Results from October 21 to November 26, 1999 are represented in the figures below. All figures represent a five day moving average with a maximum margin of error of (+/-4.7%) for any five day period based on a minimum sample of 450. The daily data point represents the average of the previous two days and the next two days. For example, October 23 represents the average for October 21-October 25. Consequently the data are shifted backward by two days.

Sample sizes vary due to survey design. Data are weighted by household size and gender and age.

Please note: All data on this page are copyright and cannot be produced in any manner without the permission of the principal researchers.

Our estimates compared to the actual result.
 

	Overall				Last five days of campaign (Nov 22-26)
	nzes	actual	difference		nzes	actual	difference
Labour	40.5	38.7	1.8		41.7	38.7	3
National	30.7	30.7	0.0		30.3	30.7	-0.4
Alliance	8.2	7.7	0.5		7.7	7.7	0
Act	7.3	7	0.3		6.7	7	-0.3
Green	3.9	5.2	-1.3		5.9	5.2	0.7
NZ First	5.1	4.3	0.8		2.5	4.3	-1.8
Christian Heritage	1.6	2.4	-0.8		2.8	2.4	0.4
United	0.3	0.5	-0.2		0.9	0.5	0.4
Other	2.4	3.5	-1.1		1.5	3.5	-2.5

The Horse Race:

Both figures above and below include those who are probed after initially responding don't know. Probing reduces the proportion of don't knows by a third.
See the questionnaire for details.

Parties that poll consistently below one percent are not depicted on the figure. The maximum sampling error for the threshold parties is +/- 2.8% based on a  minimum sample of 450.

Figure above represents the proportion of undecided voters that remain in the sample after being probed. The undecided voters and those who intend not to vote are not included in the party estimates above.


Issues

.
 
 

Methodology
Our sampling frame consists of randomly selected residential households throughout New Zealand, based on listed numbers supplied by Telecom. When we reach a household, we randomly select an eligible voter based on who in the household had the most recent birthday. We will also arrange to call back respondents if the person randomly selected in the household is unavailable. Since we are polling daily throughout the campaign we have the advantage of contacting people who are difficult to reach. We leave very few phone numbers without either a response or a refusal, simply calling and recalling until we get a reply. We try to maximize our response rate by offering respondents a chance of winning a $300 gift voucher as an incentive to participate. Our response rate is currently 58 percent. The general rule is that a response rate of about 50% is the norm for telephone interviews. Sometimes, it can be considerably lower.

Aside from having a relatively high response rate, we interview daily until the election. Other organisations typically poll over 'snapshots' of several days only. Ideally we would like to minimise our sampling error by having a large sample each day. One way to compensate for a relatively small number of daily interviews is to construct a 'rolling average' of interviews conducted over five days. For each data point, we take not only that day but the two days before and after, and move on through the campaign, moving from one five day cross-section to the next, dropping the last day from the sample and adding the next. Over a five day period our budget allows us to achieve a minimum sample size of 450 which produces a margin of error of +/- 4.7%. Organisations that can afford to contact more people over a shorter time period have the advantage of minimising their sampling error over that period. However if their response rate is low then the results might be biased toward those who are easy to reach. In the longer run, over the whole campaign our sample size will be between 3,500 to 4,000, which will make it possible, long term, to estimate an average level of campaign support for parties with quite low margins of error.

We weight our data by household size, and across six age and gender groups to make sure each day's data represents the composition of the population 18 years and over. The questionnaire takes about 15 minutes, but we have a shorter version for those people who are unwilling to go through the longer version. This is another reason we have a high response rate for the crucial voting questions.

The tables below make the best comparisons possible between our party vote intention findings and those of other polls. We interview about the same number of people per day, whereas other pollsters may do more of their interviews on some of the days than others. During a campaign, opinions may shift over that period. Like most other polls, we 'probe' those who say 'don't know' by asking them which party they lean most towards, and include those in our party vote intention reports. However, there is little significant difference whether or not we include those respondents or not.

Early in the campaign, our polling tracked closer to the Herald-Digipoll than to TV3-CM Research. Indeed, we were quite close to Digipoll's first poll (see below). We tracked a shift to the centre-left over the period when Digipoll stopped and TV3-CM continued. We continued to track National support at levels similar to Digipoll though not as low as TV3-CM.

Polls are, of course, not predictions of election results. Opinion over campaigns is volatile. Generally, the poll which best 'predicts' the election result is the one taken the closest to election day.
 
 

*Note on the Digipoll polls: The exact dates are uncertain, as Herald indicates the polls taken (1) 'between October 29 and November 4' and (2) 'during the week to November 10'.  November 4 appears to overlap – we assume that is in the second, not the first poll, given that the first poll was published on the morning of November 5.
 
 

Return to NZES