Australia set a target of cutting female intimate partner homicide 25 percent a year from a 2022-23 baseline — the rate rose 28 percent in that baseline year, and a 2025 report to Parliament found more than 1,000 recommendations from 25 inquiries since 2010 still not consistently implemented

What is happening?

In 2022 the Commonwealth, state and territory governments released the National Plan to End Violence against Women and Children 2022-2032. One of the measures attached to it is a target to cut the number of women killed by an intimate partner by 25 percent each year, measured from a 2022-23 baseline. The plan does not set out how the 25 percent figure was arrived at.

The baseline year moved the other way. The rate of female intimate partner homicide rose from 0.25 per 100,000 women in 2021-22 to 0.32 in 2022-23, an increase of 28 percent, and 34 women were killed by an intimate partner in that year. The same series records the rate for First Nations women at 3.07 per 100,000 against 0.45 for other women in that year, a ratio of about 6.8.

This plan replaced an earlier one. The National Plan to Reduce Violence against Women and Children ran from 2010 to 2022, and in 2019 policy officials working on it acknowledged that no significant reduction in violence against women would be achieved by the time it expired.

In November 2022 the Commonwealth appointed the first Domestic, Family and Sexual Violence Commissioner to track implementation of the new national plan and report to Parliament each year. The role carries no power to compel anything. It observes and it reports.

On 2025-10-30 the second Yearly Report to Parliament was tabled. The Domestic, Family and Sexual Violence Commission had gone through more than 1,000 recommendations made by more than 25 inquiries, royal commissions and coronial inquiries across the Commonwealth, the states and the territories since 2010. It found that many of those recommendations recur from report to report and are not consistently acted on. It stated that the target of a 25 percent annual reduction in female intimate partner homicide is not being met. It described a fragmented system in which jurisdictions pass responsibility between one another, said that activity is being mistaken for progress, and recommended the creation of a national body with the standing to drive implementation. No such body exists, which is why the recommendation was made.

Alongside the tabling, the government stated that First Nations women are 33 times more likely than other Australian women to be hospitalised for family violence and up to 7 times more likely to be killed.

Whose problem is this?

The body that measures whether the national plan is working is not a body that can change anything, and the governments that could change something number nine rather than one. Reporting and authority sit in different places, and nothing in the arrangement joins them.

Where does this problem end?

The boundary here separates the record of what has been recommended from the record of what has been done. The first of those is long and well documented, and the second is the one that no source opened here can supply in any itemised form.

What is the state now, and what should it be?

Now

Should be. The target state is unusually explicit in this case, and that is what makes the record legible. The plan names the number and the direction: female intimate partner homicide falling 25 percent a year against a 2022-23 baseline, inside a plan whose stated purpose is to end violence against women and children, and whose predecessor was publicly described as a commitment to end it within a generation. The gap is not that nobody said what success looks like. It is one layer down. No source opened here states how many of the more than 1,000 recommendations should be implemented, by whom, or by when, and no source names a body that would answer for it. The outcome target exists and the implementation target does not, which is precisely the distance the 2025 report to Parliament measured.

How big is it?

Between 1.7 million and 2.7 million Australian women. The Personal Safety Survey for 2021-22 records that 27 percent of Australian women, about 2.7 million, have experienced violence or abuse from an intimate partner since the age of 15. Inside that figure, 17 percent or about 1.7 million experienced physical or sexual violence from a partner, 23 percent or about 2.3 million experienced emotional abuse, and 16 percent or about 1.6 million experienced economic abuse. The components overlap with one another and cannot be added, so the interval runs from the narrowest partner measure the survey publishes to the broadest.

That interval measures exposure to the underlying harm. It is not a count of the people affected by the failure to implement recommendations, and no source opened here produces one, because a recommendation that is not implemented has no register of the people it would have reached.

The annual harm at the sharp end is a much smaller and separately counted quantity. Thirty-four women were killed by an intimate partner in 2022-23, and that is the number the national target is written against. A count of women who have experienced partner violence and a count of women killed in a single year answer different questions, and this document carries both rather than choosing between them.

For 2025 there is no single figure at all. Counting Dead Women recorded 43 deaths and Australian Femicide Watch recorded 62 women and 24 children, and the two differ because one counts deaths within intimate partner and family relationships and the other counts female homicide more broadly. Both recorded a fall against 2024. Neither is an official series, and no official series for that year opened here.

Under what conditions does it arise?

1. The bodies that recommend cannot implement, and the bodies that implement need not respond. Royal commissions, coronial inquiries and parliamentary inquiries produce findings and then dissolve or move on. Nothing in that arrangement requires a government to say whether it accepted an item, when it will act, or why it declined. More than 1,000 recommendations accumulated over fifteen years without a register that would show which of them are outstanding.

2. Nine governments share the subject and none of them owns it. Most of what a recommendation asks for lands in police, courts, child protection, housing or health, and those are run by the states and territories, while the national plan and its targets belong to the Commonwealth. The report to Parliament describes jurisdictions passing responsibility between one another, which is the predictable shape of a system where every participant can point at a different participant and be partly right.

3. The measuring role was given no authority and the authorities do no equivalent measuring. The commissioner appointed in November 2022 tracks implementation and reports annually to Parliament. That design produces an accurate description of the failure once a year and changes nothing about it, and the remedy the 2025 report proposes is a body that would have the standing the reporting role does not.

4. Activity is easier to report than outcome. Packages, plans, action plans and funding announcements are events with dates on them, and each is straightforwardly reportable as progress. The 2025 report names this directly, saying that activity is being mistaken for progress. A system that reports the things it can count will keep reporting them whether or not the count of deaths moves.

5. The harm itself has no agreed count. Two independent organisations produced 43 and 62 for the same year because they use different definitions, and the official statistical series for the most recent years did not open in this round. When the primary indicator is contested and delayed, the failure to move it is difficult to establish at the moment when establishing it would matter.

6. Money arrives as packages rather than as standing capacity. A 4.7 billion dollar allocation followed the 2024 Rapid Review and the sector then identified a remaining shortfall of 240 million dollars for frontline services, while demand on the national counselling line rose about 3,000 percent over fifteen years. Funding that arrives in response to attention recedes with the attention, and the demand curve does not.

What has been tried?

Two kinds of work have been done here and neither of them is implementation. One kind produces recommendations, and the record of that work is now more than 1,000 items long. The other kind produces money, and it arrives in packages attached to particular moments of attention. What the record since 2010 does not contain is a mechanism that turns the first into action and keeps the second in place after the attention moves.

What was found?

Why is it still unsolved?

Institutional gap — the recommendations exist, the target exists, the measurement exists, and nothing in the arrangement obliges anyone to act on any of the three.

A recommendation is an instruction with no addressee that has to answer. Fifteen years of inquiries produced more than 1,000 of them, and the striking finding of the 2025 report is not that they were rejected but that they recur — the same items reappearing in later inquiries because the earlier ones went nowhere. That pattern is what an accountability vacuum looks like from the inside. Each inquiry is competent, each report is published, and the aggregate effect is a growing archive rather than a changing system. There is no register that says which items are outstanding, no date attached to any of them, and no body that is asked why.

The second part of the gap is that the subject is split across nine governments and the plan belongs to one of them. The Commonwealth writes the national plan, sets the target and funds the national response, while most of what the recommendations ask for is delivered by state and territory police, courts, child protection, housing and health services. Fragmentation of that kind does not merely slow things down. It makes non-implementation unattributable, because for almost any outstanding item there is a defensible account in which somebody else was supposed to do it. The 2025 report describes jurisdictions passing responsibility between one another, and that is the mechanism it is describing.

The third part is that the role built to watch this was built without teeth, deliberately. The commissioner appointed in 2022 tracks implementation and reports to Parliament, and the reporting has worked exactly as designed: the failure is now documented in an official document tabled in Parliament, in numbers, with the target named and assessed as not being met. What follows from that document is nothing in particular. The remedy it proposes is a body that would have the standing it lacks, which means the current arrangement can only diagnose the condition that would have to be treated by an institution that does not yet exist.

The fourth part is what keeps the pattern stable rather than merely bad. Activity is countable and outcome is not, at least not on the timescale attention runs on. Plans, packages and appointments each arrive as an event with a date, and each is reportable as progress on the day it happens. The outcome measure moves slowly, is published late, is contested between organisations using different definitions, and for the most recent two years could not be established here at all. When the reportable thing and the real thing come apart like that, a system can produce a continuous record of effort while the number it set out to move does not move, and neither statement is false.

What observation would mean it is solved?

Candidates — (a) the number of women killed by an intimate partner falls against the 2022-23 baseline for several consecutive years (b) a body with standing to compel implementation across the Commonwealth and the states and territories is legislated, with a published register showing the status of every outstanding recommendation (c) the recurrence rate falls, meaning later inquiries stop repeating recommendations that earlier inquiries already made.

(a) is the target and it is the weakest single test. It counts a small number of events each year, so it moves substantially from ordinary variation, and the 2022-23 baseline was itself an unusually high year, which means an ordinary regression toward the earlier level would register as a 25 percent success without anything having changed. It is also the measure that cannot currently be read at all for the two most recent years, because the primary series did not open here and the values circulating in secondary summaries disagree with one another. A target that cannot be evaluated on schedule is not yet functioning as a target.

(b) is the observation that would show the pattern itself had changed, because it is the thing this record has never contained. Every instrument tried so far has been a plan, a package or an observer. A body that can require an answer would still exist after the attention moved on, which is what distinguishes it from the rest of the list. It is also the least likely to appear, since it would create a form of failure that no participant currently has to record.

(c) is the most direct measure of the actual defect and nobody publishes it. The recurrence of recommendations across inquiries is what the 2025 analysis identified, and a falling recurrence rate would mean earlier findings were being absorbed rather than restated. Establishing it requires the itemised status of the more than 1,000 recommendations, which no source opened here supplies. The three have to be read together, and the second is the one that would say whether anything structural had moved.

What is it connected to?

Fills with researchthe criminalisation of coercive control across the states and territories, family law and child protection interaction, housing and refuge capacity as the constraint on leaving, the separate implementation record of recommendations specific to First Nations women and communities, and the same recommendation-to-implementation gap in other Australian royal commission subjects. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• How many of the more than 1,000 recommendations have been implemented. This is the central missing quantity of the whole document. An itemised table showing which recommendations were acted on, which were partly acted on and which were never touched would convert the finding from a characterisation into a measurement. The commission site refused automated requests, so no such table was read here, and no secondary source opened here reproduces one.
• The homicide rate for 2023-24 and 2024-25. Search summaries carried conflicting values for both years, in one case 0.43 against 0.45 per 100,000 and in another 0.21 against 0.29. The primary statistical report that would settle it did not open. That is a failure to retrieve rather than an absence of data, and it means this document cannot say whether the target has been met in any year since the baseline.
• Whether the recommended national body was accepted. The report proposing it was tabled on 2025-10-30. Whether the government responded, accepted the recommendation, or introduced legislation is not established by any source opened here.
• Why the two independent death counts differ by nineteen. Both are reported in one opened source, which attributes the difference to definitional scope. Neither organisation publishes a reconciliation in anything opened here, and no official count exists against which either could be checked.
• When and how the commissioner role became a commission. Material from 2022 describes an individual office holder and material from 2025 describes an institution. The date of the change and the legal instrument behind it were not established.
• How the multi-year averages relate to the official series. The figure of about 30 women a year across 2017 to 2023 comes from an independent analysis blog restating government statistics. It was not reconciled here against the primary series, and no claim in the state table or the population estimate rests on it.
• What the 25 percent target was derived from. The one opened analysis of the plan notes that no basis is given for the figure. Nothing opened here explains what modelling, comparison or judgement produced it, which matters because it determines whether the target was ever achievable by the means the plan funds.

See the evidence

No Australian government statistical source was read directly. The one primary body of data that opened is the statistical agency media release on partner violence, which is where the population interval comes from and which is the only high-graded quantitative row here. Everything about the homicide series reaches this document through one secondary analysis of it, and the report that is the subject of this document — the second Yearly Report to Parliament, tabled 2025-10-30 — was not read at all, because the commission site refused automated requests. Five rows above are blank with the reason recorded rather than filled with a substitute link, and one of those blanks holds the central missing quantity: how many of the more than 1,000 recommendations have actually been implemented. The government statement on the tabling did open, and it corroborates the tabling date and the First Nations disparity, but it is an account by a party to the matter and it restates figures whose underlying series did not open. Where sources differ the difference is left visible rather than resolved: the First Nations homicide disparity appears as about 6.8 times in the statistical series and as up to 7 times in the government statement, and the 2025 death counts appear as 43 and as 62 from two organisations counting different things, with no official series available for that year to sit between them. Where a source is weak it is marked and quarantined: the multi-year averages from an independent analysis blog are graded low, are not reconciled against the primary series, and carry no claim in the state table, the population estimate or the headline. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 1,700,000–2,700,000 As of 2021-22

Sensitivity The width of this interval is definitional, not statistical. It is the distance between the narrowest and the broadest partner-based measure published by one survey in one year, and neither bound carries a confidence interval in the source opened here. Two things it does not count, in the same direction: it counts only women, so it excludes male victims counted in the same homicide series, and it counts only partner conduct, so it excludes family violence by someone other than a partner and excludes children exposed to it. Two things it does not count, in the other direction: it is a household survey, so people not in private dwellings and people who did not disclose are outside it, and it is a lifetime measure recorded since the age of 15, so it counts people who are no longer in contact with the person and is therefore larger than any count of people currently exposed. The measure is also from 2021-22 and is held here against a status check date of 2026-08-15. Most importantly, this interval measures exposure to the underlying harm and is not a count of the population affected by the implementation gap this document describes. No source opened here produces that second count, because a recommendation that was never implemented has no register of the people it would have reached. The quantity the national target is actually written against is far smaller and separately counted: 34 women killed by an intimate partner in 2022-23, a rate of 0.32 per 100,000 women against 0.25 the year before. The disparity that runs through the whole subject is not visible in this interval either, since the survey figures opened here are not broken down for First Nations women, while the homicide series records 3.07 per 100,000 against 0.45 for other women in 2022-23 and the government statement of October 2025 puts family violence hospitalisation at 33 times the rate of other Australian women.

Regional breakdown No source opened here publishes a state or territory split of the survey estimate, and the homicide series opened here is reported nationally as well. Splitting a national survey estimate by state population would be proportional allocation, which is prohibited, and it would be particularly wrong in this case: the subject of this document is the fragmentation of responsibility across the Commonwealth and the eight states and territories, so a proportional split would fabricate exactly the dimension the document is about, and would do so in a shape that carries no information from the jurisdictions themselves.