How this works

The short version

You answer up to 30 statements on a five-point scale and say how much each issue matters. For every statement we hold a position for each of the six parliamentary parties, on the same five-point scale, taken from their published policy. We compare your answer with theirs, weight it by how much you said it mattered, and average across everything you answered. That's your alignment score.

There is no AI, no model and no black box. The same answers always give the same result. Every number can be traced to a specific question and a specific published party position.

How questions were chosen

Six sectors, five questions each, so no topic dominates just because it generated more headlines. The sectors are:

• Economy, Tax & Cost of Living, Who pays tax and how much, how big the state should be, and who should own the banks, power companies and supermarkets.
• Housing & Renting, Rent rules, state house building, help for first-home buyers, and whether to loosen planning rules to build more.
• Health & Education, Free doctor's visits, how schools teach and test, school choice, and whether Māori-led health services should return.
• Climate, Energy & Environment, Emissions targets, oil and gas, farming's climate obligations, mining, and river water quality.
• Law, Order & Justice, Sentencing, prisons, police numbers, gangs, and whether justice should lean toward punishment or prevention.
• Te Tiriti, Rights & Immigration, The place of Te Tiriti o Waitangi in government, the Māori electorates, te reo Māori, and how many people should move here.

Statements were chosen to separate the parties. A question everyone agrees on tells you nothing about who to vote for, so questions where all six parties held the same position were cut. Several statements deliberately cut across the usual left–right line immigration, state ownership of banks and power companies, and the superannuation age all split the blocs in ways that surprise people.

How party positions were sourced

Every position follows three rules:

• Strongly supports / opposes (±2) requires an explicit published commitment, a manifesto line, a policy page, or a reported campaign announcement.
• Supports / opposes (±1) is used where the direction is clear from published material or voting record, but the party hasn't committed to that specific measure.
• Mixed / no clear position (0) means exactly that, and the note says which of the two it is. We don't guess.

All 180 positions are published in full, with their sources and verification dates, on the where every party stands page. Positions were last verified on 13 August 2026 and will be re-checked as parties release more policy before 7 November.

The scoring maths

Alignment score

Your answer and the party's position are both numbers from −2 (strongly disagree) to +2 (strongly agree), so the furthest apart you can be is 4. For each question:

agreement = 1 − |your answer − their position| ÷ 4

That gives 1.0 for a perfect match and 0 for perfect opposition. Each question is then multiplied by its weight, not much counts half, normal counts once, a lot counts double, skipped questions count nothing, and the whole thing is averaged:

raw = Σ(agreement × weight) ÷ Σ(weight)

One more step matters, and it is the difference between a fair result and a rigged one. Parties do not all have the same range available to them. A party that takes firm positions can score 0 against someone who disagrees with all of them. A party whose positions are mixed or undeclared cannot, because “no position” is never far from any answer: on this bank Labour's lowest possible score was 22% while ACT's was 2%. Averaging raw distance would therefore reward vagueness. So each party is stretched across the range it can actually occupy:

alignment = (raw − floor) ÷ (1 − floor) × 100

Floor is the score that party would get from someone who disagreed with it as strongly as the scale allows, calculated from the same questions you answered and the same weights you gave them. The effect is that 0% and 100% mean the same thing for every party. Scores are not otherwise massaged to make the winner look decisive, so 50% still means you agreed with that party about as often as you disagreed, and if the top two are within 3 points the result page calls it a tie rather than picking one.

The spectrum chart

Each question carries a coefficient for two axes: economic (state ↔ market) and social (progressive ↔ traditional). Your position is the weighted sum of your answers against those coefficients, normalised to a −10 to +10 range.

Crucially, the party dots go through the identical function, using their own stances at equal weight. No party position on the chart is hand-placed, which is what makes your dot and theirs comparable. Here's where that calculation puts them:

The chart and the percentages answer two different questions, so they can disagree, and when they do the percentage is the one to trust. Your percentage against a party is worked out from all 30 answers, one at a time, weighted by how much you said each issue matters. The chart takes those same answers and compresses them into two numbers, which throws away everything that doesn't load onto one of the two axes. A party can therefore sit nearer your dot while another party agrees with you on more of the actual questions.

There is a second reason they can differ. Party dots are fixed, placed from each party's own stances at equal weight, because a party has no view about which issues matter most to you. Your dot moves with your importance ratings. Mark climate as high importance and low on tax, and your dot shifts even though the party dots stay where they are.

How we test for our own bias

A quiz like this can be slanted without anyone intending it, just by writing more statements where "agree" happens to be the left-wing or right-wing answer. So an balance is measured every time the questions change, and the results have to hold before anything is published. We check that:

• Agreeing with every statement leaves you near the centre on both axes, rather than in a corner.
• Roughly as many statements favour the left bloc as the right bloc.
• Answering exactly like a party returns that party first, at 100%.
• No party's average stance across the bank is extreme, nobody is set up to be the "obvious" answer.
• Every position carries an explanation, a source and the date it was last checked.

This caught a real problem. The first version of the question bank leaned left economically: someone agreeing with everything came out at −3.3 on the economic axis, and 18 of 30 statements favoured the left bloc against 12 for the right. Four statements were rewritten so that agreeing is the right-leaning answer the same facts, framed from the other direction. It now sits at −0.1, with a 14/16 split.

“How representative is this?”

This is the first question researchers ask, and it needs a straight answer: NZVoteCheck is not a survey. There is no sample and no population estimate. One person answers, and we compare their answers to party policy. Nothing is inferred about New Zealanders as a group, so there is no sampling frame to be unrepresentative of and no margin of error in the polling sense.

That doesn't make the underlying concern wrong, it just points at three different things, and we take all three seriously.

1. Coverage, who this tool reaches

A 30-question web tool in English systematically under-serves people who are older, rural, on poor connections, not confident online, disabled, or who don't read English easily. Those groups are not politically random. This doesn't distort any individual's result, but it does mean the people who benefit from the tool are not a cross-section of the country, and a voting guide that only reaches the already-engaged is doing less good than it appears to be.

What we do about it: the site works without JavaScript-heavy interaction, runs on a slow connection, is keyboard-navigable, carries screen-reader labelling on the chart, and every question can be skipped. What we can't fix from here: it's still a website, and it's still in English.

2. We publish no aggregate statistics

We publish no aggregate statistics about how people answered. Answers are only ever stored with the sharing box left ticked, and even then they are anonymous and never published as a representative picture of the country. That is a privacy decision first, but it is also a methodological one: self-selected quiz results are not a poll. People who choose to take an online political quiz differ from the electorate in exactly the ways coverage bias predicts. If we ever did publish aggregates, representativeness and margin of error would apply in full, the results would need demographic weighting against census benchmarks, and a stated design effect. Until then, treat any “NZVoteCheck says the country is X” claim as something we did not say.

3. Measurement error, the real margin of error

The honest equivalent of a margin of error here is: how fragile is your result? Across 4,000 simulated answer sets, the median result could be flipped by changing any one of 5 questions, and 43% landed within 3 points of a tie.

But for coherent answer sets, generated from a party's actual platform with up to 12 of 30 answers randomised, the tool still returned the right party 97.7% of the time. In plain terms: if you have a consistent worldview, the result is reliable. If your views genuinely straddle several parties, the percentages on your results page will sit close together, and that closeness is the signal. Two parties within a point or two of each other should be read as equivalent, not ranked.

The moderate-party advantage, and how we removed it

Scoring by agreement distance structurally favours parties with middling positions. A party that takes a moderate or undeclared stance is never far from any answer, while a party with strong positions scores very high for people who share them and very low for everyone else. Almost no tool like this discloses that, so here is what it did to this one and what we did about it.

The test is to feed the quiz answer sets picked entirely at random. With no real person involved, every party should come top about as often as any other. Under plain distance scoring they did not: Labour came top 58% of the time and ACT 3%. Labour had not gained an advantage on the merits. It simply had the least committal set of scored positions, six questions with no published position and fourteen more at the softer end, against ACT and the Greens with none.

Rescaling each party across the range it can actually reach, described in the scoring section above, fixes it at the source. The same test now returns every party between 14% and 19%, where a perfectly even result would be 16.7%. Crucially this did not blunt the tool: a voter built from a party's own platform with nine of their thirty answers randomised is still matched to that party about 99% of the time, the same as before.

We changed the maths rather than the questions on purpose. The alternative fix people reach for, tilting the question set to compensate, would have meant introducing a second bias to cancel a first one, and it would have shown up in the balance checks below as exactly what it was. The residual spread of a few points is honest noise, not a thumb on the scale.

How the importance weighting works

Each question can count half, once or double. Formally the weight multiplies both the agreement and the total, so it shifts the balance between questions rather than inflating any score. Across simulations, varying importance changed the top party in 26.8% of cases and moved scores by about 2 points on average, so it genuinely matters.

One consequence worth knowing: marking everything “a lot” is mathematically identical to marking everything “normal”. Weighting only does something when you distinguish between issues.

How good are the sources, really?

“Every position is cited” is only meaningful if the citation actually substantiates the claim. An early version of this bank leaned heavily on generic policy index pages, 27 of 30 Labour positions and 23 of 30 NZ First positions pointed at a single page, and in several cases that page didn't contain the policy area at all. That is the kind of thing a hostile reader finds in ten minutes, and they'd be right to.

Positions have since been re-sourced against specific manifesto pages, party press releases, reported announcements and the legislative record. Every source is tiered: primary (the party's own material), reported (a news report of a specific announcement), record (legislation or a government release showing what a party did in office), or secondary (an aggregator, used only where nothing better exists). Current spread of distinct sources per party: ACT 11, NZ First 12, National 10, Labour 8, Greens 7, Te Pāti Māori 4.

Te Pāti Māori is the remaining weak spot, 27 of its 30 positions rest on its own policy page, because that page is genuinely its main published statement across all six sectors and it carries less sector-by-sector detail than other parties' material. We'd rather say that plainly than pad the citation list.

Known limitations

Two axes can't hold everything

The Greens and Te Pāti Māori land close together on the chart, because they genuinely agree on most economic and social questions in the bank. What most distinguishes them constitutional transformation and tino rangatiratanga as the organising priority rather than one policy among many, isn't really an axis. We could have spread them apart artificially. We didn't, because that would mean misrepresenting their published positions to make a prettier chart.

Policy isn't the only reason to vote

This measures agreement with published policy. It says nothing about competence, track record, whether a party will deliver, who they'd go into coalition with, or your local electorate candidate, all legitimate reasons to vote differently from your result.

Parties we left out

Only the six parties currently in Parliament are scored. That's a real limitation, not a neutral choice, so here's who's missing and why:

• The Opportunity Party (TOP), Polling strongly enough to enter Parliament: 8% in the 1News-Verian poll of 8–11 August 2026, and forecast a seat share by the Poll of Polls. It is excluded for one reason only, and not a political one. We checked all nine of its published policy documents against our 30 questions and found a citable position on 13 of them. The other 17, including the Paris Agreement, oil and gas, sentencing, police, gangs, rent, state housing, NCEA, charter schools and all immigration questions, have no published TOP position; several of its policy pages say detail is still to come. Scoring a party on 17 blanks would not be neutral: a 'no position' sits close to every possible answer, so TOP would out-match every other party for almost every user as an artefact of the maths rather than a fact about politics. We will add TOP the moment its published policy covers enough of the bank, and this note will say so.
• Outdoors & Freedom, Legalise Cannabis, Animal Justice, and other registered parties, Currently polling below 1% with limited published policy across all six sectors. Scoring them would mean inferring positions we cannot cite, which would undermine the rest of the tool.

Positions move

Parties release policy right up to election day, and some of these positions will change. Each one carries the date it was last checked. Anything stale is a bug, tell us.

Your privacy

Scoring runs entirely in your browser. Answers are kept in your device's local storage so you can close the tab and come back, and they're encoded into the share link if you choose to share one. There is no account and no login.

The “add my answers” box

At the end of the quiz, before your results, there is a sharing box directly above the “see my results” button. It is ticked by default and the tool works identically if you untick it. With it ticked, we store: your answers, your resulting scores, any optional background details you chose to give, and which version of the question bank you answered. We do not store your name, email, IP address, browser, device, location, or any identifier that could link a response back to you, so once submitted, even we cannot tell which response was yours.

Because we can't identify your response, we also can't retrieve or delete an individual one on request. That is a deliberate trade: it is the reason the data is safe. If you would rather not share, untick the box before viewing your results. The results page always states whether your answers were stored, so nothing happens silently.

Political opinions are sensitive information under the Privacy Act 2020, which is why this is anonymous by construction, disclosed at the moment it happens, and limited to the single purpose of understanding how the quiz is used and whether it discriminates properly between parties.

Independence

NZVoteCheck is not affiliated with, funded by, or endorsed by any political party, candidate, or lobby group. The site's own colours were picked specifically to avoid every party's brand colours so the design itself doesn't nudge anyone.