Thursday, August 16, 2012

Digital Politics: The Princeton Election Consortium

Today we'll take a look at a electoral analysis site that actually makes predictions: The Princeton Election Consortium! I wasn't sure what made it a "Consortium" so I looked it up: it's two or more people pooling their resources for a common goal. As the PEC contains two people it counts! I'm thinking that after going over fifty+ possible names it was suddenly 11:30 PM and Professor Sam Wang and Andrew Furgeson were out of beer and went "Well, the initials PEC are kind of nice."

But whatever the origin of the name, they have quite a site. Here is the key result:
2012 EV history with November prediction
If Romney Were Ahead Would The Obama Line Be Down At The Bottom? Or Would It Be The Romney Line?

This shows, to "95% accuracy" where the election will line up in November--the far right is the finish line. It's somewhere between 280-something and 350. I think. Maybe the yellow is 95% accuracy which would mean it could be anything from a slim Obama-loss to a total Obama-blow-out. They report: you decide.

The Algorithm
The algorithm goes something like this (if I've understood it correctly) Take All The Polls (state polls, well the three most recent anyway) and then take the median of them ... plus ... um, some other stuff, and then multiply EV value of state by %-chance of winning it. The "other stuff" looks like taking the most-current of recent polls and using their numbers get a distribution of chances of winning. Take the median of that and multiply by EV count--do this for all states and you get a number (with a chance of error, I presume).

Apparently you can do stuff to throw out outliers or take distributions other than the "standard bell curve." In any event, he does not add in economic indicators (believing they are already "priced in" to the poll) betting odds, or anything else. This is polls + math weighted by EV.

How Accurate Is It?
Like all of these? Very accurate: 2008 prediction was Obama 365, McCain 173. Real outcome? Obama 364, McCain 174. They take a victory lap over and They don't list  Election Projection but I can do the math and they were better than them too (he missed two states and therefore a BUNCH of electoral votes).

Am I impressed? Well, yes ... and no. I am impressed: I am a believer in math and numbers and these are very good. I'm not really an expert on his method (my eyes glazed over at: "DETAILED DESCRIPTION OF THE INVENTION-PREFERRED EMBODIMENT") but you can't argue with results (their results for 2004 were pretty good too--although there is some kind of story around monkeying with results that you can read for yourself). Clearly these guys know what they are talking about math wise.

On the other hand, keep in mind that by the end of the 2008 cycle everyone was pretty much saying / showing the same things. It's one thing to say "Obama seems to have a good 314 EV right now." It's another thing to say "that equates to an 89% chance of victory in November based on ..." [how other candidates with a 314 EV total in mid August did? Is that right?].

In other words, while I think his math is great, I'm not sure what goes into the future projection analysis other than just extending the line straight to November.

So we know he was one EV point off by Judgment Day (Nov 2008). How was his estimation in August of 2008? It's not that easy to navigate the site for a closer look but here's the picture I found:
Electoral votes for Obama since April 1
Okay ... Mid August ... Looks Like Obama Was Around 300 EV ... He's Doing Better NOW!?
Well, there you go ... (he was around 65 EV off. We don't get a red-line of likeliness)

The Site Itself
The site itself is functional but could use some prettying up. The charts, again, are functional, but not especially attractive. On the other hand Prof Wang is responsive and accessible: send an email, you'll probably get a response (if a short one) and you can message-board chat with the Professor too.  

Rather than linking to "today's news," instead he writes fairly regular blog posts which illuminate his thinking on the current state of the election and are, usually, informative reads.

I am not a mathematician but I would be inclined to refer to this analysis as sort of ... hmm ... what's a non-mathematical term I can use ... 'De-Bullshitting' the state polls. Clearly we see that most of these sites don't care about national polls: wanting to know "who will win" explains that--EV math makes individual state polls far more important (far more of a known quality). There are clearly different methods of de-bullshitting the current state view: this appears to be a very, very good one.

Is Obama really like 90% likely to win? That seems ... high. I'd like to know what determines the range of his red prediction bar and I haven't been able to discern it from the site itself (the FAQ doesn't address it). If it's just the middle of where the EV count is right now then it seems that in 2008 the red-bar of likeliness would've fallen well below the "win" area giving Obama like a 50% chance of losing on August 16th. Is that right?

Let's see what their entry for the 2012 election shows in 2013 (probably: "And I would have gotten away with it too, if not for that damn Mayan apocalypse!").

No comments:

Post a Comment