How CivicNZ reads affordability and cost of living official statistics

Non-partisan charter

1. Official or openly licensed primary sources only: Stats NZ, MBIE Tenancy Services, MBIE QSDEP, HUD republishing HES, and MSD Benefit Fact Sheets.
2. REINZ and CoreLogic commercial price files are not ingested.
3. Verdicts are Better, Worse, Flat, or Insufficient. They use the single CivicNZ trend rule (official significance, published CIs, like-for-like recent change, then Theil–Sen). Editors do not set them.
4. Every region, income, and household-type comparison shows the raw gap beside any official adjusted gap.
5. An explained share is shown only when both a raw and an official adjusted gap exist. It is descriptive, not causal.
6. HLPI versus LCI or QES is labelled as association, not causation. The baskets differ (HLPI includes mortgage interest; LCI holds labour quality fixed).
7. Findings carry an evidence-strength rating and a caveats note.
8. CivicNZ does not publish causal claims, cultural generalisations, or contrarian readings the table cannot support.
9. The desk is non-partisan.

How a verdict is derived

• Method: CivicNZ trend rule. Labels: Better, Worse, Flat, Insufficient.
• Threshold: Direction-of-good is taken from the series catalogue. Official 95% significance wording and published confidence intervals come first. Otherwise a like-for-like recent change (same quarter or month a year earlier, or the previous annual point) is the headline, including Flat. Theil–Sen is longer-run context and does not override that headline; it is used only when no same-season pair exists and there are four or more official points spanning at least two years. Series that do not meet the minimum observations or period, or survey shares without published CIs, are Insufficient — never Flat-as-a-stand-in. CivicNZ does not apply a homemade 0.2-unit / 2% Flat band. Verdicts are never hand-set.
• A topic verdict uses published CIs or official significance language on the primary series when the source prints them. Otherwise it is the like-for-like year-on-year change. Theil–Sen is longer-run context and does not override that headline. Companion series and short quarterly ticks are shown, not averaged into the headline. The area chip is the plurality of topic verdicts, and the lead sentence matches that chip.
• Missing values are never imputed. A one-point series, or a seasonal series without the same quarter or month a year earlier, is Insufficient — never Flat-as-a-stand-in.
• Side-by-side official series (HLPI versus LCI, rents versus incomes) are labelled as association, not causation.

Measured confounders

Every group comparison lists these factors. Status is applied, stratified, context only, or not published — never invented.

• Household income / quintile
• Tenure (rent / mortgage / own outright)
• Household type
• Region
• Ethnicity
• Mortgage interest
• Dwelling quality / size
• Hours and employment mix
• Tax and transfers

Public JSON

Summary, topics, findings, sources, methods and gaps are free. This launch does not add a paid desk route.

• /affordability/summary
• /affordability/topics
• /affordability/findings
• /affordability/sources
• /affordability/methods
• /affordability/gaps

What is not sourced here

CivicNZ lists the official extracts it does not have, instead of filling the gap.