Segmenting housing need and comparing indicators
8 min read
Lesson objectives
- Produce an indicator panel that compares groups using consistent units, periods and definitions while stating uncertainty.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
One headline percentage cannot describe housing need because households differ by tenure, income, size, location and dwelling condition. Begin with the decision the indicator panel must support: urgent relief, supply planning over three years, or prioritization of repairs. Each calls for a different observation unit. Household and eligibility data matter for relief; dwellings and expected flows matter for supply; building and service conditions matter for repairs. The OECD Affordable Housing Database distinguishes market, condition and affordability, and policy indicators. The World Bank housing market toolkit treats urban boundaries and indicator sources as central to diagnosis. Select indicators that answer the decision question and show what they omit. A count of households is not a count of dwellings, and a count of dwellings is not accessible capacity after location, condition and availability constraints.
Build a short data dictionary before drawing a chart. For every column record the indicator, numerator, denominator, unit, geographic scope, capture date, data source, missingness and comparability status. 'Cost-burdened households' requires explicit definitions of cost, income, threshold and period. 'Inadequate dwellings' requires a stated classification rule and inspection status. 'New units' may refer to permits, starts or completions; these are stages, not synonyms. Display a denominator beside every percentage. Thirty percent of 20 households has a different interpretation from 30% of 2,000, particularly if selection differs. For sampled indicators, show the number of respondents in each group. Do not compare a subgroup of five with one of hundreds and describe the difference as stable. When two definitions cannot be aligned, mark them 'not directly comparable' instead of joining them into an attractive but misleading time series.
Select segments that reveal inequity without compromising privacy. Start with renters and owners, then income bands, household size and location. If disability or health data are lawfully available and safely aggregated, use only what is needed to assess accessibility; do not publish tiny cells that identify people. A valid comparison first holds definitions and periods constant, then asks whether a difference may reflect selection or missing data. If 12 of 40 renter households and six of 60 owner households are burdened, the sample rates are 30% and 10%. The difference does not prove tenure caused the burden. Income, building age and neighborhood may differ too. The advisory note reports the difference, proposes checks on competing explanations, and asks whether the urgent problem affects a group that can be helped within the authority's mandate and budget.
Separate a quantitative deficit from a quality deficit. Total dwellings may increase while demand for safe homes near jobs still grows. An official waiting list can shrink after a registration rule changes even if conditions do not improve. Track the list's definition, opening and closure, and duplicate records, then seek evidence about excluded households where feasible. Conversely, sound existing stock should not be ignored merely because some households cannot reach it due to cost, information, discrimination or tenancy requirements. None of these mechanisms may be asserted as the cause in a real city without evidence. The lesson is to map several gaps: quantity, affordability, adequacy and location. Each needs a different indicator, test and possible intervention. Classify evidence inside the panel as directly checked, estimated under an assumption, or unknown pending collection. This classification guards against false certainty in an advisory decision.
Choose the level of analysis for the decision, not for the convenience of downloadable data. A city average may hide a remote district with costly journeys to work, while a street-level table may expose households and overwhelm the decision maker. Start with intelligible service areas and test how rankings change if their boundaries move. If reallocating ten households between zones reverses the priority order, the recommendation is fragile and that sensitivity should be disclosed. Add travel time and service conditions when comparing infill with peripheral construction, but do not add minutes of travel to a monetary burden figure as though they shared units. Put them in a matrix as separate criteria and explain the weights that require decision-maker agreement. Nor should an older indicator be compared with a later intervention and called proof of success. Specify a consistent baseline and follow-up interval before attributing change to policy.
Deliver a panel fit for a decision: one row per group, with columns for group size, rate and definition, evidence quality, and the actionable gap. Below it write two sentences: what first decision question can these data answer, and what decision remains blocked by missing data? The calculations 12÷40 and 6÷60 permit a statement about higher burden in the two observed renter and owner samples respectively, but neither citywide generalization nor a causal claim about tenure. A few interviews, a better sample, or an income-data audit may be warranted before broad assistance. Record the change that would reverse priority: if most burdened renters occupy sound dwellings whose prices are high, the intervention differs from a case of unsafe homes needing urgent improvement. The indicator becomes a condition for policy choice rather than decoration in a report.
Lesson summary
Produce an indicator panel that compares groups using consistent units, periods and definitions while stating uncertainty.