American Community Survey · journey to work · 1,742 U.S. cities over 20,000 people
The cities that bike more look the same: students, density, and commutes short enough to ride. Heat, rain, and long drives go the other way. Income does not help. A snapshot of every U.S. city of 20,000+ in the 2008–2012 Census, not a history of how cycling grew.
The short answer
Ranked by how much of the explained variation each factor carries once the others are held constant. Effects are per one standard deviation change, on a city's bike share. These are city-to-city associations, not proof of what caused cycling to rise.
The largest single predictor, carrying about a third of everything the model explains. College towns are young, low-car, short-trip places. Davis, Boulder, Corvallis, Eugene and Gainesville all fit it. County bachelor's-degree share adds a smaller positive effect on top.
Doubling density goes with about 15% more bike commuting, independent of everything else. Dense cities put destinations within riding range.
Hot summers (+6°F) cut cycling more than cold winters do. Mild winters help (+13°F → ×1.33), and 15 more inches of rain a year cut it by ~13%. Snowfall adds little once temperature is known. Within a single state these weather effects mostly vanish, so much of the climate signal is the West Coast pattern.
Where the county's average commute runs longer, fewer people bike. This holds even comparing cities within the same state, the most robust result after college share. City land area itself shows no effect.
Richer places do not bike more. Holding education fixed, higher median income goes with slightly less cycling (×0.93 per SD). Cycling in U.S. cities tracks education and age structure, not money.
Among the 50 largest cities, PeopleForBikes network scores correlate with bike share (partial r = 0.56 after density, students and winter climate). The networks are measured today and the commuting a decade earlier, so this shows co-location, not which came first.
Relative importance
Bars split the model's R² of 0.36 among the predictors (Shapley decomposition, so the bars add up to the total). Color shows the direction once other factors are held constant. The dots show the same ranking from a random-forest model that allows non-linear effects (permutation importance, scaled to the top factor).
Explore a factor
Each dot is a city. The vertical axis is logarithmic so small and large shares both stay readable. The line is the random-forest partial-dependence curve: the predicted bike share as this one variable changes, with all others at their observed values.
Geography
Circle area is proportional to bike commute share. The West Coast, college towns and Colorado's Front Range stand out. Most of the South and the Midwest outside university towns sits near zero.
Infrastructure
The PeopleForBikes network score (0–100) measures how much of a city's destinations are reachable on low-stress streets and paths. A citywide infrastructure measure was only practical to collect for the 50 most populous cities in the sample, so it sits outside the main model.
Minneapolis, Seattle, San Francisco and Portland pair the highest network scores with the highest big-city cycling. Houston, Dallas, Fort Worth and El Paso pair weak networks with almost none. The exceptions run both ways: Detroit, Milwaukee and New York score well with under 1% biking (New York's commuters mostly take transit), while Chicago scores 11 yet has 1.3%.
The causal arrow runs both ways. Cities with more riders face more pressure to build lanes, and lanes attract riders. Before-and-after studies of individual corridors are better evidence for the effect of building, and they generally find increases in ridership on the new routes.
Beating the model
The model predicts each city's share from its students, density, climate, commute time, income and demographics. The gap points to what the data can't see: bike networks, policy, culture, tourism and geography.
Ranked by the gap on the model's log scale, among cities with at least 10,000 workers.
International benchmark
Share of commuters who bike to work, from city surveys compiled on Wikipedia's modal-share tables. Survey methods and years differ, so treat these as orders of magnitude. Even Davis, the top U.S. city, sits below Utrecht, Freiburg and Amsterdam.
Measured over all trips rather than commutes, shares run higher in cycling cities: about 61% of trips in Groningen and 11.2% of trips within Paris (Institut Paris Région GPS survey, 2022–23).
Under the hood
Standardized coefficients on log bike share (log of share + 0.2 points), with heteroskedasticity-robust standard errors. Positive means more cycling. Gray entries are not significant at the 5% level. Adding region or state controls tests whether a factor still matters when comparing cities in the same part of the country.
All cities
Click a column to sort. ± is the survey's 90% margin of error on bike share; small cities have wide margins.
Methods and limits
Bike share is the percent of workers 16+ whose usual means of transportation to work is a bicycle (ACS 2008–2012 five-year estimates, all U.S. places of 20,000+). Each city was matched to its 2010 Census land area and population, to the ZIP codes within its footprint for NOAA 1981–2010 climate normals and median income, to its home county for commute time, education and age, and to College Scorecard enrollment at residential four-year colleges located in the city. The main model is ordinary least squares on 1,696 cities with complete data, checked with a random forest and with region, state and size-weighted variants.