Polling Methodology

Every poll published by Ace Strategies is fielded by DemosAU and carries a methodology statement in its own report. This page sets out what those statements contain, so a figure can be assessed before it is cited. Where waves differ — and they do — this page says so rather than describing a single procedure that no wave actually followed.

Who conducts the research

DemosAU is the research company. Ace Strategies commissions and publishes the South Australian waves; the federal waves are commissioned by Capital Brief or self-initiated by DemosAU, and the other state waves by Premier National or self-commissioned. Each report names its commissioning client, and the archive entry for each wave links the report.

DemosAU is a member of the Australian Polling Council and of the Research Society. Australian Polling Council membership requires publication of a methodology statement alongside published voting-intention figures, which is why every wave in the archive has one.

Sample construction

Respondents are recruited from internet panels. The target population is people aged 18 and over who are eligible to vote at elections — nationally for federal waves, and within the relevant state for state waves.

Two Tasmanian waves and the May–July 2026 federal MRP model used a mixed mode, combining internet panels with random device engagement. Those reports state the split: 91% panel and 9% random device engagement in February 2026, and 87% to 13% in October 2025.

Raw sample sizes in the published archive range from 931 to 2,694 for single-wave polls, and from 6,928 to 9,343 for the MRP models.

Weighting

Single-wave polls are rim weighted — iterative proportional fitting, cycling through each variable in turn until the sample matches known population margins on all of them.

The weighting variables vary by wave and are listed in each report. Across the archive they are:

• age by gender
• education
• personal income
• residential tenure
• language
• region, or state by region for federal waves
• past vote — the 2025 federal election, or the relevant state election
• left–right self-identification

Not every wave uses every variable. The Tasmanian waves, for instance, weight on age by gender, education, region and past state vote, without personal income or language.

Effective sample size and margin of error

Weighting reduces the statistical power of a sample. Each report states both the raw sample and the effective sample after weighting, and quotes the margin of error against the effective figure rather than the raw one — the more conservative of the two, and the one that should be used when judging whether a movement is real.

In the published waves the effective sample runs between roughly 50% and 65% of the raw sample. Quoted margins of error range from ±2.5% to ±4.4%.

How undecided voters are handled

Voting intention is asked as the first substantive question after demographics, so that earlier questions cannot influence it. Respondents who answer "don't know" are asked a follow-up leaner question: which option they would lean towards. Those who again decline are excluded from the published totals.

Reports state the proportion excluded — 7% on the May 2026 federal wave, for example. A primary vote total in the archive is therefore a share of decided voters, not of all respondents.

MRP models

Three waves in the archive are not conventional polls but multilevel regression and post-stratification models: October–November 2025, January–March 2026, and May–July 2026. These pool a much larger sample and model an outcome for each of the 150 House of Representatives electorates.

Post-stratification variables are age, gender, education, income, state, region and 2025 federal election vote; the May–July 2026 model adds religion. MRP models report no effective sample size, because the estimate for any given seat draws on the whole model rather than on respondents in that seat. Seat estimates carry an approximate margin of error of ±5%.

Two-party preferred

Two-party preferred is not calculated for every wave. Where a report states "no 2PP calculated", none is published, and no two-party figure should be attributed to that wave. Where 2PP is published, the report states the basis on which preferences were distributed.

What is published

Every wave is published as a page with its topline figures and as a full PDF report carrying the complete question wording, response categories, randomisation and demographic crosstabs. Both are linked from the polling archive, and each page carries machine-readable Dataset markup describing the fieldwork window, sample, margin of error and weighting.

One wave — South Australia, October 2025 — predates the methodology statements described here and its report carries none. Its sample and margin of error are published; its weighting procedure is not, and it is the one wave on this site without Dataset markup for that reason.

Citing a figure

Journalists and researchers are welcome to cite this polling. Please cite the wave rather than the site — for example, "DemosAU/Ace Strategies South Australian poll, fielded 12–18 March 2026, n=1,242, effective sample 836, margin of error ±3.4%" — and link the archive entry so readers can reach the full report. For questions about a specific wave, please get in touch.