Hmw II: SVD methods and Elections Data

Published

March 30, 2026

Important

Due : May 18 2026
Work in pairs
Deliver your work through a github repository
Present your work (15 minutes) on 2026-06-03

This homework is about

Using Matrix Factorization methods in Data analysis
Investigating voting patterns in Paris (or elsewhere)

I. Voting Data

Voting data in Paris per polling station can be obtained from a variety of websites.

Note

Many datasets are available in several formats. When possible, use parquet. parquet files can be uploaded using package arrow.

Data concerning polling stations can also be gathered from various sources.

You first task will be to design an extraction pipeline to obtain the voting data you will analyse. You will gather data corresponding to different types of elections (Municipales, Régionales, Législatives, Européennes, Présidentielles) that took place since Year 2000.

title	rounds	year
European P	1	2004
Local	2	2008
President	2	2012
Parliament	2	2017
Local	2	2020

II. Data cleaning

Some data cleaning may be necessary, for example

Some parties changed their names during the last 25 years. Defining a mapping can facilitate the comparison of results from different elections
Some parties were born during the last 25 years
Check that the names of bulletins nuls, bulletins blancs, … are consistent across the different datasets.

Note

Design a cleaning pipeline. Save the cleaned data.

III. Applying Matrix Factorization Methods (SVD)

For one election round, the outcome is summarized by a tibble where rows (individuals) are polling stations and variables/columns are the number of votes obtained by the different condidates/parties.

Perform PCA on different elections. Visualize and describe the result (attention, this is data analysis, not political science).

Choropleths are welcome

Perform CCA to compare different elections.

Feel free to combine different methods.

IV. Clustering

Perform clustering of polling stations

Compare the results of different clustering methods (kmeans with different initializations, hierarchical clustering with different distances and different merging methods).

Compare clustering

V. Regression

Consider first round of Parliementary Elections in 2024 as the reponse variables and Elections between 2017 and European Elections 2024 as explanatory variables.

Fit linear regressions. Discuss the results.

References

Grading criteria
Criterion	Points	Details
Documentation/Report	45%	English/French
Presentation	40%
Data gathering/cleaning pipelines	15%