Diagnosing housing markets and urban problems · Lesson 4

A falsifiable problem tree and conditional recommendation

8 min read

Lesson objectives

  • Frame competing causal hypotheses and a test that may change the ranking of options without confusing output with outcome.
Reading tools

After compiling indicators and mapping production, do not treat every arrow in a problem tree as established fact. Put an observed outcome at the center, such as high housing burden among sampled renters, then competing branches: low income, high rent, too few suitable homes in the needed location, or travel costs that make cheaper homes impractical. For each branch specify an independent observation that would support or weaken it. If rents are stable but the group's income has fallen, a construction-shortage hypothesis cannot explain the entire change. If dwelling counts rise but homes remain remote from jobs, unit output alone does not demonstrate effective affordability. The problem tree organizes inquiry; it is not permission to assert unmeasured causes. Mark contested arrows as provisional in the memo and name the data custodian and request date. The reader can then see where description ends and inference begins.

Classify the question attached to each arrow. 'Did rent rise?' is descriptive and needs a comparable price series. 'Did permit delay reduce completions?' is causal and requires a time or area comparison with attention to project differences and other factors. 'Do residents prefer temporary relief or repairs?' is a preference question that an administrative record cannot answer alone. 'Can the option be funded?' is a capital and operating budget question, not a satisfaction survey. This separation prevents one instrument from pretending to fill several evidence gaps. If rent data span a full year while income data cover one crisis month, disclose the temporal mismatch before interpreting burden changes. If vacancy is estimated from satellite images, do not infer that owners will let those homes. Diverse evidence is functional: each source supports only certain kinds of statement.

Specify a decision test before collecting evidence so the success rule cannot be rewritten after the result appears. In a teaching case, a municipal authority might pilot rental brokerage in a small area if an independent audit finds at least 30 safe, available homes within a journey acceptable to target households, and the full cost per household served is lower than the stated alternative over one year. Thirty is a hypothetical case threshold, not a general standard. Homes must also satisfy safety and accessibility conditions; low cost cannot excuse failure there. If the audit fails, compare temporary relief with new supply or repair depending on why it failed. Explain the threshold: perhaps brokerage has fixed operating costs that become impractical below 30 homes. If those costs are unknown, do not invent a threshold; request the estimate and keep the choice provisional.

Options need not be a simple either-or. Immediate relief for the most exposed households may be combined with a medium-term supply or planning reform, provided the quick measure does not conceal harm. For each housing option show beneficiaries, time to effect, capital and operating cost, possible price effects, spatial and travel effects, and reversibility. Demand support in a rigid market may partly flow to landlords, but the magnitude depends on the market and design and cannot be asserted without testing. Peripheral building can increase units yet weaken job access. Vacancy activation can be quicker but needs owner agreement, habitability and management. These are comparison questions rather than predetermined verdicts. End the matrix with a conditional recommendation and disclose the decision maker's priority, because option ranking may differ when urgent relief, lifetime cost or displacement prevention receives greater weight.

Test sensitivity to one input at a time and then to linked inputs. If brokerage was chosen because it is estimated at 800 currency units per household versus 1,100 for a temporary construction alternative, ask how the ranking changes when fewer owners agree or repair costs rise. Do not change the number served while silently keeping each option's total cost fixed; recalculate. Record the switching point: at what available-unit count or operating cost does the other option become preferable? If the cheaper option fails a non-compensable safety or access condition, do not hide that failure behind a large financial weight. Sensitivity analysis is not a number game. It reveals what the decision actually depends on and which datum is worth collecting first. A decision may remain stable after one value changes; report that honestly instead of fabricating a reversal for drama.

In monitoring, separate implementation outputs from household outcomes. The number of homes entering a brokerage scheme is an output; the number of households securely housed in suitable homes they can afford after six months is closer to the purpose. Many signed agreements may coexist with poor stability if families leave after two months. Set a beneficiary baseline, a desired effect, and a review date, with a way to hear from people the scheme did not reach. If the cost definition or sample boundary changes during monitoring, document the break instead of drawing a false continuous trend. Define a stop signal if safety complaints or displacement rise and a correction route if funding reaches the wrong group. The final memo does not guarantee policy success. It makes the team accountable for what was known, what was assumed, what will be measured and when the decision will be reopened if contrary evidence emerges.

Lesson summary

Frame competing causal hypotheses and a test that may change the ranking of options without confusing output with outcome.