At the most basic level: YYou take your congestion per citizen metrics before the market entry of uber, which should be linearly related, and extend that line. Then you draw another line with the metrics after uber. If you see a difference and you can't explain those away wwith other causes, you're confident it's ddue to uber.

The study very likely threw some more statistical tests, controls And models at it, but that's the gist.

Not an unexpected rresult either: personal vehicle use means the car is on the road twice per day, uber twice per passenger and the distance between passengers.