FiveThirtyEight.com: Regressions for Politics

Number two on my list (previous post) of technological innovations to political discourse and campaigning this year is regression modeling of polls. The first one that I saw doing something like this, though far more simplistically is realclearpolitics.com (RCP). Basically what they did is took an average of all the polls, without assigning any kind of weighting. This was good to give us a nice and easy quick look at where the polls stood for the election, and washing out, for the most part, the effects of left- or right-lean of polls.

I would check RCP during the primary season occasionally, just to get the sense of where things stood, but they often didn't get the right trends. Around the middle of the primary season, I came across the website fivethirtyeight.com.  At first, I was very apprehensive about the model. My initial reaction was that there would be a propagation of errors that basically rendered the results useless. But the genius of the regression model, wasn't just the polling but the way these polls were weighted, both with lean and with recency.

If you'd like to check out the FiveThirtyEight FAQ, please do so; I will not repeat work that is already excellently explained even if you're not mathematically inclined. I think many of us who were watching this campaign intently, and liked the numbers aspect, really came on board around the time of the North Carolina and Indiana primaries. It was at that point that I put 538 on my bookmarks and checked it daily. You will also notice that I now use the 538 widget on my right rail. 

A note about FiveThirtyEight.com is that the statistician in question is Nate Silver, of the baseball stats site, BaseballProspectus.com. Most of you know that I also have a baseball-related blog, Sabermetrati.blogspot.com, and am a big fan of Nate Silver's work on BP, which I reference regularly. I did not realize that it was the same Nate Silver until around early June, and it definitely lended more credence to his fine work on FiveThirtyEight.

My favorite part of 538 is the win percentage. I have been following that number since Obama officially won the nomination. It now stands at 94% chance of Obama winning the election, with a peak about a week ago of 98%.

Tuesday will prove the accuracy of Nate Silver's model; it should be interesting.

A Rocky Mountain High and a Boston Tea Party

Click Image to Enlarge

I guess $200,000,000 plus payrolls just don't go as far as they used to anymore.

The probable winners in the LCS's will be the Rockies and the Red Sox. Of course, that's not to say that they will win. As I often find myself explaining to people that the statistics show the most probable world, not the way the world will be. That's also why I don't take this information and bet on it. It's entirely possible that the Cleveland Indians can stun Boston (37% chance is pretty high). But it is unlikely. It should be a good series, though, that's for sure.

And in the NL, the ratings bonanza that is two teams formed in the Nineties in the baseball mecca of this country, the D-backs certainly could overcome an incredible home field advantage that is Rockies Stadium, but it's unlikely.

On a side note, as to method: I'm still on board with log5, and the same with pythagorean, but I'm thinking the better way to slice and dice the runs is perhaps not just home and away runs/runs against, but to look at it only against teams that are better than .500. Too often in a short series, which is a crapshoot anyway, the real way to examine teams is to look at how well they performed against their real competition, rather than against the dregs of their own division. If I had the stats to do that, I would use that method instead. For instance, the Cubs totally fattened their pythagorean record on their bad division, whereas the Arizona record didn't get properly weighted for playing against basically a whole division of playoff contenders (minus San Francisco--and even they weren't pushovers!). We'll see what we can do next year...

(Editor's Note: I changed the image for two reasons from the original post: 1. there was an error in the formula for the home series in both LCSs; 2. I added the current, as of Oct. 13, changes in the odds based on the games played.)

Maybe for the Next Four Weeks I Call This Illumi-Baseballrati...

This will be a cross-post with Big Shoulders Sports.

Two years ago on Big Shoulders Sports, I blogged about the postseason chances of the White Sox. Re-reading prior to this post surprises me that I was so accurate (I predicted a Sox victory)--perhaps I need to call a bookie!

In any case, my method is to use a combination of Bill James' Pythagorean Theorem and his Log5 Theorem for one-game win probability. My little wrinkle is using home/away numbers instead of totals, and then to total up all the probabilities adding up to 3 wins (win-win-win, win-loss-win-win, win-loss-loss-win-win, etc.). Because probabilities are additive, the chances of any scenario can just be added up (where the total equals 100%).

So what do the numbers show?


(Click to enlarge image)

The Yankees hold an advantage over Cleveland, Boston looks very tough to beat by the Angels in the AL. In the NL, the Cubs should win the first playoff series over the Diamondbacks, and the Rockies have a slight edge over the Phillies.

Now, I know what you're saying: how can your numbers show the Cubs having such an advantage over the team with the best record in the league? It's simple, the D-backs have a below-.500 pythagorean record, which is almost unheard of in the playoffs. The Cubs have a decent home-field advantage and are far better on the road--this will translate into an NLDS victory for the North Siders.

As the games are played, I will update this table and post it here (and on Big Shoulders), since the odds will change with each victory and loss.