediocre" as the Orioles may be, one would expect them to beat the Twins in a neutral setting (the Orioles have been the better team this season) and one would expect them to trounce the Twins at home, most likely taking three games and at worst settling for a split.

 

Instead, the Twins entered the series 3.5 games behind the Orioles and are now 0.5 games ahead of them. That leaves just two teams to leap to reach the postseason instead of three. Sweeping a head-to-head matchup against a team well in front of you is going to greatly impact the chances of reaching the postseason.

If the underlying assumptions about the Twins (bad, especially bad on the road) and O's (good, especially good at home) aren't affected by the sweep, any good forecast shouldn't be affected by such 'variation' even if it occurred in a group of four--a sound forecast would tend to suggest both teams will make up for/or have already made up for that variance. 

 

As you assert in your last paragraph, looking at games-back provides the same kind of swing that playoff probability does--which suggests that the probability isn't any more probabilistic than the standings.   I just don't see a lot of utility of it, and because it presents itself as a probability, it's certainly dressing itself up like it's more useful than looking at the standings.

[Otto von Ballpark]

No one is suggesting that these odds replace looking at the standings. I am not sure why you keep pitting the two things against each other. These odds augment looking at the standings.

[Mike Sixel]

This is the same math used to make billions on the stock market......not to mention most economic models used by governments. Oh, and to predict most everything on the planet. I don't get the hatred of math and science.

[Craig Arko]

This is the same math used to make billions on the stock market......not to mention most economic models used by governments. Oh, and to predict most everything on the planet. I don't get the hatred of math and science.

No, real statistical models include error analysis. And they fess up to it.

 

Here is some real scientific literature. http://arxiv.org

 

What's being discussed here is this: http://www.investopedia.com/terms/s/subjective_probability.asp

[Mike Sixel]

Wait, we don't use monte Carlo simulation anymore?

[Craig Arko]

Wait, we don't use monte Carlo simulation anymore?

Funny you should mention that. My first job out of college involved writing Monte Carlo simulations. In doing them I became quite conversant with this topic:

 

https://en.m.wikipedia.org/wiki/Circular_error_probable

 

As we were trying to compute kill ratios on the ground.

[Brock Beauchamp]

If the underlying assumptions about the Twins (bad, especially bad on the road) and O's (good, especially good at home) aren't affected by the sweep, any good forecast shouldn't be affected by such 'variation' even if it occurred in a group of four--a sound forecast would tend to suggest both teams will make up for/or have already made up for that variance.

Yeah, it should. The Orioles just lost four games in a row, two or three of which they should have won based on statistical probability. That's a huge variance that will cause fluctuation in the chances for both teams, especially considering head to head matchups where gains or losses are absolute.

 

AND let's not forget that while the Twins were shelling Baltimore, Toronto was shelling the Angels.

 

The Twins started this series behind three teams. They ended it behind one. That is going to cause a big change in percentage chances. No model can predict aberrant behavior, which is what happened over the past few days. Toronto should not have swept the Angels at home, Minnesota should not have swept the Orioles at home. With a finite number of games remaining in the season, those statistically improbable head to head matches are going to impact postseason chances.

[Otto von Ballpark]

I don't know about error bars, but it would be interesting to see a breakdown of the projection. For example, how much is it strength of schedule, home/road, strength of current roster? I think Fangraphs projected standings had some of that (coin flip mode vs depth chart mode).

[PseudoSABR]

 

Yeah, it should. The Orioles just lost four games in a row, two or three of which they should have won based on statistical probability. That's a huge variance that will cause fluctuation in the chances for both teams, especially considering head to head matchups where gains or losses are absolute.

AND let's not forget that while the Twins were shelling Baltimore, Toronto was shelling the Angels.

The Twins started this series behind three teams. They ended it behind one. That is going to cause a big change in percentage chances. No model can predict aberrant behavior, which is what happened over the past few days. Toronto should not have swept the Angels at home, Minnesota should not have swept the Orioles at home. With a finite number of games remaining in the season, those statistically improbable head to head matches are going to impact postseason chances.

Again, the standings tell me all that.  I don't see how Playoff Probability really aids my understanding. 

[Brock Beauchamp]

 

Again, the standings tell me all that.  I don't see how Playoff Probability really aids my understanding. 

Does eyeballing the standings give you an approximation of how likely the Twins are to win the WC, the Rays are to win the division/WC, the Astros are to win the division, the Angels are to win the division/WC, the Orioles are to win the WC, etc etc etc?

 

These formulas are a good way to get an general idea of how the league is shaking out at a glance and given how the baseball postseason continues to become more complex, they're handy tools to have around. Add in the fact they cover things like SoS and they're more useful than eye-balling the standings. 

[Otto von Ballpark]

Again, the standings tell me all that. I don't see how Playoff Probability really aids my understanding.

Walk me through how you get any kind of useful playoff probability just through looking at the standings? If there is some kind of trick to it, I'd like to know. I look at the standings and I guess I can put teams in a couple generic buckets - "has a chance" vs "no chance" -- but I don't know exactly where that line is, based on games back, teams trailing, and games remaining. And I know there should be shades/gradients of chances within those buckets too -- how much is a 2 game deficit worth with 10 to play, maybe relative to 1 game deficit but with 2 teams to pass, etc.

 

So, I look at the standings to get the situation, then the probabilities to help quantify what that situation roughly means for all the teams involved. I get that you don't care for that second step for some reason, but that doesn't mean there is anything inherently wrong with it.

[Craig Arko]

It think we've hit one of those magical 'agree to disagree' moments.

 

Especially since the only thing that matters is the actual final result.

[big dog]

These playoff probabilities sound really cool, but I have no idea if they are any good.  I think the biggest problem with this is that they claim to be using statistical methods, and they give us broad descriptions of the variables they use, such as "current roster".  If you understand statistics at some level, as apparently quite a few of us do, you know that using statistics requires assumptions about error distributions, weighting methods, knowledge of standard errors and correlations, etc.  They are not providing any of that information, so it's impossible to evaluate how well they are actually doing the statistics they claim they are using.

 

For example...how does "current roster" affect things?  Do they look at possible pitching matchups against near-term opponents?  Project that the whole team plays at its season averages for the rest of the way out?  Career averages?  Lots of assumptions are plausible, but we don't know

a- which ones they use; or

b- how much weight they place on the topic as a whole; or

c- whether they even really include it in any meaningful way.

 

So this is somewhere between rocket science and Magic 8-Ball, but they don't tell us anything that lets us figure out where it might lie in that range.