"Black Voters Shift Support." That's the front page headline
on the Washington Post's story
today summarizing new results from the latest Post/ABC News poll (full
results, ABC News story
& results).
Among Democrats, Hillary Clinton's "once-sizable margin" over Barack Obama has
been "sliced in half during the past month largely because of Obama's growing
support among black voters" In two polls conducted in December and January,
African-American Democrats preferred Clinton 60% to 20%, but Obama now leads by
a 44% to 33% margin.
But - as astute poll readers will wonder- the Post story notes that the survey
included an "oversample" of 86 black adults. Is that enough to make the result "statistically
significant and does it justify the front page emphasis the Post gave it? The short
answer is yes and maybe. The longer answer follows.
First, the reference to a black oversample in the Post's
methodology blurb may leave some readers a bit confused:
The Post-ABC News poll was
conducted by telephone Feb. 22-25 among a random sample of 1,082 adults,
including an oversample of 86 black respondents.
If I'm reading it correctly, this means that in addition to
the black respondents interviewed as part of an initial random sample of 996 adults
(probably 80 to 100), they interviewed another
86 another black respondents with a separate "oversample." The full sample of
1,082 adults was then weighted so that the percentage of African Americans in
the poll results matches the U.S. Census estimate for the adult population.
What we do not know from the story is (a) exactly how many total
African Americans were interviewed in the latest poll and (b) how many answered
the Democratic primary vote preference question. We also do not know the sample
size from the previous surveys, but the 60% to 20% that the Post and ABC News reported was based on
pooling results from two national surveys conducted in December and January.
Also, on re-checking the previous ABC News release
that I wrote
about in late January, the result among African-Americans cited at that
time (also based on pooled samples from December and January) had Clinton
leading Obama 53% to 27% (not 60% to 20%). I am not sure why those would
differ, but will inquire further. [UPDATE (3/2): Washington Post polling director Jon Cohen confirms via email: "the
original ABC numbers were incorrect, they've updated their analysis to the
correct 60-20."]
I emailed the Post's
Jon Cohen this morning seeking more details, and while he was up against a
deadline he confirmed that the reported shifts were statistically significant
at a 95% confidence level. That result is not surprising. My educated guess is
that the African American sample size on this survey is in the ballpark of 125
to 150 interviews, and the previous pooled sample probably about the same size.
That implies a margin of sampling error for each sub-group of roughly +/- 8. At
that level, the 27-point drop for Clinton
and the 24-point increase for Obama would both be statistically significant. More
details on that later, hopefully.
The direction of the shift certainly seems plausible. Separately,
the poll shows a big jump in Obama's favorable rating among African Americans, from
54% to 70% since December and January. Ironically, the initial Post/ABC poll
result (along with similar findings from other pollsters) helped generate a
spate of stories speculating about Obama's appeal (or lack thereof) among black
voters, including a profile on 60 Minutes. These stories all included significant
detail on Obama's background and racial heritage, which likely contributed to greater
recognition and favorability.
Of course, there is also good reason for caution about these
shifts. First, keep in mind that a 95% confidence level still leaves open a 5%
chance that the shifts occurred by chance alone. In other words, we should
expect one sample in twenty to produce a "significant" result that is not
really significant.
Second, as we remind
readers often, statistical sampling error is just one source of potential
variation (and error) in surveys. Other factors - including rates of response
which tend to be a bit lower with African Americans - can also introduce variation
not accounted for by the "margin of error."
Finally, while these results provide evidence of a recent shift
toward Obama among African-Americans, that shift may be less "dramatic" than
today's numbers make it appear. A range of random variation of +/- 8 to 10
points in support for either candidate on either set of polls allows room for the
possibility of considerably smaller shift within the range 95% confidence.
Given that Clinton
enjoys nearly universal name recognition, while roughly half of African
Americans did not know Obama well enough to rate as of about a month ago, I
would expect to see Obama gain among African-Americans gradually over the next
year. How high he goes is anyone's guess, but his support should increase. So I
would not be surprised to see other polls showing a shift to Obama among
African Americans that is less "dramatic" than the one in evidence in the current
Post/ABC poll.
Of course, I am just speculating. If we can be patient, more
polls should bring clarity. But the trend is intriguing and the race for the
White House more fun to watch than ever.