Ring of Conservative Sites Ring of Conservative Sites

[ Prev | Skip Prev | Prev 5 | List |
Rand | Next 5 | Skip Next | Next ]

Monday, November 06, 2006

Pay no attention to that poll behind the curtain!

Mike Taylor provides this election coverage:

This morning's news about the "generic poll" may give heart to those of us who favor conservative candidates. Today's headline from USA Today is:

You know that "generic" question that news analysts constantly refer to, it goes something like this: "If the elections for Congress were being held today, which party's candidate would you vote for in your Congressional district?"

Smart voters, such as yourself, realize that this poll question is meaningless and generally unpredictive of actual election results. (See, I respect your intelligence, whether you're in the military or not, unlike some Massachusetts Democrats I could name!)

Why is the predictive value of this question so poor and why report it at all? I don't know the answer to the second part of the question. But there is one important reason why this generic question is unpredictive: sample stratification.

There are 435 separate House races in the United States. Each one of those districts is holding a unique and mutually exclusive House election. If you live in Connecticut's 4th District, as I do, you can't vote in Missouri's 4th District nor in California's 28th. I can only vote it one house race, I can not vote in all of them. (Insert Democratic voting irregularities joke here..)

That means that each race is highly, exclusively STRATIFIED. Let's look at a simple example of stratification:

District A: 9 of 10 voters will vote for the Democrat.

District B: 6 of 10 voters will vote Republican.

District C: 6 of 10 voters will vote Republican.

TOTAL SEATS WON: 2 Republican, 1 Democrat
VOTE COUNT: 30 Total: 17 Democrat, 13 Republican

That means that there will be a majority of Republicans elected in these three races... but generically, Democrats "win" by capturing 17 of the total 30 votes (57% of total votes cast). How meaningful is that "win" to Democrats? I guess it might make their egos feel a littel less battered come Wednesday.

All this is simple statistics, yes? Because you're a very smart voter I can anticipate your follow-up question: Why don't pollsters know this and adjust their polling techniques?

The simple answer is: to correctly collect a stratified sample among 435 Districts would cost time and money. A LOT of time and money.

By now, you've all seem sample summary statistics and have probably noticed that most polls are conducted among 400 respondents or so. (I say again, you noticed because you're a smart person! Good looking, too.) A sample of 400 voters will yield a very low "expected error rate".

Generally, in the survey business a sample of 125 is considered to be fairly predictive, stable... and cheap. A sample of 400 approaches the point of diminishing returns, collecting more than 400 responses does not significantly reduce the expected error rate, but it does cost money to keep collecting responses!

If you read the sample statistics for this generic poll (Gallup) you will find:
Poll results are based on telephone interviews with 1,516 National Adults, aged 18+, conducted November 2-5, 2006.
A sample of 1,500 adults seems very large in comparison to 125... and even robust when compared to the average poll sample of 400. But since those 1,500 respondents are divided among approximately 435 House races, that makes each individual District sample about 3.5 responses per District! That is, if the sample is sufficiently stratified.
How does that sample size of 1,500 look now?
If you wanted to conduct a relatively stable generic poll taking into account all 435 House Districts you would need 125 responses X 435 Districts or 54,375 responses... that's Fifty-Four Thousand responses.
You can see how the money needed to conduct a good generic poll in an off-year election becomes an important factor.
So, although the generic poll trend looks better for Republicans this morning, it is all going to come down to individual races.

My prediction: I am with Hugh Hewitt (http://hughhewitt.townhall.com/) and I predict that Republicans will hold both the House and Senate come Wednesday morning.
I can't wait for the Main Stream Media meltdown that will follow.

-- Michael Taylor

Excellent job, Mike. This, loyal readers, is from someone who actually knows about polls from a professional point of view.

I might add:

We will find out tomorrow from what I like to call an "actual election" with voters cast by "actual Republicans'. That may skew some of the results a little.


Post a Comment

<< Home

Ring of Conservative Sites Ring of Conservative Sites

[ Prev | Skip Prev | Prev 5 | List |
Rand | Next 5 | Skip Next | Next ]