How this site works

The rule everything else follows

Every figure carries its source: who published it, a link to the original, when they published it and when we checked it. A number without a source cannot be added to this site — the site fails to build rather than showing you something unattributed.

When a source does not report something, you will see not reported. We never fill a gap with an estimate, carry a figure over from an earlier release, or infer one from a remainder and then present it as if it were measured.

Polling data

The tracker currently holds 134 national polls from 13 polling organisations — Agenda C Synesis, DemosAU, DemosAU (MRP), Essential, Fox & Hedgehog, Freshwater, Newspoll, RedBridge/Accent, Resolve, Roy Morgan, Spectre Strategy, Wolf & Smith, YouGov — with fieldwork between 18 May 2025 and 16 July 2026.

Each poll is placed at the midpoint of its fieldwork, not its publication date. When a poll is released tells you nothing about when opinion was measured, and some fieldwork runs for weeks.

Pollsters report differently. Some publish a two-party-preferred figure, some do not. Some split the Coalition into Liberal, Liberal National and National columns; others report a single figure. Where a pollster publishes a two-party-preferred pair that includes undecided voters — so it does not add to 100 — we exclude it rather than mix a non-comparable number into the trend.

How the trend line is calculated

The line is a weighted average of nearby polls, recalculated for every day on the chart. Polls closer to a given date count for more, on a smooth curve with a 21-day span, and larger samples count for more than smaller ones.

A wider span gives a smoother line that is slower to react to real change; a narrower one chases noise. 21 days is a judgement call, not a fact.

Why the shaded bands are wider than the pollsters’ margin of error

The margin of error a pollster quotes measures one thing only: sampling error, the luck of who happened to answer. It assumes a perfect random sample of the population.

Real polls are not perfect random samples. The errors that actually dominate are house effects, weighting choices, differential non-response and late swing — none of which the quoted margin captures. So we widen the interval by a factor of 1.7.

That figure is a stated judgement about how much real polling error exceeds theoretical error, not something derived from the data. We would rather show you an honestly wide band than a precise-looking wrong one.

Reading a movement

If two shaded bands overlap, the honest reading is no clear change. Most of the week-to-week poll movement reported as news sits comfortably inside the noise. A single poll showing a party up two points usually means nothing at all.

Political funding data

Funding figures come from the Australian Electoral Commission’s Transparency Register, published on the first working day of February each year. The current data covers 2024-25.

The register contains two things that matter: what each party reported receiving in total, and each individual receipt large enough that its source had to be named. The difference between them is money with no public source — $144.1M, or 29.4% of all party income. That gap is the legal consequence of a disclosure threshold, not evidence of wrongdoing.

Three choices shape those numbers, and each of them cuts:

• Public money is not a donation. Electoral commissions pay parties for each first-preference vote, and revenue agencies make refunds. Listed alongside private donors, the AEC would appear to be the largest single funder of Australian politics. We report public and private sources separately.
• Branches are combined.The disclosure threshold applies per branch, so a donor can give just under it to each of a party’s branches separately and never be named. Combining branches reflects how the money actually works. The trade-off: parties also move money between their own branches, and the register records that identically to an outside donation — so the traceable share is an upper bound.
• Names are consolidated. The register does not standardise names — the same body can appear three different ways. Left alone, one large source is counted as several small ones. We merge known variants, which means our totals can be higher than any single spelling in the register.

Colour, and reading the charts

Charts use authentic party colours, because a wrong-coloured party line is itself an error. But some of those colours are genuinely hard to tell apart with colour-vision deficiency — Greens green and One Nation orange are nearly identical under the most common form of red-green colour blindness.

So colour never carries meaning on its own. Every party has its own marker shape, lines are labelled directly, hovering a party fades the others, and a tested colour-blind-safe palette is one click away above any chart. Every chart also has a table view, and untraceable money is shown as a hatch rather than a colour.

Mistakes and corrections

We will get things wrong. When we do, the correction is made and noted — not quietly overwritten.

Every data change is recorded in version control, so any figure on this site can be traced back to when it changed and why. We also publish the known problems in our own data rather than hiding them: if a pollster’s numbers do not add up, you will see that said plainly on the page rather than silently tidied away.