More today on the debate over why the Rasmussen automated
survey is showing
a closer Democratic race than other pollsters. Today Rasmussen released new
results showing dead heat between Hillary Clinton (32%) and Barack Obama
(32%). Although the current results are not significantly different (in a
statistical sense) than last week, Rasmussen tell us:
Obama has been steadily gaining
ground during April. Last
week, Clinton
had a two-point lead. Two
weeks ago, it was Clinton
by five. The week
before that,
the former First Lady was up by seven. Our last
release in March found Clinton enjoying a double digit lead.
Over at MyDD (a liberal site whose contributors are not
particularly fond of Sen. Clinton), Chris Bowers sees this new result as
further evidence that the other national polls are inflating Clinton's lead. He does the math and
concludes that the Democrats sampled for this most recent Rasmussen survey
represents 19.5% of the voting age population, compared to between 38% and 50%
represented by the Democratic subsamples of other recent national surveys.
Bowers concludes by asking a very good question:
Is the difference between
Rasmussen's national Democratic primary preference numbers, and the numbers of
other polls, the result of the different universes the polls are sampling? To
put it another way, does Clinton
perform better in non-Rasmussen polls because those polls include a far greater
percentage of "unlikely" primary voters?
The answer is elusive. I looked at some other potential
explanations for the difference in the Rasmussen numbers last
week. The problem with all of these comparisons and theories is that we are all
speculating. What we really need to test the Bowers theory is a very large
national sample that could allow a comparison between hard core likely Democratic primary
votesr and the larger universe sampled by the other national pollsters,
typically registered voters that identify as Democrats.
Back in November
and December
2005 the inaugural Cook/RT Strategies polls did just that. Each survey started with a sample of roughly 1,000
adults. RT Strategies provided cross-tabulations for each
survey (November
& December)
that included results for two subgroups of Democrats and Republicans: All
voters that identify or lean to a given party and the "hard core" primary
voters for each.
They defined "hard core" primary voters using two questions:
How often do you personally have
the time to vote in the primary elections or participate in political caucuses
when parties select their candidates for office-all the time, most of the time,
some of the time, or never?(BASED ON THOSE VOTING IN PRIMARIES
AT LEAST SOME OF THE TIME) And do you generally participate in primary
elections or party caucuses with the (ROTATE: Republican Party / Democratic
Party?)
Respondents qualified as a "hard core" Democrats if they
said they voted in primary elections or caucuses "most of the time" on the
first question and said they generally participate in Democratic Party
primaries or caucuses on the second. The November 2005 survey yielded 476
registered voters that identified or leaned Democratic, the December 2005
survey yielded 460 -- roughly 46-47% of all adults. Appropriately, hard core
Democratic primary voter universe was much smaller, just 169 respondents in November
and 181 in December -- or about 17-18% of all adults.
Neither survey yielded much in the way of big differences
between hard core Democratic primary voters and all other Democrats. The
November survey asked a complex question about Hillary Clinton (that took
different forms for different randomly selected respondents). When they rolled the difference versions together, the percentage agreeing that Clinton "would be a good candidate" was
five points lower among hard core Democrats (60%) than among all Democrats and
leaners (65%). While that difference is in the direction that the Bowers' theory would
predict, it was not quite statistically significant given the small sample sizes.
Only the December survey asked about Democratic vote
preference directly, and it showed virtually no difference. Clinton led among both groups, receiving 32%
from hard core Democratic primary voters and 33% from all Democrats and
Democratic leaners. Kerry and Edwards trailed with roughly 15-17% in both
universes. Of course, the survey did not include Obama among the potential candidates, and
a lot has happened in the nearly 16 months since.
Demographic differences, on the other hand, are still valid: The hard core
Democratic primary electorate was both older (53% over 50) and better educated
(46% college degree) than all Democratic identifiers and leaners (43% over 50
and 39% with a college degree).
But the main point here is that the only way to really test
the Bowers thesis is to do a similar test involving a very large national sample
of adults, or successive surveys rolled together to produce a large sample. Given
the heavy attention being paid now to the 2008 nominating contests and the easily
documented mismatch
between past primary turnout and the universe of respondent asked primary vote
questions (see also Bowers),
it is the least the national pollsters can do.