Correlation vs causation in property research

Rail. A station opens. The suburb median rises over the same years. Mechanism that would count: shorter commute for people within walking distance, so the effect should be concentrated near the stop, not suburb-wide. Confounders: the rate cycle, a rezoning that arrived with the corridor, a mix shift toward houses in the sales set. Until you can separate those, “the station caused the median” is a story.

School. A catchment change, then asking prices rise on one side of a street. Mechanism that would count: a documented catchment line and a buyer set that names that school. Confounders: renovation quality on that side, a flood line on the other, a small sample. A catchment is a fact. A price effect is a separate claim.

Population. ERP rises. Rents rise. Mechanism that would count: more households than dwellings, visible in occupied-dwelling counts and listings. Confounders: household size falling, a new apartment block that lifted ERP through construction workers counted elsewhere, or a rent series that changed publisher. People are not dwellings. See population.

How, specifically, would the cause produce the effect? Who would pay more, for what, and where? If the mechanism predicts a local effect and the statistic is suburb-wide, the statistic is the wrong object. If the mechanism needs households and you only have ERP, the input is the wrong object.

These mistakes skip the confounders.

• Treating a station announcement as an operating station. See catalysts.
• Using an all-dwellings median as the house market next to a unit pipeline.
• Reading “prices rose after X” without asking what rose in the control suburbs.
• Letting a small sample correlate with almost anything.
• Calling a Districts nearby project a price driver.

A mechanism you cannot locate on a map is not a mechanism. It is a mood.

It does not establish that the station, the school or the population change caused the median. Districts can show that a median rose in a period and that a project sits in the same period. It cannot separate the project from the rate cycle, the sales mix or a zoning change. Planning changes are recorded differently in each state and council. An unknown register next door is a gap, not a finding that nothing changed.

Recorded observations (sales, rents, pipeline, overlays, population) are labelled by source and period. “Before you act” groups findings by topic, not by claimed cause. Nearby development is context. See how nearby development can affect a property.

Write the mechanism in one sentence. Write two confounders. Write whether the statistic’s geography matches the mechanism. If you cannot do those three, you have a correlation. Stop there. A planning change recorded in one state’s register and missing from the next council’s feed is a coverage difference. It is not evidence that nothing happened in the second place. Small suburbs make the problem worse: a median from a dozen sales will sit next to almost any other series you pick. That is a sample problem wearing a story about infrastructure.