www.electroral-vote.com and Andrew Tanenbaum

Yeah I have loved this site over the past couple of months ... I donated to him as well. It's got a ton of great stats to analyze. It's a stats-dork wonderland.

I loved Andrew's Operating Systems book in college, but unless I'm misunderstanding his methodology, his use of statistics here is poor.

I'm *horrible* with statistics, but one thing I know is that if you do a poll that has a +/- 3% margin of error, and the score is, say, 49% Kerry, 48% Bush, then you can't call that poll "barely Bush." The poll is, in reality, a dead heat, and NO realistic prediction or projection can be made from it, because the spread is within the MoE.

I understand that what he's trying to do is provide a lot of different pictures of how the vote looks using different algorithms, different polls, etc., and a map with 10 or 12 states that are "too close to call" doesn't look very interesting... But the way he's splitting hairs with these poll results, he might as well investigate further into a 49% - 49% poll and see whether it's really 49.3% - 49.1%, just so he can color one of the states "barely Bush" or "barely Kerry."

I guess the biggest problem I have is that anyone who casually glances at this site might be misled into thinking there's some sophisticated statistical jiggery-pokery that's allowing Tanenbaum to make projections that are somehow more accurate, so he can color in more white states either reddish or blueish... And that's just not the case.

Or, to put it another way, statistics don't lie, but disgruntled Computer Science professors do.

While I agree statistics can be used to mislead, I think he's very open about his intepretations of the data.

In his FAQ:
"What do the terms "strong," "weak," and "barely" mean?
By definition, "strong" means support of 10% or more; "weak" means 5% to 9%, and "barely" means less than 5%. The states marked barely are statistical ties."

Also, I would read:
http://www.electoral-vote.com/info/polling.html
Where he says:
"In the first above example, with a 3% MoE, the 95% confidence interval for Kerry is 49% to 55% and for Bush 45% to 51%. Since these overlap, we cannot be 95% certain that Kerry is really ahead, so this is called a statistical tie. Nevertheless, the probability that Kerry is ahead is greater than the probability that Bush is ahead, only we cannot be very sure of the conclusion. When the ranges of the candidates do not overlap (i.e., the difference between them is at least twice the margin of error), then we can be 95% certain the leader is really ahead."

I think anyone looking at ANY statistics needs to take it upon him/herself to understand how the interpreter sliced and diced it to give you the view you're seeing, because polling is inherently "fuzzy math". Personally, that he backs up all the stats he puts up there with explanations of how he arrived at the final numbers along with the actual source data feeds is far more intellectually honest than almost any other polling/estimations that I've seen on the web.

I didn't say his intentions were to mislead, and, unlike the Slashdot crowd that's mulling this over, I don't think he's got any particular candidate he's trying to twist the statistics in favor of. But I just have a problem with distilling very tenuous, fuzzy statistics down to a little website image that says "Kerry: 298, Bush: 231." It's about six too many levels of abstraction and extrapolation, and unless you drill down to all of his qualifiers and disclaimers about the "fuzzy" data, you might think he's actually got better data than someone else. All he's really doing is over-simplifying and making judgement calls when most responsible statisticians concede that no real judgement can be made. It's surely a harmless oversimplification, but it still irks me. Not that I'd have any problem with a 298-231 Kerry victory.

Tangentially, the electoral college blows.

Hey, it's all a moot point anyway - the only important statistic came in yesterday...

Redskins lost.

Another good site for "meta" analysis of electoral polls is done by a biophysics professor at Princeton. No mystery about who the author is, plus much more rigorous statistics, including MatLab code if you want to run your own numbers. On the other hand, Tanenbaum's site is much prettier and better organized. If I had to hazard a guess, Tanenbaum has pretty much given up on his day job to keep that web site going, while the Princeton guy is still trying to make his day job work for him.



I feel sympathy for the Princeton guy. My phone has been ringing like crazy with press inquiries about tomorrow's election. I gave interviews today for BBC Scotland (but the interviewer had a British accent) and for the local ABC news channel. Tomorrow will be the big blitz. My university's PR guy has set up an entire camera studio on campus with a terrestrial fiber network feed so that CNN or whoever can punch me up directly. I've also been posting on a new E-Voting Experts Blog and written up a detailed guide for election observers. Sleep? What's that?

Great work, Dan. Keep it up, eh!

Awesome. Now give me the backdoor codes to the Diebold machines so I can break into them and inject 120,000,000 votes for Homer J. Simpson!

Checks and balances...now *there's* a concept...!

Checks and balances...now *there's* a concept...!

You know you are getting a bit cynical (and are from New Jersey) when you read that line and immediately respond - "Those that cash the checks and have the biggest balances..."