How we got to 64%
Weighting the council’s own result to the town’s real population — across every dimension they measured, at once.
Only about 4.6% of residents responded to the 2024 consultation, and the people who did skewed older and more affluent than the town as a whole. The standard statistical fix is post-stratification: re-weight the responses so the sample matches the actual population, then recompute support.
Buckinghamshire Council did a limited version of this — weighting by age alone, which moved support from 60% to 64%. We did the fuller, more valid version: weighting to the population across age, sex, ethnicity, disability and deprivation jointly. That matters because the under-represented groups overlap — younger, more diverse and less-affluent residents are substantially the same people — and only a joint weighting handles that correctly without double-counting.
What we used
Two real, public sources — and nothing else:
- Buckinghamshire Council’s published support figures for each age band, sex, ethnic group, disability status and Acorn group (their May 2024 Consultation Analysis Report).
- ONS 2021 Census population counts for the same categories, for all 236 neighbourhoods (Output Areas) of the unparished town — roughly 66,000 adults.
What we did
We built a population model of the town across all five dimensions at once, calibrated the council’s published support rates onto it, and re-weighted to match the real population. As a check, when we weight by age alone our method returns 63.7% — reproducing the council’s own 64% and confirming the approach.
What we found
| Raw headline (council) | 60% |
| Age-weighted (council’s own) | 64% |
| Age-only, our method (check) | 63.7% |
| Joint, all five dimensions | 64.0% |
| 95% confidence interval | 61–67% |
The result is stable: dropping deprivation gives 65%, dropping disability gives 63.7%, and the confidence range never falls below 60%. Weighting by everything confirms the council’s own number.
Being straight about the limits
We believe in showing our working, including where it’s imperfect. The council only published support one dimension at a time, so our model assumes the dimensions combine additively and can’t capture interactions between them; the deprivation measure is a neighbourhood-level proxy; and one alternative assumption about who responded gives a lower figure of 58.5%. The properly rigorous version needs the council’s cross-tabulated data, which we have formally requested. None of this changes the headline: support sits around 64%, and 60% is, if anything, an undercount.
This analysis uses only published council figures and ONS Census data. The full method and code are documented in the campaign’s technical notes and are reproducible.