Are Geneva's rivers cleaner than they were?

Find your issue, and check it is measured

An environmental issue: something happening, to something, because of something, with proof that somebody measures it.

Sewage reaches the rivers that drain into Lake Geneva. It carries faecal bacteria into the Aire, the Arve, the Versoix and nearby rivers. Geneva has spent thirty years trying to stop it.

Proof it is measured

E. coli in Geneva's rivers and at the lake shore, opened: the canton's own monitoring, 184 stations, 1995 to 2026.

The system

Input: sewage from overflowing combined sewers and treatment works, and runoff from farmland. Store: the river. Output: flow into the lake. Rain is the trigger: a storm overloads the sewers and the bacteria arrive with it.

Write your research question

One sentence naming the indicator and its units, the years and the places. No strategy in it.

A comparison of mean E. coli (CFU/ml) at the same 48 Geneva river stations in 1995 to 2000 and in 2019 to 2024.

Independent variable

Period: 1995 to 2000, or 2019 to 2024.

Dependent variable

Mean E. coli, in colony-forming units per millilitre.

Not obvious

The lake is clean, but that does not mean every river improved. Station by station, it could go either way.

Find a real-world strategy and its tensions

One real-world strategy people are genuinely arguing about, two named positions, and what the tension does.

Geneva's river restoration programme, and its first large project: the Aire.

The strategy

In April 1997 the canton added river restoration (renaturation, the word to search for in Geneva's documents) to its water law. Restoration means taking rivers out of concrete channels and giving them room to move. A fund of about 11.8 million francs a year pays for it. The Aire was the programme's first large project: bill PL 8490, debated in Geneva's parliament, the Grand Conseil, on 30 August 2001.

Position 1 · environmental, political

The canton, in the bill's report: stopping treatment works from discharging into the river, and restoring it, meant "la qualité des eaux va donc en principe sensiblement s'améliorer": water quality should improve substantially.

Position 2 · economic, cultural

Marc-André Chevalley and Pierre Dupraz, speaking for the farmers in the same debate: the project was far too big ("sur-dimensionnée, relevant même de la science-fiction"). And the order was wrong: "améliorer d'abord la qualité de l'eau", improve the water first, then free the river.

The tension

Land for the river against land for farming. Giving the Aire room to move takes fields, some of them the canton's protected best arable land.

What it does

Uneven costs: the whole river benefits, but only a few farmers lose land.

For the environment

Both sides agreed the water had to get cleaner. They disagreed about whether room for the river is how you get there.

Link to the question

The farmers asked for cleaner water first. The question measures whether Geneva's rivers got cleaner over the years the programme ran.

Select and document your data

A dataset you can defend: counted, documented, and cut to a rule fixed before you looked.

Source

Canton de Genève, LCE_HYB_TM_BACT on the SITG catalogue (Geneva's official map and data service): a CSV inside a zip, extracted 23 August 2026. Read the metadata PDF in the zip's DOC folder.

One row

One station in one year: its mean E. coli, and how many sampling campaigns (visits) that mean came from.

The first thing to fix

Every row appears about fifty times. Remove duplicates on station code plus year, or every count and every test is wrong.

Inclusion rule

Keep years with 10 or more campaigns (fewer visits make a mean you cannot compare). Keep river stations with at least one such year in both periods. That drops the one lake site, Pâquis.

Derived quantity

For each station, the average of its annual means in each period: two numbers per station.

Credit

Canton de Genève, SITG open data, with the extraction date.

Treat your data

Run a statistical test, leading to figures you built and a statistical result.

Mann-Whitney U test, two-tailed, at p = 0.05.

Why this test

A difference between two periods, and E. coli counts are skewed, so a rank test rather than a t-test. Mann-Whitney U is in the guide's table and needs 5 or more per group; there are 48.

Also an option

Because the same 48 stations are in both periods, the data is paired, and the Wilcoxon signed-rank test (the paired version of Mann-Whitney) would use that. Mann-Whitney ignores the pairing. That makes a real change harder to detect, never easier: say so in the evaluation.

Report

The median of each period, how many stations improved and how many got worse, and p. Free calculators: Mann-Whitney U and, for the paired option, Wilcoxon signed-rank, both from Social Science Statistics.