Over the next few months, the media will be talking about presidential polls taken nationwide. Most of the polls you hear about are done nationwide and assume we Americans for our president the way the French vote for theirs (and may the Lord bless and keep Nicholas Sarkozy): that is, by popular vote.
Some of the smarter ones amongst you are toying around InTrade.com to predict the upcoming election. Well, that's nice and all, but how 'bout those of us for whom statistics matter? We're looking at FiveThirtyEight.com which breaks out poll numbers state-by-state and evaluates the pollsters themselves based on their prediction accuracy.
Anyone who knows anything about and/or remembers the 2000 race -- and that number dwindles every year amongst the growing crop of imbecile analysts -- knows that nationwide horserace polls are more like horses--t polls. The presidential race is really 50 races at once. (Note that I say 50 and not 51. Giving DC electoral votes is, after all, Democrat shut-out insurance, which is what would've been needed had they not stopped counting votes in Minnesota back in '84 when Dutch Reagan eviscerated Fritz Mondale). Calculating the probabilities based on a each state is the only way to go.
Still, there are some cool things we can learn about nationwide polls. As the conventions come around in a week and a half, the guys (or guy, I think) at FiveThirtyEight have come up with a rather simple equation to guesstimate what the bounce to the individual candidates will be in the polls:
Ln(d)*20.1 - 31.8*Ln(d+4) + 43.4
where d = number of days elapsed from the start of the convention
Well, that's easy enough, right? Obviously! A candidate can expect a 6-point bounce after the convention. Duh!
Okay, now for the rest of you, they include a chart which accounts for the fact that the Republican convention will be like four days after the Democrats hold theirs:
In other words, the Obama gets a huge up-tick for a few days, followed by a swing to McCain. However, they see McCain's bounce as being muted by the GOP's convention so closely scheduled after the Democrats rally in Denver.
Of course, this is based on previous results and doesn't account for the fact that a) Hilary Clinton is going to have her name brought up for nomination and; b) Obama has a strange, Marshall Applewhite-like control of his followers and; c) John McCain may not show up to the GOP convention if his speech doesn't fall between the Early Bird Special at Applebee's and his bedtime.
Hey, second-year analyst, come up with a strategy to play the spread between InTrade's numbers and FiveThirtyEight's numbers. Then do it because you're not getting a bonus this year.