A New Zealand political poll estimates how a defined group of people say they would vote at the time they were asked. Its percentages are not a count of votes cast, a measure of every New Zealander’s views, or a guarantee of the election result. To interpret a poll, check who was asked, how they were sampled, what question they answered, which responses were included in the calculation and how much uncertainty surrounds the estimate.
How a political poll is conducted
A pollster defines the population the poll is intended to describe, recruits or selects a sample from that population, asks questions and analyzes the answers. Because it is not practical to ask every person in the target population, the result is an estimate based on the people who responded.
As an Amazon Associate I earn from qualifying purchases.
New Zealand polls commonly use phone interviews, online panels or a combination of methods. The New Zealand Population Review guide Political Polling in New Zealand (Satherley, Greaves and Sporle, 2023) describes several approaches to building a sample:
| Approach | What to look for |
|---|---|
| Random sampling | People are selected through a defined sampling process. Check which population the sampling frame can reach and whether some people may be missed. |
| Stratified sampling | The sample is organized around groups in the population. Check which groups were used and how the results were analyzed. |
| Quota sampling | Recruitment aims to reach specified numbers in selected groups. Quotas can help match measured characteristics, but do not by themselves make participation random. |
| Open online poll | People choose whether to take part, for example through a news site or social-media post. The guide cautions that this self-selected response is not a sound basis for estimating the views of the wider population. |
Sample size matters, but it does not prove that a sample represents the population. The 2023 guide gives 500–1,000 people as a typical range for nationwide polls in New Zealand and stresses that a well-designed smaller sample can be more informative than a larger, poorly designed one. The target population matters too: eligible adults, enrolled adults and people likely or intending to vote are different groups.
#1 Best Overall
Every recruitment method has possible gaps. A landline-only poll may overrepresent older people; an online panel may miss people without internet access or people in lower socioeconomic circumstances. Ask whether respondents were recruited by phone, from a panel, through quotas or by a mix of methods, and read that alongside the poll’s stated target population and weighting.
What a poll percentage actually measures
A result such as “Party A: 32%” is an estimate of the share selecting that answer within the poll’s analysis base. It does not automatically mean 32% of all New Zealanders, all enrolled adults or people who will ultimately vote for that party.
Check the denominator
Some party-vote results exclude respondents who are undecided, unsure or unwilling to answer. If they do, the published party shares are calculated among the remaining respondents, not everyone who was asked. For example, in a hypothetical poll of 100 people, if 40 choose a party and 10 are undecided, that party’s share is 40% of everyone asked but about 44% of the 90 who gave a decided answer. The poll report should make clear which base it uses.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesThe denominator can differ by question. A party-vote result might exclude undecided responses, while a preferred-prime-minister question might handle “don’t know” answers differently. Do not assume that percentages in different questions, polls or reports use the same calculation.
Read the question, not just the headline
“Who would you vote for?” asks about stated voting intention; “who do you think will win?” asks for a prediction. They measure different things. Exact wording, answer options, question order and timing can also shape responses. A useful report identifies the question as asked, the polling organization, the organization that commissioned it and the field dates.
What the margin of error does—and does not—tell you
A margin of error describes the random variation expected from sampling under stated assumptions, usually at a specified confidence level. It is commonly shown as a number of percentage points to add to and subtract from an estimate. It is not a promise that the true result must fall inside that range, and it does not account for every way a poll can be wrong.
The margin depends on both the sample size and the estimated share. It is generally largest near 50% and smaller for more extreme estimates. The 2023 New Zealand Population Review guide’s table gives illustrative 95% margins of 3.1 percentage points for a 50% estimate and 1.9 points for a 10% estimate, each with a sample of 1,000. Those are examples under standard assumptions, not a universal uncertainty guarantee for every poll design. In particular, weighted or non-probability samples may need a different treatment of uncertainty. A headline maximum margin should not simply be applied to every party result; look for the uncertainty associated with the specific estimate.
Small subgroups have more uncertainty
When a poll divides its respondents by age, region, gender or another characteristic, each subgroup has fewer people than the full sample. A subgroup result therefore needs its own sample size and uncertainty information. Treat claims based on small cells cautiously; the overall poll’s margin does not describe every subgroup result.
Sampling error is only one source of error
A poll can miss the mark because its sampling frame fails to reach part of the target population, because some types of people are less likely to respond, because an interviewer affects an answer, or because the question is unclear or leading. A reported sampling margin does not quantify these coverage, nonresponse or measurement problems.
How weighting helps—and where it stops
Weighting gives more or less influence to responses so that measured characteristics of the sample better match population benchmarks. Pollsters may weight on characteristics such as age, gender, education or ethnicity. A transparent poll report should identify the variables used.
Weighting can address measured imbalances; it cannot correct a flaw that was not measured. It cannot, for example, repair poor question wording or guarantee that respondents and nonrespondents differ only in the characteristics used for weighting. Treat weighting as one part of the method, not proof that all bias has been removed.
Recommended Free Tools
How to compare polls and read changes over time
A one- or two-point change between adjacent polls may reflect random variation rather than a real shift in support. Avoid assigning a cause to a movement from the poll alone. Compare a sequence of polls, their field dates and the methods used; changes in events, party leadership or the time remaining before an election may be relevant context, but are not established as the cause by a poll result.
Rank #4
- Used Book in Good Condition
Before comparing two headlines, check these details in each report:
- Population: Were respondents eligible or enrolled adults, likely voters or another group?
- Sample and recruitment: Was the poll conducted by phone, online or using mixed methods? Was the sample random, stratified, quota-based or self-selected?
- Weighting: Which variables and population benchmarks were used?
- Question: Was the wording, order and set of answer options comparable?
- Denominator: Were undecided, unsure and refused responses included or excluded?
- Timing and uncertainty: What were the field dates and sample size? Is the reported margin specific to the result or a headline maximum, and what is the base size for any subgroup?
The Research Association of New Zealand (RANZ) describes its revised 2020 Political Polling Code as rules for member organizations and best-practice guidance for publishing stories about polls. RANZ says the revision accounted for changes in data collection and reporting since 2014. The code is guidance and member rules, not legislation.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why a party-vote poll is not a government forecast under MMP
New Zealand’s mixed-member proportional (MMP) system means party-vote support does not mechanically determine who will govern. The 2023 New Zealand Population Review guide describes a 5% party-vote threshold for entering Parliament unless a party wins an electorate seat. A party near that threshold deserves especially careful interpretation: uncertainty around its poll estimate may cross the boundary. Check the rules applicable to the election being discussed rather than treating a poll guide as an election-specific ruling.
Free tools Windows power users keep installed
One-click scans. No signup required.
A party’s poll share is not a seat total. Representation also involves electorate-seat outcomes, and a party leading a poll is not automatically able to form a government. Seat projections therefore depend on assumptions about electorate results and potential governing blocs. Treat a projection as conditional on those assumptions, not as a direct reading of the poll percentages.
Best Value
Be particularly cautious about applying national results to Māori electorates. The 2023 guide notes that polling in those electorates is relatively scarce, undecided shares can be high, and the smaller target populations and available sampling frames create constraints. A national party-vote result does not establish support in a particular electorate.
The Electoral Commission’s 2026 Broadcasting Allocation Decision illustrates a separate use of polling: it considers averages across multiple named poll series and specified historical date ranges for an allocation decision. Those averages belong to that decision’s context; they are neither a current election forecast nor the result of one pooled poll.
What a responsible poll report should include
When a news story gives a poll headline, look for the full report or enough details to judge the estimate. A useful report states:
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →- the polling organization and the commissioner;
- field dates and target population;
- sample size and recruitment method;
- weighting variables;
- the exact question and response options;
- how undecided, unsure and refused responses were treated; and
- the margin of error or other uncertainty information, including subgroup bases where relevant.
The Broadcasting Standards Authority’s decision Chapel, Garbutt & Hopcroft and Television New Zealand Ltd — 2024-042, dated 2 September 2024, records that presenter Simon Dallow described a poll as “a snapshot in time.” The decision also discusses public understanding of polls as a rough tool for gauging opinion from a sample, not as definitive results. That is the right frame for a headline: a poll reports answers gathered from a particular sample at a particular time.
Limited election-advertising caution
The Electoral Commission’s page “Telephone canvassing, polling or surveys,” dated 1 March 2011, says that telephone canvassing or a survey or poll that can reasonably be regarded as encouraging or persuading people to vote a particular way may count as an election advertisement under the Electoral Act 1993. Promoters publishing an election advertisement must include the required promoter statement. The page is dated; anyone making decisions for a live campaign should check current legislation and Electoral Commission guidance.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
