About RRG’s Processes

Outlined below are the processes used by Remington Research Group for conducting political surveys. This memo – available in PDF format here – will address the following processes:

– Instrument Creation
– Universe Data Acquisition
– Setting Turnout Demographics for Weighting
– IVR Deployment
– Results & Weighting

POLLING INSTRUMENT CREATION

All surveys from Remington Research Group include a screening question at the beginning to ensure we are only speaking with likely voters. Self-identified demographic questions are also included to ensure we are weighting the interviews correctly post-interview.

Remington does not utilize an intensity ballot (Definitely Candidate A, Probably Candidate A, etc.) in its instruments. Ballots are read stating candidate name and political party affiliation of each candidate. Respondents are then given the choice of each candidate read in addition to an undecided option. We do not branch undecided voters to a “leaners” question where they would indicate which candidate they lean towards. Remington utilizes a rotating ballot to eliminate any bias towards the first choice.

A sample Remington Research Group Instrument is attached to the PDF version of this memo as Addendum A (click to view or download).

NO STRATIFICATION/QUOTAS

Remington does not stratify samples. We also do not utilize a quota system when collecting interviews.

Many pollsters only stratify based on age, race, and party registration. Due to their small sample sizes, they may not have a representative sample based on differences in response rates within demographic subgroups by geography. There are likely differences between individuals in the same demographic group depending whether they live in urban, suburban, or rural areas.

RANDOM SAMPLES

For a statewide survey, the goal is to collect a minimum of 1,600 IVR landline interviews and 400 live cell phone interviews. The size of the sample enables us to install our weights, leaving us with a significant effective sample size. Additionally, not using stratification or quota systems allows us to speak with voters that represent every geographic area of the state. Quota systems limit the reach of a survey.

Landline IVR Random Sample: Criteria for voters included is that they possess some type of voter history in the past eight years or be a newly registered voter.

Remington acknowledges that IVR to landlines is not without fault. We have found that IVRs miss certain subsets of voters that are necessary to compile a full sample that is reflective of the electorate. Remington uses live calls to cell phones to capture these subsets of voters that an IVR is unable to. For this reason, Remington pulls random oversamples within chosen subsets for the cell phone component (i.e. African Americans, Hispanics, young voters).

The table below shows the landline and cell samples from the latest polls in Florida, North Carolina, Ohio, and Pennsylvania conducted by pollsters with a FiveThirtyEight rating of A- or higher. As you can see, Remington surveys include live cell sample sizes that are similar to what other pollsters collect but also includes significantly larger landline samples.

The noticeable difference between Remington and other pollsters is the large landline sample. The larger landline sample reduces risk of missing subsets of voters, specifically geographically (urban vs suburban vs rural).

IVR DEPLOYMENT

Remington Research Group conducts automated telephone surveys using Interactive Voice Response (IVR). This technology radically reduces the expenses related to research and offers much faster survey completions.

In addition to cost and time advantages, IVR technology has enabled us to poll more accurately. Interviewer bias has been reduced to zero by eliminating the inflection and tone of a live caller. Every survey respondent hears the exact same question read in the exact same way. Social desirability bias are also eliminated using IVR mode.

RESULTS & WEIGHTING

The most important aspect when running an automated survey is to weight the responses to mirror the actual turnout demographics for the electorate. Setting the weighting ratio is what separates accurate numbers from inaccurate numbers.

Remington uses hard data from the voter file to appropriate weighting ratios. This means we weight per the participation of actual voters, not per the raw full voter file or Census data.

Remington utilizes traditional weighting methods. We weight per:

– Geography (Media Market and Congressional District)
– Party Affiliation
– Gender
– Age
– Race

Weighting raw data will always reduce the effective sample size of the survey. Remington’s original samples are large enough to leave an effective sample that is still significant.