Political Blog

Find posts by keyword
Find posts by date

Print
Email
|

Under fire in VA, Survey USA offers two polls

Under fire in VA, Survey USA offers two polls

by Joe Arnold

WHAS11.com

Posted on October 14, 2010 at 2:29 AM

Kentucky Democratic candidates Jack Conway and John Yarmuth have both complained that Survey USA, the pollster behind the WHAS11/Courier-Journal Bluegrass Poll, has been oversampling Republicans in polls of their respective contests.

In our polls, Survey USA filters the results only to determine likely voters but does not apply any projections of voter turnout.

Yarmuth told me that his concerns about a WHAS11/Courier-Journal Bluegrass Poll that showed a virtual dead heat with GOP challenger Todd Lally mirrored concerns by incumbent Democrat Tom Perriello (D-VA) who is trailing in Survey USA polls. 

SurveyUSA has now conducted two polls applying different methodology in the Virginia race.

Once in a while, SurveyUSA pre-election polls become a lighting rod, and in 2010 in Virginia's 5th Congressional District this is the case. 3 SurveyUSA polls released in July, August, and September, all using identical methodology, had incumbent Democrat Tom Perriello trailing Republican challenger Robert Hurt by more than 20 points. At the same time, polls by Democratic interests had the Democrat down by 1 or 2 points, and a poll by a Republican interest had the Democrat down by 6 points.

For today's 10/12/10 release, SurveyUSA conducted parallel research, two separate but otherwise identical polls, interviewing 613 likely voters using RDD sample and a separate 628 voters using RBS sample. In the RBS poll, which is apples-to-apples comparable with SurveyUSA's 3 previous 2010 polls in VA-05, Republican Hurt today leads by 17 points (down from 23 on 09/28/10). In the RDD poll, Republican Hurt today leads by 11 points, 52% to 41%. It is impossible to know (at this hour, before votes are counted) which of these two polls is a better snapshot; both sampling methods can be defended.

The largest difference between the two polls, which accounts for almost all of the variability, is among black voters: SurveyUSA's RDD sample is 20% black, and Perriello leads among those blacks 7:1. The RBS sample is 12% black, and Perriello leads among those blacks 7:2. Whites vote approximately 2:1 Republican in both polls, and each black vote is a vote that makes the contest closer.

Common to the two polls: Independent voters break 2:1 Republican; the Republican leads in all age groups; the Republican leads among less educated and more educated voters; the Republican leads among less affluent and more affluent voters; the Republican leads materially among men and less materially among women. SurveyUSA includes below the results of the RDD sample, as the more defensible data set. But: SurveyUSA has collected no data, and cannot weight any of the data it has collected, using any sampling method, to mirror the carbon-copy polls released by Democratic pollsters Penn-Schoen, Global Strategy Group, and Benenson, and the lock-step data from the SEIU labor union, which is working to re-elect Perriello.

 Historical Context, Definitions, Filtering: In 2008, polling in Virginia's 5th District with RDD (Random Digit Dial) sample, SurveyUSA's final pre-election poll, compared to the actual election result, was 3 points too Republican, within the margin of sampling error, and included this caveat in the analysis: "If black turnout increases from the 22% SurveyUSA shows here to 25%, the Democrat may take the seat." The Democrat did take the seat, by 745 votes. In 2010, SurveyUSA switched in this and other congressional districts from RDD sample to RBS (Registration-Based Sample, aka: Voter-List sample). RBS sample allows the pollster to know it is reaching a house with at least one registered voter. There are pros and cons to each sampling method. "Sampling" is the way the pollster draws the phone numbers that will be called for a telephone survey. RDD starts with the assumption that every possible residential phone in the geography can be dialed; RBS starts with the assumption that only those registered with the state, who have a phone number that can be matched to the address, can be dialed. In 2006, the Republican won VA-05 with 59% of the vote. In 2004, the Republican won VA-05 with 64% of the vote. SurveyUSA's RDD poll, reported below, interviewed 850 adults from VA-05, of whom 780 were registered to vote, and of whom, SurveyUSA determined that 613 had already cast a ballot, or were likely to do so on or before election day. The results from today's RDD poll are not backwards comparable to SurveyUSA's previous releases, and therefore do not contain interactive tracking graphs. It is incorrect to infer that Perriello has gained 12 points since SurveyUSA's last release.

 

Print
Email
|