Draft note

### 1. Data

- Voting districts / polling stations
- Etalab
- Insee / `mapvotr`
- official contours
- Census tracts
- IGN
- socio-demographic data: variable selection
- Electoral results
- lists / candidates selection

### 2. Interpolation

References:

- https://r-spatial.org/book/05-Attributes.html#sec-area-weighted
- https://cloud.r-project.org/web/packages/areal/vignettes/areal-weighted-interpolation.html

Packages (with equivalent code on [Joël's example result](https://github.com/joelgombin/spReapportion/blob/master/CS_ParisPollingStations.rds)):

- `spReapportion`
- https://r-spatial.org/r/2022/04/12/evolution.html
- `sfReapportion`
- `areal` / `st_interpolate_aw`

See also https://github.com/walkerke/umich-workshop-2022/blob/main/acs-data/population_weighted_interpolation.R

### 3. Nonprobabilistic sampling of polling stations for exit polling

- nonprob
- https://www.ceeol.com/search/article-detail?id=1127833
- https://www.taylorfrancis.com/chapters/edit/10.4324/9781315669083-7/high-frequency-polling-non-representative-data-andrew-gelman-sharad-goel-david-rothschild-wei-wang
- distinct PCA + HCPC
- https://www.nature.com/articles/s41598-021-86703-3
- https://journals.sagepub.com/doi/full/10.1177/20597991251369126#con
- [MCA](http://factominer.free.fr/factomethods/multiple-correspondence-analysis.html)
- robustness
- Silhouette scores: https://link.springer.com/article/10.1007/s40808-025-02455-3
- compare to k-means?
- practical considerations

### 4. Example results

1. Avignon
2. Bordeaux
3. Clermont-Ferrand
4. Grenoble -- see #1
5. Lille
6. Lyon
7. Marseille
8. Montpellier
9. Nantes
10. Paris -- see #5
11. Roubaix
12. Rouen