Football Outsiders puts Seattle as favorites to win the SB

[RolandDeschain]

FO's analyses, and every similar bottom-up statistical analysis I have seen, projects current rankings onto the future with the assumption that future outcomes are independent from one another. This severely underestimates the likelihood of unexpected outcomes, because many of those unexpected outcomes contain dependent variables.

For example, the money line suggests that Denver is perhaps 80% likely to win in San Diego this week. Let's say they are also 70% likely to win vs. Kansas City the following week (in Denver). It is trivial to calculate how likely they are to win both games, so long as you assume that they are independent events. The more complicated reality is that if Denver loses in San Diego it will definitely affect their odds against KC, because both outcomes are dependent on the same unknown variables: ie. how good is Denver? Did Peyton Manning just get injured for the season?

I think NFL fans have a good intuitive understanding of this when they aren't letting simplified models sway their thinking. That is why anybody on these boards could concoct a "plausible" scenario in which Denver would lose their next five games, probably mostly revolving around key injuries, whereas rudimentary projections based on only the current status quo would have you think it was close to impossible.

Great post.

Except that everything he said is made null & void by the fact that Vegas does not make lines based on who they think will win, they make and adjust lines over time to keep betting even between the two teams so they can't lose.

Also, anybody that thinks DVOA is "simple" doesn't know what all actually goes into it.

 

[formido]

If it were true that NFL fans' intuitions were better than "simplified models" then NFL fans would be able to predict future results better than the models. But, in fact, NFL fans almost never can. When DVOA ratings were pitted against a dozen or so public Power Rankings, which are voted on by NFL fans that call themselves experts, DVOA beat every one.

 

[Kixkahn]

Lets not jump the gun, too much time still ahead of us. I know it seems grumpy to say that but injuries could still come out and bite us in the butt.

 

[Popeyejones]

Except that everything he said is made null & void by the fact that Vegas does not make lines based on who they think will win, they make and adjust lines over time to keep betting even between the two teams so they can't lose.

While it's true that Vegas ultimately ends up "crowd sourcing" who is predicted to win a game, it doesn't make what he said even remotely invalid, let alone "null & void."

He was talking about serial autocorrelation, which is true for Vegas or Football Outsiders.

Also, anybody that thinks DVOA is "simple" doesn't know what all actually goes into it.

He was making a point about what DOES NOT go into it. It's an intractable problem.


Also, worth saying that it seems like a meme has developed that AgentDib was pitting FO against Vegas. He wasn't and that wasn't his point. He was saying that FO and Vegas suffer from the same problem.

 

[RolandDeschain]

Lets not jump the gun, too much time still ahead of us. I know it seems grumpy to say that but injuries could still come out and bite us in the butt.

Yeah, we could finish the second half of the season missing 2 or 3 starting O-linemen. That would suck.

 

[Popeyejones]

A VERY SIMPLE AND INTUITIVELY OBVIOUS EXAMPLE OF THE POINT AGENTDIB WAS MAKING:

If you go to the link in the OP, the Packers' chance of winning their division decreased by 23% in one week. We all know why and we all know what happened.

While the example is an injury, the takeaway isn't "a player could get injured." The point is a more general one: what happens in week 9 has an effect on what happens in week 10, and weeks 9 and week 10 have an effect on week 11, which with week 9, 10, and 11 have an effect on week 12, and so on.

Simply for modeling reasons FO doesn't acknowledge this. In order to have a product to sell, they pretend that what happenss in week 14 is independent of what happened in Weeks 13, 12, 11, and 10 which are all also independent of each other. That's simply factually inaccurate. An injury is an obvious example, but there are tons of examples. As a 9ers fan, the 9ers offensive adjustments are the first thing that comes to mind, but I'll spare the details.

 

[RolandDeschain]

they pretend that what happenss in week 14 is independent of what happened in Weeks 13, 12, 11, and 10 which are all also independent of each other.

Uh, what? No, they don't.

 

[Popeyejones]

they pretend that what happenss in week 14 is independent of what happened in Weeks 13, 12, 11, and 10 which are all also independent of each other.

Uh, what? No, they don't.

Sorry, I wasn't clear. In the time series I meant to clarify "at right before Sunday of Week 9", as in right now. On Friday of x9 they treat their prediction for x14 as being independent of x9, x10, x11, x12, x13, when it is clearly and obviously not. They statistically treat each game in the time series as an independent event, despite the fact that there probably isn't a single person in the world who actually believes that.

 

[ImTheScientist]

Let's not forget that Football Outsiders says we're the only team in the league with a clear, noticeable home field advantage, too. 8)

We're probably getting a 10% bump in Super Bowl chances by their calculations if we get #1 seed. Good luck to the poor suckers that have to come to our house for the playoffs, lol.

Tell that to Tampa Bay! They almost skinned our ass last Sunday......at the Clink! :roll: It almost gave me nightmares!

We won.