A new Washington Post poll of likely voters in the Alabama U.S. Senate race that shows liberal Democrat Doug Jones with a three-point lead over conservative Republican Roy Moore, 50 percent to 47 percent, was based on a methodology that includes an oversampling of Democrat voters.
The sample of 739 likely voters upon which the poll was based included 38 percent who self-identified as Republicans, 31 percent who identified as Democrats, and 27 percent who identified as independents. (See question 14 in the poll methodology here.)
This sample, which includes only seven percent more Republicans than Democrats (R +7), is not reflective of how Alabamians have voted in recent elections.
In 2016, for instance, President Trump defeated Hillary Clinton in Alabama by 28 points, 63 percent to 35 percent.
The Fox News Poll released on November 16 showing Jones up by eight points contained a similarly disproportionate Democrat sampling, as Breitbart News reported:
The Fox News Poll sample of 649 likely voters included 48 percent Republicans, 42 percent Democrats, and ten percent independents, which appears to be an unusual oversampling of Democrats and undersampling of Republicans.
Another unusual aspect of the Fox News Poll released on Thursday that shows Democrat Jones with an eight point lead over Moore is that it appears to contain a sample of Alabama likely voters in which 51 percent voted for Donald Trump and 48 percent voted for Hillary Clinton in the 2016 Presidential election.
The new Washington Post poll is the only one of the four polls included in the latest Real Clear Politics (RCP) Average of Polls (which shows Moore with a 2.5 point lead over Jones) that shows Jones in the lead.
According to its own methodological description, the Washington Post poll demonstrated how selectively respondents were moved into the likely voter category: “This Washington Post-Schar School poll was conducted by telephone November 27-30, 2017, among a random sample of 1,304 adults in the state of Alabama, including landline and cell phone respondents. The error margin is 3.5 points among the sample of 1,110 registered voters and 4.5 points among the sample of 739 likely voters. Sampling, dating collection, and tabulation were conducted by Abt Associates of Cambridge, Mass.”