Thursday, September 18, 2008

Campaign 2008: The Year That Polls Died

A number of polls were released today and by "a number" I mean far, far too many. The Hedgehog Report alone reported on 37 different poll results from 13 different sources--that's just for Thursday. The problem is not really the number of results--it's the number of sources. Everyone and anyone seem to be conducting polls. With little history of these agencies it is difficult to know whether to even start believing the numbers or not.

When faced with a poll from an unknown source about the only thing you can do it look at raw numbers or compare it with polls from other agencies. That is a dangerous tact to take as even a reputable agency can have an anomalous results and even a shady agency can get lucky.

My basic reaction is to just stop looking in the number in detail after I start seeing silliness. For example, today we have:
Strategic Vision for NJ: Obama 47%, McCain 44%.
Insider Advantage for CO: Obama 51%, McCain 41%.
Now a friend has warned me against comparing state to state. He argues that particulars of the candidates could cause a traditional red state to flip while a traditional blue state to show little movement, for example. His advice is good up to a point. But those two results just can't be correct. Colorado was red by about 4% in 2004. New Jersey was blue by double digits. McCain being behind 10% in Colorado would indicate a landslide for Obama--he should be ahead in NJ by twenty plus in that case.

The real problem is that the poll aggregation sites are not filtering many polls. RealClearPolitics now has Obama as a notable favorite in Colorado and will for some time--until the +10% number falls off. It makes the aggregation sites meaningless.

My only advice to poll junkies at this point is just to wait it out. Eventually people will realize that a poll from an brand new source that shows a results that is +/-10% from the results everyone is showing for the same time period is not to be trusted. Until then, your best bet is to ignore those numbers that looks suspicious to you.

No comments: