[And implicitly: problems with a fundamental mismatch between the exit polls and the tallied vote. The reason none of the internals match up with the county level data is because the final exit polls still said Kerry won. They had to weight Kerry voters down and Bush voters up to match the votes. And when they do that, all the internals, not just the Hispanic ones, seem to go blooey when compared to the county-level numbers]
[Even if you don't think there was voter fraud, that still seems to pose the question how come the exit polls were so suddenly and unprecedently incompetent that nothing matched up. (None of this was true in 2000, according to Teixiera.)]
[BTW -- the Teixeira analysis, while a little dense, gets better and better as it goes on. My favorite: that for the numbers to match up, Hispanics in the states that don't break them out would have had to vote for Bush at rates of over 100% to make it all work out.]
http://www.alternet.org/story/20606/
November 24, 2004
Public Opinion Watch
By Ruy Teixeira
Did Bush Really Get 44 Percent of the Hispanic Vote?
I very strongly doubt it. This claim is based, first and foremost, on
the finding in the National Election Pool (NEP) exit poll, the
nation's largest and by far most influential, exit poll. But that
finding, if carefully scrutinized, seems highly implausible for a
variety of reasons. I lay these out below and conclude that a more
reasonable estimate for Bush's Hispanic support this year is around 39
percent.
Start with the Texas state exit poll. That poll shows Bush with an
astonishing 59 percent of the Hispanic vote. That's an increase of 16
points in Bush's support over 2000 and a shift in margin of 29 points
(from an 11-point deficit to an 18-point lead).
The poll also claims that this mega-shift happened at the same time
that Bush's support was being compressed among whites. Bush's support,
the exit poll claims, dropped by a point among Texas whites compared
to 2000, at the same time as Kerry's support among Texas whites rose
by four points compared to Gore's. So Texas's favorite son runs for
re-election and widens his margin among white voters practically
everywhere - except Texas, where he loses ground! But among Hispanics
in Texas, he gets a massive 29-point shift in his favor?
This pattern just doesn't make sense. But where the Texas poll makes
the least sense of all is when you try to match them up with the
county-level voting returns. If Bush was pulling over 70 percent of
the white vote and almost 60 percent of the Hispanic vote, how on
earth did he lose any counties in Texas? Consider these (racial
composition figures based on voting age population):
Brooks county: 90 percent Hispanic, 10 percent white - 68 to 32
percent Kerry.
Dimmit county: 83 percent Hispanic, 16 percent white, 1 percent black
- 66 to 33 percent Kerry
Duval county: 86 percent Hispanic, 13 percent white, 1 percent black -
71 to 28 percent Kerry
El Paso county: 75 percent Hispanic, 20 percent white, 3 percent black
- 56 to 43 percent Kerry
Hidalgo county: 85 percent Hispanic, 14 percent white - 55 to 45
percent Kerry
Jim Wells county: 73 percent Hispanic, 23 percent white - 54 to 46
percent Kerry
Maverick county: 94 percent Hispanic, 4 percent white - 59 to 40 Kerry
Starr county: 97 percent Hispanic, 2 percent white - 74 percent to 26
percent Kerry
Webb county: 94 percent Hispanic, 6 percent white - 57 percent to 43
percent Kerry
My, my, where could those 59 percent Bush-voting Hispanics be hiding
in the great state of Texas? Perhaps in the big urban areas such as
Harris county (Houston)? Well, let's see, if we figure Hispanics are
at least a sixth of Harris county voters (probably more, but let's be
conservative), then, by themselves, they would push up Bush's margin,
compared to 2000, by five points if they really voted for him at the
59 percent rate (and it should be even higher - to balance the
apparently way-under-59 percent Hispanics in these other Texas
counties). But wait! Bush's margin actually contracted in Harris
county by a point. Maybe black voters (18 percent of the Harris county
voting-age population) moved the needle back the other way? Seems
unlikely if we believe the Texas exit poll: it says Bush improved his
margin among black voters by 19 points in 2004!
That just deepens the mystery. To account for the slight shift away
from Bush in Harris county, we would then have to assume that Harris
county whites reduced their margin for Bush by 12 points or more in
2004.
Similar exercises could be performed on other counties, but these
examples should suffice to make the point: the 59 percent figure, as
common sense would suggest, is clearly a gross overestimate of Texas
Hispanics' support for Bush in 2004.
That puts the national exit poll figure for Hispanics off to a bad
start. In 2000, Texas Hispanics were 10 percent of the national exit
polls' Hispanic sample and this year they will likely be substantially
more (the latest census population projection put Texas Hispanics at
19 percent of the nation's Hispanic voting-age population and the
Texas exit poll has Hispanics at 23 percent of Texas voters this year,
compared to just 10 percent in 2000). And we would expect Bush's
support in the southern region of the national exit poll, which
includes Texas, to be particularly skewed by the Texas figure. That it
is, it's... 64 percent! Wait a minute - 64 percent? That's even higher
than the Texas figure! Maybe it's the inclusion of Florida in the
southern region sample? Nope, the Florida exit poll says Hispanics
voted 56 percent for Bush, three points less than their Texas
counterparts (amazing in and of itself!).
Only two other states in the southern region (Georgia and Oklahoma)
have Hispanic breakouts available, so we can't directly find all the
missing pro-Bush Hispanics. But, as the astute conservative analyst
and number-cruncher, Steve Sailer, has calculated, if you take the
given Hispanic Bush support rates for the four available states and
figure the number of Hispanic Bush votes that implies from those four
states, you can then estimate how many Hispanic Bush votes must have
come from the non-broken-out states (given their percentage of overall
voters in those states, which the NEP has released) to produce the
number of southern Bush Hispanic votes indicated by the 64 percent
support figure. Well, I suppose the Hispanics in those other states
could have produced those missing votes - but only if they voted early
and often: they would have had to support Bush at the rate of 190
percent! (Read Sailer's analysis in its entirety for all the details
on these calculations.)
There are similar problems with the other regions of the national exit
poll. In the west, the NEP says that Bush's Hispanic support rose by
eleven points (from 28 to 39 percent). But the NEP California state
exit poll says that Bush's Hispanic support in that state only rose by
four points over 2000 (from 28 to 32 percent). Given that California
Hispanic voters are over three-fifths of this entire region's Hispanic
voters, that puts a heavy burden on the other states of the west to
produce this eleven-point jump in support for Bush. Indeed, as Sailer
has calculated, once you take into account the other released Bush
support rates for Hispanics in western states, Hispanics in the
remaining states in the west must have supported Bush at the rate of
167 percent to reconcile the released state figures with the western
region figure.
Sailer's similar calculations for the midwest (123 percent Bush
support among Hispanics in non-broken-out states) and the east (95
percent) show this problem affects all regions, albeit not as severely
as the south and west.
Okay, so what's the explanation for this particular set of anomalies?
That is, even accepting all the various state-level Hispanic figures
as gospel, including the absurd Texas figure, why do we get these
crazy mismatches between the state figures and the regional figures
from the national poll?
It seems to me there are two logical possibilities. One is that the
Hispanic respondents included in the national poll systematically
differ from those included in the state poll. So, for example, if
Texas Hispanics in the state poll support Bush at 59 percent, those
Texas Hispanic respondents included in the national poll support him
at, say, 67 percent. Or California Hispanic respondents in the
national poll support Bush at 39 percent, not 32 percent. And so on.
That strikes me as less likely than the other possibility. We know the
national exit poll took some pretty serious weighting to get it to
match up with the actual election figures. This suggests that, for
example, even Hispanics that were already sampled/weighted in the
Texas exit poll to have a 59 percent support rate for Bush were
probably further weighted toward Bush in the process of getting the
national exit poll "corrected." The same logic would apply to the
other states - Hispanic respondents from those states in the national
poll got an additional push toward Bush that makes their Bush support
rates higher than those measured at the state level.
If this has happened, it's worth noting that in the 2000 Voter News
Service poll this problem does not appear to have occurred. If you
take the Hispanic proportions of voters in each state in the 2000 poll
and the Hispanic support rates for Bush in each of those states, you
can calculate a state-based 2000 Bush support rate and compare it to
the national rate. They are very close: the state-based rate is 34
percent and the national rate is 35 percent.
All this leaves us with a question: if 44 percent is the wrong level
for Bush's support among Hispanics, what is the right level? Of
course, we'll never really know for sure, but I am persuaded, by
playing with the numbers and making some reasonable assumptions to
correct the anomalies in the NEP that it is somewhere around 39
percent. That is also Sailer's conclusion and that of the National
Council of La Raza, whose extremely useful review of 2004 poll and
voting data on Hispanics I recommend to you.
If the 39 percent figure is about right, that would mean Bush improved
his standing among Hispanics by four points - about his gain in
support among voters overall. That makes sense to me and is certainly
no cause for complacency among Democrats. But there is no reason to
panic either: Bush made gains among Hispanics, as he did among most
voter groups, but not a breakthrough.
Source used for this section:
Edison Media Research/Mitofsky International exit poll of 13,360
voters for National Election Pool, released November 2, 2004
(conducted November 2, 2004)