The City of Zurich says cycling nearly doubled between 2012 and 2024 (source). We try to reconstruct the claim using the public bike-counter data.
Verdict: plausible, but not proven.
«In der Stadt Zürich hat sich der Veloverkehr von 2012 bis 2024 beinahe verdoppelt.»
"Cycling in the City of Zurich nearly doubled from 2012 to 2024." Based on the city's official traffic index, which puts 2024 at 193.4 against 100 in 2012 (Statistik Stadt Zürich; reported by Pro Velo, November 2025).
The city averages all counters that existed in a given year. Using the open data, a median over counters gives ×1.94, almost exactly the official ×1.93.
Counters were added, removed and replaced over the years. So we kept only the 8 counters that ran in both 2012 and 2024, and adjusted for weather and holidays.
We set "nearly doubled" at ×1.8 or more in advance, then resampled counters and weeks a thousand times to see how often the growth clears that bar.
2012 → 2024, all counters active each year.
The 8 long-running counters, adjusted for weather. 95% range ×1.63 to ×2.37.
Treating two 2019 device swaps as measurement changes, not real growth.
Confidence that growth was at least ×1.5. Nobody should doubt the trend.
Andreasstrasse grew faster than any other long-running counter. Its level relative to the median counter stepped up by about half when a new type of device was installed in October 2019, after a seven-month outage. Nothing else in its data changed: the split by direction and the daily rhythm look the same.
Count that step as real and the estimate is ×1.90. Count it as a change in measurement and it falls to ×1.79, just under the bar. Leave Andreasstrasse out entirely and it is ×1.79. The public data cannot tell which is right.
Roadworks and outages. Several counters dip for a year and recover. Dropping those years moves the estimate by about 1%.
COVID. Cycling surged in 2020 and fell back in 2021. Leaving both years out changes little, and 2012 and 2024 are unaffected.
Changing counters. Quiet counters were removed, busy ones added. A plain average would show ×2.14 instead of ×1.75. The city's median avoids most of that.
Few counters in 2012. With only 11 counters, the median rests on one of them. Dropping any one moves it between ×1.53 and ×2.16.
If yes, the honest summary is "up about three-quarters" (confidence about 0.24 for "nearly doubled"). If no, "nearly doubled" holds up (about 0.72). The city can answer this from its maintenance records.
Only roughly. We hid each counter in turn and rebuilt its 2025 traffic from the others, the weather and the street network. The best method is off by about 45% hour by hour and 42% on daily totals, and it runs about 15% low. A graph neural network did not beat simpler methods. Adding 29 cantonal counters around the city helped a little; steering routes along the planned Velonetz did not.