Before you quote that statistic about Australian boys
A new University of Melbourne report makes five recommendations to ASIO, the AFP and the eSafety Commissioner. I read it twice and checked it against itself.
————————
In June, the University of Melbourne published Misogyny, Online Radicalisation, and Youth Extremism in Australia, funded by the Office of National Intelligence. You have probably seen the numbers even if you haven't seen the report: roughly a third of Australian teenage boys holding misogynistic attitudes, more than a quarter agreeing with right-wing extremist statements, 23% of young men saying feminism should be resisted with violence.
Those figures moved fast. They ran in the Herald Sun and the Advertiser in January and on Triple J's Hack in March, months before the full report appeared.
This is not a think piece. It makes five recommendations, and two of them are consequential: amend Australia's national definition of violent extremism to include gender-based violent extremism, and build screening tools for gender attitudes into the counter-terrorism risk assessments used by police, CVE practitioners and social workers.
That is a high bar. When research asks a government to change the operating definition of violent extremism and to start screening people against it, the analysis behind it has to be checkable by someone outside the project. So I checked it, using nothing but the report itself.
Three things I found
The "right-wing extremism" number measures something else.
The report defines six extremist-attitude scales on page 17. "Right-Wing VE" is not one of them. But it appears in every results table, and it is the source of the executive summary's claim that over 26% of teenage boys agree with right-wing extremist attitudes.
Follow it through the document and it turns out to be the Economic Reactionary scale, renamed. You can confirm that two ways: the regression output on page 43 labels the identical coefficients "Economic Reactionary VE", and the narrative on page 22 quotes "economic reactionary (ρ = .53)" for the value tabled as Right-wing VE.
Its two items are about defending private property with force, and violent resistance to a tax system that supports "freeloaders". Then page 18 records that one of those two items was dropped because the scale did not hold together. One item remains.
That number is in the executive summary. It went to the papers.
The main regression figure disagrees with the table that summarises it.
The report's headline conclusion rests on nine regression models, shown as a coefficient matrix on page 43 and summarised in a table on page 44. Five match. Four do not.
The clearest case: for religious violent extremism, the figure gives misogyny a coefficient of −0.053. Negative. The table on the facing page gives +0.049. Positive, and significant. Three further coefficients that are not statistically significant in the figure appear at p < .001 in the table.
A reader has no way to tell which set of numbers is the analysis.
Age and sex are not in the model.
The figure's own caption says the models control for "politico-religiosity and online behaviours". Not age. Not sex.
The entire descriptive half of the report is about age and sex — teenage boys score highest on very nearly every measure in it. Leaving both out of the model means it cannot distinguish "attitudes permissive of violence against women predict extremism" from "young males score high on both sets of questions".
Why this matters more than a footnote
Here is the part that should concern anyone who works in risk.
Recommendation 2 proposes validated screening tools for gender-based attitudes, for use in counter-extremism risk assessment. The report's own thresholds classify between 26% and 35% of adolescent boys as positive. The report also tells us, on page 5, that Australia has roughly 200 active participants in countering violent extremism programs nationally, two-thirds of them aged 10 to 24.
Run that through. A screen that returns positive on a quarter of a demographic cohort, deployed to identify a population of a couple of hundred, has a positive predictive value indistinguishable from zero. For every genuine case you would generate an extraordinary number of false positives — each one a child, brought into contact with a security process, on the strength of a survey answer.
The report gives no sensitivity, no specificity, no positive predictive value and no false-positive cost anywhere in its 51 pages. That is not a statistical nicety. For anyone who would have to operate the instrument, it is the whole question.
What I am not saying
I am not saying the problem is not real. It is, and the evidence for it does not depend on this report.
I am also not saying the report is worthless. Getting attitudinal data from 1,122 Australians aged 13 to 17 is difficult and valuable, and the matched adult comparison is a real contribution. Publishing the actual survey wording is a good-faith act — it is the reason I could write any of this. The teacher survey contains the single most persuasive observation in the document: boys who comply with instructions from male staff and refuse identical instructions from female staff.
And the report's most defensible finding is one it undersells. General social media use shows almost no association with extremist attitudes. What matters is the content, not the screen time. That is a genuinely useful correction to how this debate is usually conducted, and it is the finding most likely to survive scrutiny, precisely because it is a null result.
The real risk
The report records that around 65% of public comment on its preliminary findings was hostile, and that commenters specifically attacked the methodology.
It then published a relabelled scale behind a headline number, an asterisk with no footnote, and a figure that contradicts its own summary table.
That hands the critics something legitimate. And the predictable outcome is not that sceptics are persuaded. It is that the general reader concludes the whole area is contested and stops paying attention. Where the underlying problem is real, that is the worst result available.
None of it is unfixable. Publish the dataset and the model specifications. Re-run the nine models with age and sex as controls. Report the reliability statistics and a confirmatory factor analysis. Restore the scale names. Put confidence intervals on the prevalence figures. And run a positive predictive value analysis before anyone builds a screening instrument.
Until then, the two security recommendations should wait for the appendix.
————————
The Full Appraisal
Scope, six verified internal discrepancies, eight design limitations, a risk assessment of the five recommendations, and a SWOT — is here: Does the evidence carry the recommendations?
Source
Meger, S. (2026). Misogyny, Online Radicalisation, and Youth Extremism in Australia. Research Report. The University of Melbourne, June 2026. All page references are to the 51-page web edition. You can find the full original report here.
The Biggest Risk
The remaining risk is not only in the report. It is in how quickly a contested survey becomes a public fact. On 25 August the ABC transmitted the findings as the “biggest red flag” and the “clearest warning sign” of violent extremism, and as a warrant to treat misogyny and anti-feminism as a national-security category. That is a diagnostic claim. The underlying study is a single wave of self-reported attitudes. Two problems in the coverage itself should have stopped that leap.
First, the ABC quoted an “extraordinary” result: that misogyny, attitudes to violence against women, and the belief that feminism has gone too far account for “approximately 70 to 80 per cent of all the explained variants” in the extremism answers. The report’s own models give R² figures of 34 to 59 per cent. One of those extremism scales includes the item “Feminism is damaging to our society and should be resisted with violence if necessary.” A large overlap between the question and the answer is not a discovery about national security.
Second, the same article listed “far-right” as one of the six extremist movements measured. The report never defined a right-wing violent-extremism scale. It defined an “Economic Reactionary” scale about defending private property and resisting taxes said to support “freeloaders,” then reported that scale in public tables as right-wing extremism. That label is what a general reader hears.
When those two points are not carried with the headline, the audience receives a screening rule and a political category that the published analysis does not support. Hostile gender attitudes are a real problem. Treating an unvalidated attitude score as a red flag, and then putting that flag on national television, is a different problem. It is a societal risk, and it is one the press can still correct.
You can find the ABC article by Georgie Hewson here.
Comments