Research methods

Some people answer an optional survey after they've built their comparison. This page explains what we do with those answers — for anyone citing the figures on What the numbers say, or checking how the research is run. Everything here is aggregate and descriptive; it is never a prediction, and it is not a representative poll.

The survey is run as a confirmatory repeated cross-sectional trend series — the design official statistics use — not an open-ended data archive. Under the aggregate-counters model (ADR-0007) the research store holds only the counter tables listed below: the raw answers never leave your device, and the server only ever increments totals. Nothing else is collected, so nothing else can ever be analysed.

What it can and can't measure

The estimands are a closed, pre-registered list — each maps one-to-one to a counter table, and new ones can only be added going forward, never computed over past responses. In plain terms, we can count:

• how many people took part, by election, collection period and state;
• top-party match against one characteristic at a time (say, age, or education) — never two at once;
• support for each proposition (agree / neutral / disagree);
• whether people who match a party tend to agree with its positions;
• a weighting frame (age × gender × state, with no opinion attached) so the whole series can be weighted against ABS and AEC benchmarks;
• response coverage by electorate (a count only, with no characteristics attached);
• which consent notice was in force, and when responses arrived (to the calendar quarter).

Because there is no per-person record at any point, there is nothing to cross-tabulate beyond these totals — that is a deliberate design property, not a limitation to be worked around. The full registry and the question codebook are published: analysis plan · codebook.

Disclosure controls

The rule that governs every stored total:

No counter may key an opinion together with a sensitive attribute and a geography finer than national.

Party match and proposition stances are political opinion. The 12 sensitive characteristics (the Privacy Act's sensitive-information categories — union membership, birthplace, language, Indigenous status, religion, sexual orientation and the political-opinion items) are therefore stored nationally only, never against a state or electorate. The nine non-sensitive characteristics (age, gender, education, work, household size, children, tenure, income, financial security) may be keyed to state. This caps the meaning of the worst case — a single small total — at "one person somewhere in Australia".

Every time-varying total is also keyed by collection cohort (pre-declaration, live campaign, post-election, historical), classified against the AEC timetable, so cohorts are suppressed independently and never silently combined. On top of that, publication applies a minimum cell size (at least 10 responses per cohort), a 50-response board minimum, a rule that shown percentages sum only over surviving cells, and a differencing check when figures are regenerated.

Keeping the series honest

• Scoring is fixed per wave. Each election is scored with that election's engine and dataset; historical waves are never re-scored. A method change is dual-run for a wave and published as a bridged, marked series break.
• The instrument is stable. The question set and its ABS/AES-aligned buckets are held steady across waves; an option-set change creates a new codebook version and is documented as a series break.
• New estimands are additive and forward-only. Adding a total is a governed change to the plan, reviewed against the disclosure rule, and applies only from the next collection.
• Every election is reviewed. After each federal election we confirm the estimands are still needed, re-run the disclosure review over anything published, and reassess re-identification risk.

Retention

The counters are genuinely aggregated statistics that don't relate to an identifiable person, so they are retained indefinitely — there is no fixed deletion clock. A clock never mitigated the residual risk anyway: a total that is one after five years is one because only one person ever matched that combination. That risk is handled by the key rule, publication suppression and access control. Deletion is therefore purpose-based: a wave's counters are deleted when an after-election review finds its purpose has ended, or the privacy impact assessment directs it.

Ethical by construction

Taking part is voluntary and never a condition of using the tool: the survey is offered only after your comparison and plan already exist, and the "skip and build my voting plan" action is always there, equally weighted, going straight to your card. Consent is express and opt-in (unticked by default), you must be 18 or over, collection is minimised to the aggregate totals above with no direct identifiers, and a "prefer not to say" option is present on every question — including all 12 sensitive ones. See the privacy policy for how the answers are handled and why a contribution can't later be withdrawn once it's merged into the totals.

Standards we run against

The research programme is run against these external standards. The current consent notice is version 2026-07.2.

• ABS Data Quality Framework (Australian Bureau of Statistics) — In place.
• ABS confidentiality and statistical disclosure control (Australian Bureau of Statistics) — In place.
• National Statement on Ethical Conduct in Human Research 2025 (NHMRC, ARC and Universities Australia) — In place — awaiting an external determination.
  Outstanding: Written ethics determination (HREC approval, or documented reasons the research is exempt / not requiring review) — restricted legal file.
• AIATSIS Code of Ethics for Aboriginal and Torres Strait Islander Research (Australian Institute of Aboriginal and Torres Strait Islander Studies) — In place — awaiting an external determination, because the survey collects Aboriginal and Torres Strait Islander status.
  Outstanding: Indigenous data-governance decision covering collection and analysis of Aboriginal and Torres Strait Islander status — restricted legal file.

The machine-readable register that binds each standard to the control and tests that implement it is published in full: standards register.