Reply #23: Republicans may be less likely to agree to be interviewed than [View All]

[ProgressiveEconomist]

Democrats ("differential nonresponse bias"), and failure of exiting voters to complete interviews ("nonresponse bias") has grown dramatically over time.

Thus proportions of Republican voters in exit polls may be lower than proportions of Republican voters in final precinct vote tallies.

There is a similar kind of potential bias in pre-election telephone surveys, but it runs the other way: Since pollsters must exclude cell-phone exchanges from their "random digit dialing", "cell-phone-only" households, which tend to be younger and lean more Democratic than landline households, are excluded from pre-election polls. Virtually all pre-election telephone surveys except those done by Gallup try to correct for this bias by giving Democratic-party-affiliated responses higher weight than Republican-party-affiliated responses.

But such reweighting would be unreliable in an exit poll to CHECK election results; party weights used for exit polls generally come from precinct tallies. Such weights may help pollsters explain who voted Republican and who voted Democratic, but can't be used in auditing elections.

According to http://www.mysterypollster.com/main/2004/12/what_about_thos.html , the "nonresponse rate" in 1992 national exit polls was 40 percentage points. Eight years later, in Dubya's initial "victory", it was 23% higher, at 49 percentage points. Higher nonresponse bias over time may also mean higher differential nonresponse bias over time.

See also Febble's DU posts today on the "nonresponse" and "differential nonresponse" issues, at http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=364&topic_id=2654934&mesg_id=2654934 .

Also, increasing percentages of voters are using absentee ballots or voting before Election Day in the 15 states that allow early voting. Differential (between Rs and Ds) trends in these modes of casting ballots could be compounding any differential Election Day nonresponse bias against inclusion of Republican voters in exit polls.