I think it's finally time to revamp polling during election cycles. The current system is bad for two reasons. First and foremost, they influence the outcome of what they're measuring. Media outlets use constant opportunity to call it as if it was a horse race. This gives the perception of momentum and, in turn, plays a role in the choice of some portion of those voting (and being polled). Second: they don't actually measure what needs to be measured. As i recall, it's the delegate count that determines the winner - not the popular vote, and the latter is what the polls are measuring.
On a side note, I was appalled at people who continued to point out that Al Gore won the popular vote in 2000. Considering this was the 57th presidential election, i would hope that we'd all remember that has no bearing on who's elected. We might as well have considered who had more myspace friends for winner in 2004 and then who has more people following their twitter streams this year - all would be just as relevant.
Back to the point - you'll notice I said we should revamp polling. My first point should be enough to eliminate it. Since I doubt that will happen, i think the media should all chip in and do it right. Instead of one poll to judge the popular vote, they should run fifty state-based polls. Each of those could then be placed into some cool delegate map to show the outcome. A few outlets, the LA times for example, had great maps during the primaries. They could even throw in a slider to let us play with the margin of error (come on, can't they just borrow the code from Karl Rove's I/T guy?)
I realize that there are two potential areas of contention in my theory: work and accuracy
Work: to run a single poll with +/- 3 percentage points you need to survey 1111 people. To run 50 such polls would be quite a few phone calls - especially if there doing two a week. reducing the margin of error to +/- 4 percentage points would only require 625 people per poll. This is still 31,250 phone calls, but that's quite a bit less work.
Accuracy: It's been some time since my statistics classes, so I'm not exactly sure how to roll up the margin-of-error in a case like this. Simple combination of the data is pretty well documented, even weighting the margin-of-error for different size samples. This part needs a better mathematical mind than mine.
So, I'm appealing to the pollsters to update their methods, the media to refuse to publish polls that are based on total popular vote. If both of those fail, I'm calling on anyone who gets a phone call from a pollster to tell then they're voting for Rush Limbaugh. That should show the validity of the poll.
Sincerely,
Checking my caller ID before I pick up.
Comments