Wednesday, October 8, 2008

Election Polls

A short note about election polls, which I've been following somewhat religiously for the last few weeks.

Election polls differ in at least four significant ways from actual voting.

First, polls are typically of around 1,000 people or less, which means that at best, they are statistically precise to within plus or minus three percent. This means that a six point difference between 2 candidates may be nothing more than sampling error (i.e., a statistical anomaly).

Second, polls tend to be of the general population and not of likely Electoral College votes, which is how the election is counted (but see for a count of Electoral votes, according to polls). As we know from recent elections, the Electoral vote percentages frequently (and seemingly increasingly) do not correspond to popular vote percentages.

Third, polls are snapshots on how people feel on a certain day. Americans seem to be particularly fickle in their opinions recently, perhaps due to the economic turmoil, so don't trust that today's lead won't disappear tomorrow.

Finally, many polls do not remove unlikely voters (though you do see some figures concerning "likely voters"). Polls of people who do not vote are fairly useless, but pollster's haven't been very successful in predicting who will actually vote. Thus, the tendency is to include respondents who are registered and say they plan to vote, without looking at their demographics to see what they've done in the past.

For all these reasons, if you're an Obama supporter, you should be worried and if you're a McCain supporter, you should have some hope. Either way, vote!

1 comment:

feroz said...

You forgot the Bradley effect.

People who lie to the poll taker because they want to seek approval or are embarased by their real thoughts- such as they won't vote for a black person if there is a white person they can vote for. Despite that they disagree with the white candidate.
And sometimes this is unconscious (recent Stanford report).
Bradley was suppose to win the governorship in CA in the 80s, lost,almost 10 pt lead in last polls. Wilder who did win in Va in 90s for Governor in VA had something like 5 pt lead, won by 1/2. Dinkins, mayor NY same.

The consensus is that a 5-6 pt lead means a tie. 8-10 more likely. Add in polls stated margin of error of course.

But this has never been tested on a national scale. (what the percentages should be. And how the nationwide poll numbers will reflect the electoral count. How will the Bradley effect each state.)

The one thing that has been shown to be consistent, is that the younger the voter, the less one has to worry about the Bradley affect.

Hence the big, on the ground-get out the vote-on the day campaign, etc effort by Obama's people.

Also the polls under represent these young voters since they can't get them via a land line. I know of so many 20-30 years olds who have one number, their cell phone line.
Polls only call land lines, -verifies an actual residence, etc.)

But then again this is not a typical election year. One during an financial meltdown.

So many factors. Which is why truly the only real poll that counts is 11/4, and that no one during these last days should take anything for granted, or let up.

Allen Salzberg
(yes same name, no joke. And since my parents came over in 1950 I doubt related.)