Cell phones, GOP bias and Oklahoma

SoonerPoll, like every other pollster in the nation, is facing a new challenge due to the rise of cell phone only households. The advent of cellular phones has made political polling more challenging, but not impossible.

Land line telephones reached their highest penetration into homes in 1997 when 98 percent of homes had a land line phone. Since then, the number of households with a land line phone has dropped to a level not seen since 1963. This is a worrying revelation for many pollsters, after all in 1963 random digit telephone dialing was not considered viable for polling since it excluded such a large percentage of the population.

Consider this: if a pollster in Oklahoma were to pull a random sample of addresses throughout the state and then run a phone match, about 65% would match with a phone number. This means that 35% would not be included in an outbound data collection of live interviewers.

At SoonerPoll, we use a list-based methodology (list of registered voters) and then quantify the likeliness of voting, which calculates a "score" for every registered voter. If we were to pull a random sample of likely voters throughout the state and then run a phone match, about 92% would match with a phone number.

Why are these two phone matches so different? First of all, one must keep in mind that only 57% of the general public is registered to vote and only about 50% of those registered will vote on election day. Likely voters are typically stakeholders within the community, less likely to move or change their home phone number, and more likely to port to a cell phone only than the general public.

A recent study conducted by Pew Research Center shows that as the number of Americans who rely solely or mostly on a cell phone has grown, the potential bias in landline only surveys in favor of Republican candidates has grown as well. In fact, the study finds that in the 2008 election landline only surveys were on average bias in favor of the Republican candidate 2.4 points, in the 2010 election that number more than doubled to a 5.1 point bias.

Interestingly the SoonerPoll results released just days before the election did not have a Republican bias, in fact in every race SoonerPoll had the Democrats doing better than they actually did on election day -- that is to say every race in the SoonerPoll had a Democrat bias. Keep in mind that polls rarely call out the exact result on election day and the result will either have a bias toward the Republican or Democratic candidate (in those races with a Republican and Democrat only).

The table below presents the Republican's spread in the statewide ballots and SoonerPoll's spread released in just less than a week prior to the election. For example, Mary Fallin won by 20.9 points, we had Fallin winning by 18 points in our last poll, and therefore favored Jari Askins, the Democrat, by 2.9 points.

Spread-WinSoonerPollBias

Governor 20.9%R, 18, 2.9%D
Lt. Governor 31.52%R, 17, 14.52%D
State Auditor 11.88%R, 6, 5.88%D
Atty. General 30.22%R, 20, 10.22%D
State Treasurer 33.14%R, 10, 23.14%D
School Super 18.2R%, 9, 9.2%D
Labor Commis 28.34%R, 13, 15.34%D

Missing from the list above is the Insurance Commissioner race which the Republican won by 8.96 points and SoonerPoll, in its final release of polling results, had the Democrat with a 2 point lead which is still within the poll's margin of error. This 8.96 point spread was the smallest win of any Republican on the statewide ballot and SoonerPoll had the Republican candidate winning in every other pre-election poll released.

In order to understand why the SoonerPoll results did not have the same bias as the rest of the country, you must first know why the rest of the country has had such trouble with Republican bias in the polls.

The Pew Research report suggests the Republican bias is due to the fact that cell phone only respondents are younger and more likely to be black or Hispanic, and therefore less conservative and more Democratic in their vote preference than landline respondents. Though this may be true on a national scope, much of this conventional wisdom does not hold true to Oklahoma.

We have mentioned the youth vote in a previous post, but it is of value to mention it again, in brief, here. State election board data has consistently shown that not only is the youth vote too small to make a large impact on the vote in Oklahoma, but the youth in Oklahoma are slightly more Republican than Democrat in registration and turnout.

Oklahoma is also a state with very few black or Hispanic voters (in comparison to the white vote at roughly 85%), and has only about 10% of voters who identify themselves as liberal.

Though the 8 percent coverage issue SoonerPoll faces with its list based sampling is better than the 35 percent coverage issue that many pollsters are faced with, it is still a coverage bias that needs to be addressed. We do address this coverage bias by giving potential poll respondents with no landline phone the opportunity to participate via the internet or an in-bound call, although the response rates are much lower than traditional out-bound.

In the end, the problems of cell phones and the national pollsters' Republican bias hasn't hit Oklahoma as hard. At SoonerPoll, we are committed in adapting mixed modes of data collection in order to produce the best polling results possible.