Reply #5: A look at exit poll adjustments made in NC. [View All]

[skids]

Here is a cursory examination of what's useful in the available North
Carolina data, from the perspective of comparing it with the exit poll
data. The URL for that is:

http://www.sboe.state.nc.us/enrs/main_primary.asp?ED=11xx02xx2004&EL=GENERAL&YR=2004&CR=A

One of the files, voterstats11xx02xx2004.txt, contains very granular
demographic information on voter registrations, providing not only
counts by each demographic group, but by all combinations of
demographic groups, on a per-precinct basis. That is to say, instead
of knowing that we have 45 Black voters registered in a specific
precinct, 60 registered Democrats, and 34 seniors, we know precisely
that we have 18 Black Democrat Seniors. This level of detail is rare
and useful. However, to collapse the numbers into the simpler,
ordinary terms people are used to looking at, here is how the
registrations in that file break down statewide:

Party DEM: 2580437 voters
Party LIB: 13045 voters
Party REP: 1909589 voters
Party UNA: 1024341 voters
Party RFM: 1 voters

Sex F: 3021379 voters
Sex M: 2488016 voters
Sex U: 18018 voters

Race : 19 voters
Race A: 14951 voters
Race O: 78740 voters
Race W: 4226473 voters
Race B: 1114798 voters

Race ?: 50 voters
Race M: 9253 voters
Race I: 47308 voters
Race U: 35821 voters

Age 18 - 25: 618622 voters
Age < 18 Or Invalid Birth Dates: 26222 voters
Age Over 66: 876676 voters
Age 41 - 65: 2388889 voters
Age 26 - 40: 1617004 voters

In addition to the voter registration file, we also have a turnout
statistics file, turnout11xx02xx2004.pdf. This gives turnout by race,
gender and party on a county-by-county basis. For this overview I'll
not get into individual counties and stick to the statewide totals.

NC SoS turnout:
Race : 19 voters
Race ?: 50 voters
Race A: 74% of 14951 = 11064
Race O: 48% of 78740 = 37795
Race W: 66% of 4226473 = 2789472
Race B: 59% of 1114798 = 657731
Race M: 76% of 9253 = 7032
Race I: 43% of 47308 = 20342
Race U: 68% of 35821 = 24358

We can use this to build a chart against the state NEP exit poll final
weighted results, and I am also including a count from the NEP raw
data. Note this count only represents all people interviewed by NEP.
It does not take into account which survey group they were in, so my
speculations as to how the weighting was done are just that and not to
be taken too seriously. Though these questions do seem to have N
values approaching the total number of NC respondants in the NEP
data set so if I am lucky, this is a valid assessment.

                 White  Black Asian Other Unlisted
Exit poll         71    26     0     2
Raw Exit poll     73    22     0     2    2
SoS               79    19     0     2    1

This would suggest that the exit poll failed to interview whites, but
in the final adjustment, the numbers were tweaked to include less
whites (which should favor Kerry). NEP claims that whites had a 2%
lower completion rate than non-whites in North Carolina, so why they
did this is a mystery.

NC SoS turnout
DEM: 65% of 2580437   = 1677284
REP: 69% of 1909589   = 1317616
UNA: 54% of 1024341   =  553144
LIB: 57% of 13045     =    7435

                Dem    Rep    Oth
Exit poll       39     40     21
Raw exit poll   38     38     23
SoS             47     37     16

Note that this one is a little squirrely. The actual question asked
by NEP was "do you consider yourself" not how the voter was
registered. Neglecting that, this would suggest that the NEP severely
under-interviewed democrats, but then weighted in slightly in favor of
Republicans. The more likely scenario, as we'll see below, is that
Democrats were more likely to "consider themselves" independents.
In either case, the weighting advantage goes to Bush here.

NC SoS turnout
Male 63% of 2488016    = 1567450
Female 65% of 3021379  = 1963896
Unlisted 70% of 18018  =   12613

              Male   Female  Unlisted
Exit poll     41     59
Raw exit poll 44     55
SoS           44     55      0.3%

This would suggest that the NEP actually got an equal proportion of
males and females, but applied a final weighting *favoring* Kerry
(more females) on this demographic. NEP claims completion rates were
3% higher for females in North Carolina.

Unfortunately the SoS turnout figures do not include age. We have the
raw number of registrants, but not the percentage who voted. To see
what the NEP did, though:

                 raw            weighted
Age  - :       3 voters  0%
Age 18 - 24: 224 voters 10%    18-29  14%
Age 25 - 29: 175 voters 08%
Age 30 - 39: 453 voters 21%    30-44  33%
Age 40 - 44: 265 voters 12%
Age 45 - 49: 279 voters 13%    45-59  30%
Age 50 - 59: 365 voters 17%
Age 60 - 64: 179 voters 08%    60+    22%
Age 65 - 74: 169 voters 08%
Age 75 - :    68 voters 03%

The younguns were weighted down in the final numbers and the seniors
weighted up. According to their report, completion rates were
extremely low among seniors in NC, so weighting them up is to be
expected. However, completion rates among the young were lower than
the middle-aged, so why they were weighted down rather than the
middle-aged group is a mystery. That would be a weighting that would
benefit Bush.

I am not going to compare the "hispanic" ethnicity question because
the compatibility between the SoS and NEP figures might be iffy.

The end result here is that either the SoS's data is very inaccurate,
or the NEP severely overweighted Black and female Republicans (and
Republicans among the entire population in general) and underweighted
the youth vote.

There is some very interesting work to be done combining the SoS
turnout demographics file with the voter registration demographics
file. Because demographics are broken out by county, it should be
possible to enhance the demographics to more decimal places by rubbing
these two files together, and the demographics could be divided
further into subgroups (e.g. a separate turnout number for Black
Non-latino Democratic Seniors.) In addition, the calculations of the
sampling error of the NEP presented above could be broken down both
regionally and into these more specific demographic categories, though
with the limited sample size and distribution there will be
granularity limits.

Finally, there is the absentee ballot file absentee11xx02xx2004.txt.

There are 814004 absentee voters listed, by name, in the absentee
ballot file. These include both mailed absentee ballots and "One
Stop" early voting ballots. However another analysis that peeled out
the results marked "absentee" from the precinct-by-precinct results
file (priprecinct11xx02xx2004.txt) showed another 200K worth of
absentee votes in the final totals. This would seem to indicate some
sort of reporting inconsistancy, for example, absentee lists from a
few counties missing. Further investigation of this file might
indicate wherein that problem lies. The analysis I refer to is at the
following URL and attempts to suggest if there was anti-Kerry fraud, it
was in the polling, not in the absentee vote. (And of course we
cannot rule out anti-Bush fraud in the absentee vote, though that
would run counter to the exit poll disparity.)

http://www.yuricareport.com/ElectionAftermath04/NorthCarolinaAnalysis.html

Because absentee voters are listed by name and with demographics, it
is possible to get exact demographics for the group of voters that are
in the file (this can also be used in the enhancement of the general
demographics.) The NEP poll also will give us a raw breakdown of the
respondants to telephone surveys. How meaningful these two results are
when compared depends entirely on the demographic disposition of that
extra 200K absentee voters which we know nothing about yet, so this
would be data best considered "preliminary"

Absentee file versus absentee/early voters polled by NEP callers:

   SoS absentee data                vs.        NEP raw telephone poll data

Party DEM: 395505 voters        49%    35%      136 respondants
Party LIB: 1198 voters           0%
Party UNA: 113172 voters        14%    32%      122 respondants (includes
nonanswers)
Party REP: 304126 voters        38%    32%      122 respondants
Party RFM: 2 voters              0%
Sex F: 460924 voters            56%    55%      210 respondants
Sex M: 349524 voters            43%    45%      170 respondants
Sex Unlisted: 3556 voters        0%

What we can take from the above might be a suggestion that registered
Democrats consider themselves more independent when asked the NEP
question. Were this true it would emphasize the anti-Kerry weighting
above in the party affinity question.

No race or age figures are available in the absentee file, but for
comparison with the overall results, here are other NEP raw telephone
poll demographics:

Race Asian: 1 voters 0%
Race White: 286 voters 75%
Race Unanswered: 4 voters 0%
Race Hispanic/Latino: 5 voters 1%
Race Black: 71 voters 19%
Race Other: 13 voters 3%
Age 50 - 59: 67 voters 18%
Age 65 - 74: 59 voters 16%
Age 18 - 24: 21 voters 6%
Age 40 - 44: 30 voters 8%
Age Unknown: 1 voters 0%
Age 30 - 39: 63 voters 17%
Age 60 - 64: 53 voters 14%
Age 75 - : 27 voters 7%
Age 25 - 29: 20 voters 5%
Age 45 - 49: 39 voters 10%


Another interesting thing about the absentee file is that the ballots
are dated. It should be possible to construct a turnout graph of
early voting, by demographics, over time. In addition, the NEP data
was collected in three separate calls, so time-wise comparison of the
sampling may be possible, though the sample sizes would be very low.