2008 NZES
Variable Frequencies
The following links display frequencies from questions asked in the
2008 New Zealand Election Study and the full dataset is available at
the bottom of the page. The questionnaire can be downloaded from here. Other
detail concerning sponsors and collaborators can be accessed from the NZES website.
The dataset contains 3042 respondents, of whom 636 are an oversample
from the seven Maori electorates. Part of the data is a three-wave
panel to 2002 (948 respondents). Among the freshly-sampled respondents, the response rate was about 40 per cent.
The data is weighted by education (adjusting down the proportion of
university educated), age, gender, and voting as recorded at the 2008
election. Sample size is weighted down to reflect the size of
the general electorate sample, rather than the total sample. Voting data has been validated and corrected from the marked
electoral rolls.
IMPORTANT: TURNOUT ANALYSIS IS NOT RECOMMENDED
However, there are serious problems with the validated data.
Comparing our validated sample, including both respondents and
nonrespondents, to the official figures, we noticed that nonvoting was
about 4 per cent higher in our sample, and that special votes were
undercounted by about the same degree. This is well outside the limits
of sampling error (and with no non-response error to give any 'benefit
of the doubt'). We confirmed that the marked rolls themselves are
the source of this error with independent analysis of one of the
worst-affected electorates, and pointed this out to the Chief Electoral
Office. After investigation of several electorates that we had
highlighted, they have reported:
'the under-recording was due to
electorate headquarters staff not consolidating one or more certified
rolls used to record special votes in electorate headquarters onto the
master roll'. They were 'unable
to confirm this now, as the certified rolls have been destroyed by the
Clerk of the House in accordance with s. 189 of the Electoral Act'.
This was a final recording error, and the Chief Electoral Office
assures us that in all other respects the processing of the special
votes was as it should have been, and that the official figures are the
corrrect ones.
From the point of view of the NZES, this remains a serious problem. The
Chief Electoral Office has told us it will address the problem for
future elections. But the data to correct the error in 2008 is lost.
(Worse, although to a lesser degree, our validated data from 2002
and 2005 is subject to the same errors).
An error of 4 per cent in the sample may seem small, and, indeed, the
removal of this margin spread across the votes cast for political
parties in our sample is unlikely to do significant damage. But
from the point of view of analysis of nonvoters, the problem is severe
because of differential response rates. NZES data from 1999, before
this problem appears to have emerged, indicate that, then, voters had a
response rate of 50 per cent and nonvoters only 25 per cent.
Extrapolating this to a hypothetical sample of 1000, in the original
frame we might expect 44 special voters wrongly classified as
nonvoters, and 207 true nonvoters. Applying the expected response
rates, we might expect 22 wrongly classified special voters, and 52
true nonvoters. The extent of error in the sample of nonvoters is
therefore 2/7, too high for any comfort. Since 1999, response
rates have fallen, we suspect more so among nonvoters, so the problem
could be worse.
Now that it is clear that the data to correct this error no longer
exists, the NZES will continue to work on the problem. Those
electorates affected can be identified and deleted from the sample (a
strategy already being used for work on turnout drawing on the 2002 and
2005 samples).
Electorates dropped from turnout analysis of the 2002 and 2005 elections
In the longer run, we may seek to impute vote among those respondents
affected if a sufficiently robust predictive model can be specified.
The dataset now on release may therefore be improved upon in the
future, but other than for the matter of turnout we would not
expect anything to change except
marginally. For the time being, we do not recommend any analysis of turnout from the 2008 NZES. The
alternative strategy of relying on reported vote has been
considered, but with the data unweighted there are very few
self-reported nonvoters - 3.4 per cent of the general electorate
sample (N=77). (There are more in the Maori electorates, but
these are oversampled). To weigh the general electorate nonvoters
to their level in the official data would require multiplication by 6
or 7, an extent not to be recommended.
Section A
New Zealand Politics and You
Section B
The 2008 Election Campaign
Section C
Issues, Problems, Parties and Leaders
Section D
What are Your Opinions?
Section E
Party Preferences and Voting
Section F
Government and the Electoral System
Section G
Representation and Participation
Sections H, I
You and Your Background, Your Spouse or Partner
The 2008 NZES as an spss-saved file (zipped)