Found, Not Reported

1A fingerprint for found offences

When police find an offence, the suspect is usually standing in front of them. When a victim reports one, police often start with nothing. Search below for an offence you know, such as theft of a bicycle, to see where it sits.

2The total held still; the mix did not

Leave out the two offences that laws switched on and off (public drunkenness and COVID-19 directions), and police-found offences per head were only 3% lower in 2025–26 than in 2016–17. Inside that steady total, single offence types swung by far more.

Switch the chart to Per 100,000 people and drunkenness, a grey bar, dwarfs every other change; in percentage terms, blue bars such as begging and drug use fell almost as far. Chart 3 shows what grey bars look like up close; Chart 4 turns to the blue ones.

3When the law moves, the count follows

Public drunkenness stopped being an offence on 7 November 2023 (Department of Health, 2023). Two bail offences were repealed on 25 March 2024: contravening a conduct condition of bail (Bail Act s 30A) and committing an indictable offence on bail (s 30B) (Galbally Parker, 2024). Failing to answer bail stayed an offence, and from 2 December 2024 so did a narrower one: committing a Schedule 1 or 2 offence, the Bail Act’s lists of serious offences, while on bail (Crime Statistics Agency, n.d.). Both repealed offences returned from 26 March 2025, one of them in a different Act (Judicial College of Victoria, 2025). Affray was rewritten in 2017 (Premier of Victoria, 2017).

Press Codes combined and affray runs straight through its 2017 rewrite: a new label, not a new level. Bail-condition breaches fell to 933 in 2024–25, a transition year in which no such offence existed until 26 March 2025, then came back at 5,301 in 2025–26, below the 7,100 a year recorded before.

4Same law, different counts

These six offences sit under the same Act and section in every year of the data.

Press Shared scale to put all six on one footing: 1,4-butanediol climbs furthest, cannabis use falls furthest. The law stayed the same; the counts did not. These shifts may reflect changes in behaviour, in detection or in how offences are recorded, and the available data cannot determine the contribution of each. Two details are worth knowing. Police-recorded offences for using cannabis fell from 468 to 20 a year while recorded offences for possessing it barely moved (8,217 to 7,886). And 1,4-butanediol, which Victoria Police describes as slower-acting and more unpredictable than GHB, made up most of the drugs seized in St Kilda over six months to March 2025 (Victoria Police, 2025). Advocates for the begging repeal had already pointed to a fall in the number of people charged with begging (Payne, 2025).

5Where the shifts happened

The map takes three of Chart 4’s shifts down to council level; pick one from the menu above it. Every council has the same laws, so differences between them reflect local behaviour, local policing, or both.

The Melbourne local government area (LGA) fell furthest, from 519 to 228 public-order offences per 100,000 residents a year; its count halved too, from 844 to 439. Precursor offences rose fastest in Latrobe, Frankston and Wangaratta (Crime Statistics Agency, 2026a). The map shows where the shifts happened; it cannot show why.

Seven offences, under the old law

Parliament voted to repeal the offence of begging in September 2026. In 2025–26, police recorded 7 begging offences, 94% fewer than in 2016–17, under a law that had not changed. A crime count can move that far with no new law, and with no evidence in the data that anyone’s behaviour changed. Between now and 28 November, when a crime number is quoted, it is worth asking what kind of offence it counts, and what else might be moving it.

Method and data notes

Data. Two independent sources. (1) Crime Statistics Agency recorded offences data tables, year ending June 2026, released 24 September 2026 and covering July 2016 to June 2026. Statewide offences by offence subgroup with rates (Table 01), by investigation status (Table 04) and by offence code, Act and section (Table 05); offences by Local Government Area (LGA Tables 01–02). (2) ABS ASGS Edition 3 LGA 2025 digital boundaries, for the map only. Both are downloaded directly from the publisher by the R code on first run and are licensed CC BY 4.0. Recorded offences count each offence police record, so one incident can produce several. Table 05 omits offence types that could identify a person.

Groups (an author-defined proxy). Every CSA offence subgroup is assigned to one group. Usually found by police = divisions C (drug), D (public order and security), E (justice procedures) and F (other/regulatory), plus B45 receiving or handling stolen goods and A22 assaulting police or emergency workers, which arise from police contact. Exceptions sent to the reported groups: D31 privacy offences and D32 hoaxes (harm to a person), D43 hacking (harm to property), and E21/E22 breaches of family violence and intervention orders (harm to a person, because a protected person usually reports the breach). Usually reported: harm to a person = the rest of division A; Usually reported: harm to property = the rest of division B. This is a judgement about how police usually learn of an offence, not a field in the data. Chart 1 is a consistency check, not proof of how any offence was discovered: weighted across offence types, 5.8% of police-found offences are unsolved, against 8.6% for harm to a person and 63.7% for harm to property.

Law change test (Chart 2). Grey marks an offence type where at least 20% of offences in either period sit under offence codes that were not used in the other period (an offence created, abolished or re-coded), or where a documented law change applies (drunkenness, bail, affray). Blue does not prove the law was untouched, only that the same offence codes carried the count. Chart 4’s six offences were also checked individually: each appears under the same Act and section in all ten years.

Robustness checks. The grouping was re-run two other ways (table below). The share of offences that police usually find ranges from 14.4% to 30.5%, so “a fifth” describes the main grouping only. Under every version, police-found offences are far less often unsolved than all others, and their total per head (excluding public drunkenness and COVID-19 directions) moved by 10% or less over the decade. Chart 2 marks an offence type grey at a 20% code-change threshold. At 10%, drug trafficking and drug precursors (mostly 1,4-butanediol) also turn grey; at 30%, nothing changes. Drug use, offensive language, begging and trespass and entering without authority stay blue at every threshold. Chart 4 checks its six offences code by code.

Automated checks. The report will not knit unless: every offence subgroup is assigned to exactly one group; each downloaded file is complete and readable; all 79 councils match a boundary; council population denominators are positive and add up to within 1% of the state population (largest gap: 0.01%); each of Chart 4’s offences keeps one Act and section and is recorded in all ten years; and the three groups sum to the CSA’s statewide total. Population denominators are recovered from the CSA’s own published rates, which use ABS resident population, not the number of people present in an area. Built with R 4.3.3, plotly 4.10.4, crosstalk 1.2.1 and sf 1.0.15.

*Excluding public drunkenness and COVID-19 directions, which laws switched on and off.

Design choices. The design follows the course text (Baglin, 2026). Each chart has a short headline title that states its takeaway, with details in a subtitle, and series are labelled directly where possible (Evergreen & Emery, 2016, as cited in Baglin, 2026). The main comparisons use position and length, the most accurately read visual variables, with circle area (Chart 1) and colour (Chart 5) in supporting roles (Cleveland & McGill, as cited in Baglin, 2026). No chart maps more than three variables plus one facet, and none uses a dual axis; Chart 4’s shared scale indexes each offence to 2016–17, the alternative the course text recommends, on a log scale because the changes differ so much in size (Few, 2008; Rost, 2018; both as cited in Baglin, 2026). Small-multiple panels keep one 1.8:1 shape on every screen so slopes are not distorted, and circle area is proportional to the count it shows (Baglin, 2026). Group colours use the Okabe–Ito palette (Okabe & Ito, 2008), which is colourblind-safe; the group colours and every colour used for data lines and markers have at least 3.3:1 contrast with the page, above the WCAG 3:1 minimum for graphics. Two deliberate exceptions are paler: Chart 2’s grey bars (1.4:1, so they recede behind the blue ones, each with a darker 3.7:1 outline) and the map’s lighter colours near zero, whose exact values are in the hover text and data table. Gridlines are faint on purpose. The map uses ColorBrewer’s colourblind-safe RdBu diverging scale (Neuwirth, 2022), centred on zero. The story moves from author-driven annotated charts (Charts 1–4) to a reader-driven map (Chart 5), and each control serves one interaction intent: select (Chart 1’s search), reconfigure (Chart 2’s re-sort, Chart 3’s combined codes), encode (Chart 4’s shared scale, Chart 5’s measure) and details on demand (hover text and data tables) (Segel & Heer, 2010; Yi et al., 2007; both as cited in Baglin, 2026).

Integrity choices. Rates rather than counts wherever populations differ; three-year averages to smooth single-year noise; regulatory offences excluded from Chart 2 because they are small and dominated by one-off operations. Small multiples use separate scales, labelled on each panel, because the offences differ in volume by orders of magnitude. The map’s diverging scale is symmetric about zero with a grey midpoint. Investigation status excludes 2025–26 because recent investigations are incomplete. Explanations for the rises in Chart 4 are left open because the data cannot separate behaviour from police attention. Recorded offences are a full administrative count rather than a sample, so no sampling error is shown; the uncertainty lies in recording and classification (see the robustness checks) and in revisions, since the CSA extracts data 18 days after each period and figures can change between releases (Crime Statistics Agency, 2026b). Colours were checked for colour-vision deficiency (worst pairwise ΔE 9.2 in OKLab), every series is also named in a label or legend, and every chart has a full text description for screen readers and a data table. Chart buttons and menus can be reached with Tab and pressed with Enter or Space, and Chart 1’s search box works by keyboard. Text colours meet WCAG AA contrast (4.9:1 or higher) and all chart text is at least 12px.

Acknowledgements

I acknowledge the use of Claude (Opus 5.5; Anthropic, 2026; https://claude.ai/), a generative AI tool, in preparing this assignment. I provided the assignment brief, checklist and rubric with the prompt “Implement the assignment”, chose the story angle from options it proposed, and asked it to make sure the story was not a clichéd one. I used it to help develop and test the story, locate and check sources, write and debug the R code for data preparation and the interactive plotly charts, and draft and edit the narrative text. I checked the figures against the source data, reviewed and edited the outputs, and take responsibility for the final work. Charts were built in R (R Core Team, 2024) with plotly (Sievert, 2020), crosstalk (Cheng & Sievert, 2023), sf (Pebesma, 2018), readxl and the tidyverse (Wickham et al., 2019). Thanks to the Crime Statistics Agency for publishing offence-level data with the Act and section attached, which made Charts 3 and 4 possible.

References