“Using a list-based methodology to quantify likely voters increases the likelihood that a phone match can be found for participants in a random sample, but, as more and more families become cell phone only households, the writing on the wall says this kind of methodology will soon be unsustainable.”
– Bill Shapard, CEO of SoonerPoll.com, on SoonerPoll.com’s July 2012 methodology change.
Why Did the Methodology Change?
Since July 2012, SoonerPoll.com’s methodology has included a combination of landline and cell phone random digit dial samples. This type of sample is known as a dual-frame sample. Prior to the methodology change, SoonerPoll used an internal database of registered voters, or a list-based methodology.
By July of 2012, only about 65 percent of likely voters in Oklahoma could be reached via a landline phone. If you compare that to the 94 percent of likely voters that could be reached via a landline phone in 2004, the year SoonerPoll.com began polling, it’s clear to see that a methodology change equated to a value added to the company.
“The world is changing, and we want to continue to make sure that our sampling remains representative of all voters likely to vote on Election Day,” Bill Shapard, CEO of SoonerPoll.com, said in a July 2012 press release.
Was the Previous Methodology Reliable?
By July 2012, it was clear that a change in methodology would soon be needed. However, it is important to note that SoonerPoll.com had great success with the list-based methodology for years before switching to the dual frame sample it uses today.
SoonerPoll was extremely accurate with the with the list-based sample, boasting an error rate smaller than all other pollsters in Oklahoma’s last four general elections before the methodology change.
To find out more about SoonerPoll.com’s reputation for accuracy, visit the “How well did we do?” page on SoonerPoll.com.
More about SoonerPoll.com’s Original Methodology
SoonerPoll conducts two types of public opinion polls: polls of likely voters and polls of Oklahoma residents.
Likey voters: how do we identify them?
Using our own internal database of registered voters – perhaps the most comprehensive and up-to-date voter database in the state – SoonerPoll identifies likely voters by using a statistical model we developed exclusively for this purpose. No other pollster has a comparable tool for determining likely voters, and the difference has been notable.
After using qualitative data on key factors, such as prior voting history, (among other factors) to construct our statistical model, we randomly pull our sample of likely voters. The list of likely voters in Oklahoma changes often. We keep the data current, and we continually monitor the growth of the state’s voting rolls. After determining which newly registered voters belong in the likely voter category, we add those individuals to our sampling frame.
Once likely Oklahoma voters are selected at random using the sampling frame described above, they will be given the opportunity to participate in the poll by phone or online. Most respondents who take our poll receive a call from our outbound call center, which you will read more about below, but our database of likely voters does contain some entries that do not have a valid phone number.
In order to give likely voters without a valid telephone number an opportunity to respond, postcards that contain both a unique identifier code for respondents who would prefer to take the survey online and an inbound phone number for respondents who want to complete the survey by phone.
SoonerPoll conducts Oklahoma resident polls among a representative sample of Oklahomans aged 18 and above. We develop samples of the Oklahoma telephone household population by using a random digit dialing (RDD) technique.
RDD selection starts with identifying all telephone exchanges that serve a desired “universe.” For example, a statewide survey would include all of the prefixes in the 405, 918 and 580 area codes.
After identifying the prefixes to be targeted, we eliminate non-working and business blocks of numbers. After valid blocks of numbers are weighted proportionally by the number of listed phone lines, we systematically select numbers to yield a probability sample of base numbers. Finally, random digits are added to the base number to create a random, ten-digit telephone number.
After the data collection process is complete, all of our public opinion polls are stratified using multiple variants that include, but are not limited to:
- political party affiliation
- predetermined geographical areas (municipality, county, etc.) within the poll’s “universe.”
The stratification process was developed by SoonerPoll using its own proprietary election turnout models. The success of our methods for determining likely voters becomes clearer with every election as we establish ourselves more firmly as Oklahoma’s leading pollster.