Skip to content

Constituency weights produce unrealistic rent estimates for Scotland and Wales #316

@vahid-ahmadi

Description

@vahid-ahmadi

Summary

When computing constituency-level rent statistics using the parliamentary constituency weights (parliamentary_constituency_weights.h5), Scotland and Wales produce significantly underestimated private rent figures compared to ONS/VOA official statistics.

Evidence

Area PE estimate (private rent) ONS approx Gap
Scotland avg £555/mo £950/mo ~40% too low
Edinburgh £530/mo £1,100/mo ~50% too low
Glasgow £445-590/mo £850-950/mo ~50% too low
Wales avg £731/mo £800/mo ~10% too low
Cardiff £750-770/mo £900-1,000/mo ~20% too low
England avg £1,167/mo £1,326/mo (mean) ~12% low
London (central) £1,700/mo £1,750-2,100/mo Close
Manchester £1,030/mo £950-1,050/mo Close

England constituencies — especially London and major cities — produce plausible estimates. Scotland and Wales are systematically too low.

Root cause

The FRS has limited Scottish/Welsh sample: only 626 Scottish private renters and a similarly small Welsh sample across all ~53k households. With 57 Scottish constituencies and 32 Welsh constituencies, the weights cannot differentiate high-rent areas (Edinburgh, Glasgow) from rural Scotland. The weight vectors are not sparse enough — each constituency has ~33k non-zero weights out of 53k households — so the weighted averages regress toward a smoothed national mean.

Key observations:

  • Weight correlation between a London and Scotland constituency is near zero (-0.005), so the weights do differentiate across countries
  • But within Scotland, there isn't enough FRS variation to produce realistic inter-constituency differences
  • Edinburgh North and Leith (£530/mo) vs rural Angus (£480/mo) — almost no difference, when ONS shows a ~£400/mo gap

Impact

We had to remove the constituency-level map from the rent control dashboard and fall back to region-level bars because the Scotland/Wales estimates were not credible.

Possible fixes

  1. Calibrate weights to ONS regional/LA-level rent benchmarks — add rent distribution targets to the reweighting objective so constituency weights reproduce known regional rent gradients
  2. Augment FRS with Scottish Household Survey / National Survey for Wales — more Scottish/Welsh observations would give the weights more to work with
  3. Post-hoc rake — after generating constituency weights, rake the Scottish/Welsh constituencies so their weighted mean rent matches ONS regional figures

🤖 Generated with Claude Code

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions