How can we predict the One Nation vote at the upcoming Victorian election?

As shown above, plotting the cleaned actual 2026 ONP first preference against the cleaned notional 2025 ONP + Family First + Trumpet of Patriots first preferences sum – and removing five outlier seats where they drastically under- or over-performed – provides us with a clear relationship, with an R² of 0.89 with an average residual of 2.59 points and 22 seats within +/- 2% of the regression equation.

This relationship would require tweaking. While its currently quite likely that the ONP will poll similarly in Victoria to their South Australian result based on polling, the intercept and the slope of this relationship could vary, and it may be far more viable to make clearer distinctions between regions, candidate types, etc, but as a starting point this is a fairly logical relationship that captures right-wing and third-party supporting sentiment. The combined sum of the three parties does indeed indicate a stronger relationship than plotting the actual results against 2025 ONP first preference votes alone.

Secondly, taking a look at the most recent demographic data in South Australia sourced from the Australian Bureau of Statistics, we can observe the following relationship.

I’ve chosen to model this relationship with a logarithmic equation to capture the diminishing returns of increased education, a choice that essentially strengthens the support for ONP among electorates with low levels of university education. I’d argue the data reflects this on both a state and federal level.

This relationship is thought-provoking, with an R² of 0.89 excluding a single seat, and given that the demographic data was captured in the same year for all states and uses entirely consistent methodology, I greatly appreciate the strength of the trend. Clearly useful to model urban and rural differences.

I will note that these predictors – cleaned 2025 right-wing votes and higher education percentage – are correlated themselves with an R² of 0.77, so they’re better represented as two representations of the same data-generating process rather than separate relationship – arguing so would be inaccurate.

Its also worth noting the following trend – not sure how useful this would be, but Its certainly funny. Unsurprisingly, the GRN vote is also strongly correlated with the prior two predictors.

Some poor indicators include log distance from CBD:

The proportion of the population born overseas:

And age, which is so shockingly uncorrelated that I must admit that I’m surprised a positive trend wasn’t at-least somewhat present:

So what can we do with this information? First, lets take a look at Victoria to see if we can determine if the South Australian trend holds as strongly within the state we're looking for forecast. Demographic data pre-dates the 2025 federal election, so we can observe how well the education predictor informed 2025 ONP support within the federal boundaries.

As you can see the relationship is not as clear-cut, but realistically, I'd still have a great deal of faith using these statistics as the groundwork for an ONP vote distribution model. Fundamentally, it's unlikely you'd be able to derive a more accurate trend than the two we've identified, particularly considering that fluctuations around the trendline are inevitable due to candidate selection, other parties in the field, controversies, and demographic / electoral changes since the data was recorded.

Significantly, at the Federal election ONP was polling far lower than they did at the South Australian election, so it's really a matter of a ~2.5 average residual compared to a ~3.5 average residual, which isn't as drastic of a difference.

So, taking our education model and our right-wing support model, where do they agree and where do they disagree most in forecasting Victorian ONP support in each seat?

These results are actually very promising considering they're completely unfiltered. They take no account of independent candidates at the 2025 or 2026 elections, they make no regard for controversy or candidate favourability, and yet they still have a high level of agreement. I'll also preface that I've conducted this test attempting to align the means and the standard deviations between the two data sets, and I'm assuming a slight slump in ONP support on the day of the election (not that it matters here what the mean is).

We can see off the bat that 19/88 or 22% of the seats agree within plus or minus 1%, which, even considering that the two metrics are correlated themselves, is interesting. A total of 44% agree within plus or minus 2%. The biggest differences can in-part be explained-away by the presence of anomalies at the last federal election, such as independent candidates in Wannon, Indi, Calwell, and Monash reducing right-wing party support in SW-Coast, Benambra, Bass, Narracan, and more.

Other differences with a correlated direction of bias seem to lie in the northern fringes of Melbourne, where there is apparently a slight disconnect between the education data and right-wing support in Mill Park and Thomastown, as well as in the east with Packenham, Berwick, and Narre Warren North, all of which have a stronger right-wing support than their education levels would suggest.

All in all, these are the estimates that I'd be willing to put some weight on in the lead-up to November's election. After making seat-by-seat adjustments to account for differences in both the 2025 results and the 2026 candidate field, I expect that these figures will be within the realm of reasonable. I'd also want to make state-wide adjustments for where ONP is likely to have gained support most since last year, after all, this is the largest flaw with the above models.