Friday, October 3, 2008

Polling Cell Phones - Missing Obama Votes?

There is an emerging challenge for pollsters in the 2008 election - polling cell phone users.

Cell phone only users are usually those under 30, a group that has been strongly supporting Barack Obama. For their samples most pollsters use a random digit dialer that call only landlines. Therefore they are missing a lot of younger voters in their polling. To make up for this most (respectable) pollsters stratify their sample, making sure they get enough voters in each subgroup including young voters (<30).

The assumption being made with this stratified studies is that the landline user group adequately represents the whole population of cell phone users. Pew Research group recently looked into this problem and found some interesting results...

More specifically.....
"Combining polls it conducted in August and September, Pew found that of people under age 30 with only cell phones, 62 percent were Democrats and 28 percent Republicans. Among landline users the same age, that gap was narrower: 54 percent Democrats, 36 percent GOP.
Similarly, young cell users preferred Democratic presidential candidate Barack Obama over Republican nominee John McCain by 35 percentage points. For young landline users, it was a smaller 13-point Obama edge."
-Baltimore Sun

What could the reason be for this difference? Possibly, cell phone only users are those living in cities and more technologically enabled - two groups that seem to go strong for Obama. In epidemiology speak, age is an effect modifier in this election, and it appears Pew believes "cell phone only use" is a confounder. Nate Silver mentions "Urban voters are about 50 percent more likely to be cellphone-only than rural voters, for instance, and while some pollsters weight by geography, others do not. Thus, you may wind up with a biased sample." So this leads to the possibility that other polling outfits are slightly (unintentionally?) biased towards McCain.

So what can we infer from this?
1. Pollsters are using data from the group of young landline users, who are currently +13 Obama.
2. They are inadvertently excluding from the group that is +35 Obama
3. I have seen data stating about 50 percent or more of adults under 30 are cell phone only users (though less likely to vote than their counterparts).
4. Pollsters weigh younger voters under 30, as about 20 percent of the sample.

So according to Pew, pollsters are inputing the +13 Obama data, when they should be inputing +24 Obama data. Assuming that this data is worth 20 percent of the sample, that equals to a 2.2 percent difference in Obama's favor.

-Polls appear to be under polling Obama by a percentage point or two (estimated 2.2 points) based on this variable alone.

I'm not sure how I missed this. It's a good summary of what polls use cell phone data. Nate comes up with a similar value +2.8 Obama comapring two sets of data.


Cugel said...

The problem with this conclusion is that you have to treat the sub-sample of 433 out of over 6,000 as a separate poll.

The margin of error would be much higher of course if you did this.

Further, unless there are other polls on this subject you'd be extrapolating a HELL OF A LOT from one tiny sample.

I thought some other pollsters did studies and found little difference.

My own view is that there probably IS a major difference, but I certainly can't substantiate it. And NOBODY has any idea how big it will be.

And frankly, we don't know what percentage of 1st time voters will actually show up and try to vote.

Here in Colorado thousands will try and send in mail-in ballots. But, the idiots who are registering them don't know enough to tell them that state voter ID laws require a color copy of their photo ID be included with all first time voters requesting an absentee ballot.

Since they don't know this, they'll just send in the ballot, which will be thrown in the trash -- perhaps somewhere between 10,000 - 20,000 or more votes will be lost in just this fashion, in just this one state.

We will need to have a new national voter law in January to eliminate the ID requirements that are designed to block minority and first-time voting.

Dan said...

Very nice points cugel. it's a shame that there are those extra regulations like that in colorado. What is wrong with a black and white copy? I'd like to see some studies on how productive voter fraud measures (regulations) are for each state. How many voters are they turning away vs. how many incidents of fraud they are preventing. I doubt the data is available though.

I agree, we can't really say much since this is just one very small sample. You can't really make adequate conclusions on 433 people. Something I should consider for next time...

I think Nate Silver's study, which I mentioned in the update, gets at the problem a little better. He saw a D-leaning of 2 points in all the combined polls that used cell phones. However those 2 points could be due to other factors, so we still can't conclude much.