Methodology / FAQ
Frequently Asked Questions
Answers to common questions about how the Transport Score works, written for a general and skeptical reader. Every number traces to data, not opinion.
Why does my suburb have a low score when there is a train station nearby?
A suburb's score is computed by dividing the entire suburb into 250-metre grid squares and weighting the result by where homes actually exist. A single train station provides excellent service for the people living right next to it, but if 90% of the suburb's residents live too far away to walk to the station, the suburb's overall score will reflect that lack of access. The suburb page's verdict will tell you which story applies — check the "reach" figure to see what share of residents are within walking distance of a viable route.
Where does the data come from?
All timetables, route alignments, and stop locations come from the official Public Transport Victoria (PTV) GTFS schedule data on Data Vic. Suburb boundaries are from the authoritative Vicmap Admin locality dataset. Population weighting is from the ABS Estimated Resident Population (2024-25), distributed by current G-NAF address data. See the full data inputs table for every source, vintage, and caveat.
Does the score account for late, cancelled, or "ghost" services?
No. The score measures the promise of the published timetable, not what is actually delivered. This is deliberate: a cancelled service is worse than one that runs as scheduled, so the score is always at least as generous as reality. Real-time punctuality tracking is a separately planned feature, pending data access.
Why isn't my suburb listed in the league table?
A suburb must have at least 3 active, scored stops inside its official Vicmap boundary to qualify for the league table. This prevents a single flag stop from putting a suburb on the worst-20 list. Additionally, the grid-based scoring is scoped to Greater Melbourne and the surrounding commuter corridor — suburbs outside this scope still get a score page using the stop-mean fallback but are not ranked in the main league table.
How is the proposed "Fix" (the Network Plan) designed?
We run a computer model that identifies areas where residents live too far from high-quality transit. It proposes feeder bus routes along existing road corridors to connect those residents to existing high-frequency train, tram, or bus hubs. The model uses a greedy maximal-covering heuristic: each round, it picks the highest-value remaining corridor (most newly-covered residents per dollar), repeating until coverage targets are met or no further candidates add coverage. The assumptions (10/15 minute headways, $6.50/km operating cost, 4am-midnight span) are stated policy choices, not data findings — see the plan for details.
Adversarial Q&A
For the skeptical reader
These are the questions we would ask about our own numbers if we were reading them for the first time. They are published here because an answer we write ourselves is more honest than one we are forced to write under hostile questioning.
You designed this to make Labor seats look bad.
The score is blind to electorate boundaries. Suburbs are attributed by their real Vicmap locality boundary, computed from government timetable data, weighted by ABS population estimates. Neither the scoring code nor the build pipeline reads an electorate map or political affiliation. The robustness harness (see methodology) sweeps every major constant and publishes which suburbs move band — if the score were rigged, the adversarial perturbations would expose it. The master principle — every ambiguous choice resolves in the network's favour — means the score is if anything biased toward the suburbs that score worst. We invite anyone to reproduce our numbers from the published formulas and data.
The 28-minute constant is arbitrary.
The headway curve has exactly two tunable parameters (half-saturation of 28 minutes and exponent of 2.2), replacing what used to be eight independently hand-picked breakpoints. The robustness harness sweeps both parameters (22, 25, 31, 34 minutes for the half-saturation; 1.8, 2.6 for the exponent) and publishes which suburbs' bands survive the sweep unchanged. Suburbs whose band is stable under all perturbations are identified honestly on the suburb detail page; suburbs near a parameter boundary show as methodology-dependent. This is more transparent than a step-table approach where the choice of breakpoints is never publicly tested.
Crow-flies walking overstates access.
Yes — deliberately, and in the network's favour. A freeway, rail corridor, or river can only make a real walking catchment smaller than the 800m circle, never larger. This is the first of the three choices that carry the master generosity principle (see Overview). If we used real walking paths, low scores would be even lower — and since the point of a low score is to identify places that need investment, the overstatement does not change the policy conclusion.
Why should I trust a political party's numbers?
Every input is a government-published dataset: PTV GTFS timetables, ABS population estimates, Vicmap locality boundaries, G-NAF address registers, and Census mesh block counts. Every formula is published on this site — not described in prose, but reproduced as actual equations with named constants. The robustness harness publishes the effect of every perturbation, including adversarial ones designed to break the result. The changelog records every change with its quantified effect on real suburbs, and entries state what did not move alongside what did. If you find a number on this site that you cannot reproduce from the published formulas and data, that is a bug we will fix.
Your plan costs are made up.
The plan's cost assumptions ($6.50/km operating cost, 10/15-minute headways, 4am-midnight span) are declared policy proposals — the network Fusion would fund if elected — not measurements of existing spending. That distinction is stated on every plan page and in the methodology. The $6.50/km figure is sourced and dated, linked to a transit-planning analysis of Victorian state budget papers; it refreshes on the same cycle as every other price anchor on the site. A plan with stated, defensible, conservative cost assumptions is more useful than one that hides its budget behind vague language, and the plan page reports honestly when it falls short of its own targets rather than inflating its reach.