The flood paper compares neighbourhoods just inside an inundation outline with their neighbours just outside it, in the same towns, around six dated floods. This page sets out the analysis variables, the sources, their summary statistics, and the estimator. The variable roles are those of the estimation code: the headline outcome is the log of enquiries per active listing, the treatment is exposure to a flood outline, and enquiries enter as the regression weight.
Variables
Outcome variables (dependent)
- Enquiries per active listing: inbound search for the area’s property divided by the stock of live listings, the cleanest measure of demand because it holds supply fixed.
- Composition-adjusted sold price: the log Land Registry price with the mix of what sells partialled out, tested for capitalisation.
- Outbound destination risk: the flood risk of the places an area’s own residents search, testing whether they look for safer ground.
Independent and control variables (the inputs)
- Treatment: a postcode within 500 metres of a recorded flood outline; the control is a postcode in the same local authorities, outside 500 metres of any flood within six months.
- Continuous flood-risk score: a distance-decay exposure score from 0 to 1, falling with metres from the nearest outline.
- Enquiries: the search count in each cell, which enters as the regression weight.
- Active-listing stock: the number of live listings, the denominator of the demand rate.
- Neighbourhood deprivation: the 2019 English Index of Multiple Deprivation, used for the equity cut.
Data source
Rightmove property-search enquiries, about 294 million from 2010 to 2025, restricted to 2018 onward where enquirer-location coverage exceeds ninety percent; the cleaned Rightmove listings panel for the active-listing stock; Environment Agency Recorded Flood Outlines, from which six major fluvial floods of 2019 to 2024 are drawn; HM Land Registry sold prices; and the 2019 English Index of Multiple Deprivation. England is the canvas.
Summary statistics of the analysis variables
Exposure and inputs:
| Variable | Role | N | Mean | SD | Min | Max |
|---|---|---|---|---|---|---|
| Treated postcodes (within 500 m of a flood) | treatment | 36,603 | — | — | — | — |
| Control postcodes (same towns, unflooded) | control | 368,692 | — | — | — | — |
| Flood-risk score, treated postcodes | input | — | 0.833 | — | — | — |
| Flood-risk score, control postcodes | input | — | 0.412 | — | — | — |
| Enquiries per cell (regression weight) | input | 3,273 | 431 | 747 | 1 | 7,679 |
| Active-listing stock per cell | input | 3,273 | 1,656 | 2,026 | 1 | 12,040 |
| Deprivation (IMD), flooded towns | input | — | 22.41 | — | — | — |
| Deprivation (IMD), control towns | input | — | 21.82 | — | — | — |
Outcomes:
| Variable | Role | N | Mean | SD | Min | Max |
|---|---|---|---|---|---|---|
| Enquiries per active listing | dependent | 3,273 | 0.260 | 0.335 | 0.017 | 11.00 |
| Composition-adjusted log price | dependent | 3,144 | −0.057 | 0.356 | −1.084 | 1.462 |
| Outbound destination flood risk | dependent | 6,243 | 0.032 | 0.059 | 0 | 1 |
Counts pool the six events (treated postcodes range 2,106 to 21,331 per event). The flood-risk score and deprivation are released as treated-versus-control means at the event and town level, not as a property-level distribution, so their standard deviation, minimum, and maximum are not shown. The enquiry and stock statistics are per analysis cell (event by relative-month by local authority).
Method
A stacked event study with a within-town control. Dynamic event-time regressions of the log demand rate are weighted by enquiries and absorb event-time and event-by-authority fixed effects, then collapse to a single difference-in-differences coefficient over the first two months after each flood. Inference uses 133 event-by-authority clusters together with an event-level wild cluster bootstrap, which is deliberately conservative given only six events. Supply enters only as the denominator of the demand rate, never as a conditioned regressor, because listings are a post-flood mediator. Prices and outbound search are tested with the same design.