Victoria, September 2026

01

The revised preference flows are worth two-tenths of a point of the three-point two-party move

This wave changes two of the three flows it uses. One Nation moves from 75–25 to the Coalition to 81–19, and others reverse from 55–45 to the Coalition to 58–42 to Labor. The table applies both sets to the same September primary votes.

Both columns use the September primary votes of Coalition 30, Labor 25, One Nation 20, Greens 14 and others 11. The published figure is 52 and the September arithmetic gives 52.22, a difference of rounding. Applying the June flows to the June primary votes gives 54.8 against a published 55, so both waves reproduce from their own stated flows.

The two revisions pull in opposite directions and nearly cancel. One Nation preferences flowing harder to the Coalition are worth 1.20 points to it, and the reversal on others, from a 55–45 split in the Coalition's favour to a 58–42 split in Labor's, costs it 1.43. The net effect of changing the method is a fall of 0.23 of a point. The three-point narrowing from 55 to 52 is therefore in the primary votes, not in the assumptions.

The revision is also a change in the kind of claim being made. From February to August the One Nation and others flows were taken from the 2025 federal election result in Victoria and applied to a state contest. This wave takes them from respondent-allocated preferences in DemosAU's own 2026 Victorian polling, which is a measurement of this electorate rather than a transfer from another contest. The report does not say that the earlier waves would read differently under the new basis, and on these primary votes they would not read differently by much.

02

No party's first preference moved outside sampling error since August

A change between the August and September waves carries about ±5.8 points at 95 per cent, from effective samples of 559 and 572. The largest movement is two points.

Each bar is the change in a party's first preference from August to September 2026, in points. The pale band is the ±5.8 points a change of that size carries.

The three-point two-party move is inside the interval as well. What the wave changes is the order of the three leading parties rather than any party's measured support: Labor's 25 is above One Nation's 20 where June had One Nation above Labor, and neither party moved by an amount this instrument can detect. The five-point gap between them in September carries a within-poll interval of about ±5.3, so the new order is no better established than the old one.

The effective sample is 572 from a raw 1,020, a design effect of 1.78. That is the second-highest in the Victorian series, behind February's 1.78 on a much larger raw sample, and it is why this wave's margin of error is ±4.1 per cent where June's was ±3.8 on a raw sample of almost the same size. A reader comparing September with June is comparing a slightly blunter instrument with a sharper one.

03

Labor's strongest cohort is voters who speak another language at home

This wave adds language to both the weighting and the crosstabs, so it is the first in the series to report the split. DemosAU publishes no cohort sample sizes; if the thirds each hold about 191 of the effective 572, a gap between two of them carries roughly ±8 to ±9 points.

Every row sums to 100 in the report. Cohort figures rest on subsamples and carry wider margins than the ±4.1% statewide topline.

Labor takes 34 per cent of voters who speak another language at home and 22 per cent of those who speak only English, a gap of 12 points. The interval on that gap is about ±9.5, so it is outside subgroup error, and 34 is the highest figure Labor records in any cohort in this wave. The Coalition is almost unchanged across the same split, 28 against 30, and One Nation is six points lower among the non-English-speaking group.

Two other gaps are outside subgroup error. One Nation takes 28 per cent of TAFE-qualified voters and 13 of the university-educated, a gap of 15 against about ±8.0, which is the same division the June wave recorded at 31 and 15. The Coalition takes 38 per cent of outright owners and 18 of renters, a gap of 20 against about ±8.8. The Greens' 23 among voters under 35 against 7 among those over 55 is a 16-point gap against about ±7.0. The Coalition's eight-point spread between university and school education, 34 against 26, sits inside about ±9.2 and should not be read as a difference.

04

Three of the four leaders are rated net negative, and none of them moved

Net personal rating, positive minus negative. The axis is zero; bars run left for a net negative and right for a net positive. This is the first Victorian wave to rate all four party leaders.

Each of the four nets is outside zero, and each of the four movements since August is not. The changes are two points for Carroll and one each for Sandell and Pickering, against intervals of about ±6 on a net rating, and Wilson's fall of three is inside hers as well. The ratings describe a settled set of positions rather than a shift.

Carroll's net of 14 points negative is the best any Labor leader records in this series, and he reached it by replacing one who stood at 39 points negative in June. His positive share of 23 per cent is five points above Allan's June reading and his negative share is 20 points below it, which is the change a new premier brings rather than a change in the government's standing: the Labor primary vote in September, at 25, is four points above June's and inside sampling error of it.

05

September against August and the 2022 election

Percentage points, first preferences. No movement in the final column reaches the ±5.8 points a change between these waves carries.

One Nation has an em dash in the election column because it did not contest the lower house statewide in 2022. The election column is the result as the Victorian poll register records it. DemosAU's report states the two-party swing against Labor since the election as 7.0 points and the September figure as 52, which are consistent with the 45.0 in the election column.