How Cities Actually Measure Whether a New Bike Lane Worked

A bike lane opens. Someone posts a photo of it empty at 2pm and declares the project a failure. Someone else posts one of it full at 8:30am and declares a triumph. Both photos are real. Neither means anything.

Cities do have methods for answering this, and they're more fragile than the press releases suggest. Know where the soft spots are and you can read an evaluation report in a few minutes and tell whether the city is being straight with you.

"Ridership went up" is the weakest possible claim

It's the headline every project gets, and alone it proves nothing.

Bike counts can rise for reasons that have nothing to do with the lane being good. Riders shift over from a parallel street, producing a big increase on the new lane and an equal decrease one block away that nobody measures. The after-period had better weather. The count happened in June instead of March. A university term started. A bikeshare dock opened nearby.

None of that makes the lane bad. It means a single before-and-after number on one street is a story, not evidence.

What a serious evaluation adds is a control: count comparable corridors that didn't get a lane, over the same period. If volumes rose 40% on the new lane and 12% citywide, you can argue about 28 points. If they rose 40% and 38%, you've measured a nice spring.

How the counting actually happens

Four methods do most of the work, and each one lies in a different direction.

Pneumatic tubes

Rubber tubes taped across the pavement for a week or two. Cheap, fast, the standard tool for a short before-and-after. They struggle to distinguish bikes in mixed traffic, undercount riders passing side by side, and only capture whatever weeks they were installed for.

Inductive loops

Cut into the pavement, permanent, counting for years. Vastly better data: seasonality, day-of-week patterns, long trends rather than a snapshot. Expensive enough that a city has a handful, not one per project, so your specific lane probably doesn't have one.

Video and computer vision

Cameras plus classification software. Where the useful stuff lives now, because you get behavior rather than just counts: which direction people ride, lane versus sidewalk, whether drivers yield at the conflict point. Video also raises privacy questions cities handle with wildly varying care.

Manual counts

A person with a tablet at one corner for two hours. Still valuable, because a human records what software misses: age, apparent confidence, whether a rider had a child aboard or gave up and dismounted.

Method	What it's good at	Main weakness	How it gets gamed
Pneumatic tubes	Cheap short-term volume	Two-week window, classification errors	Choosing favorable weeks for the "after" count
Inductive loops	Continuous multi-year trend	Few locations, costly to install	Quoting a peak month as the annual norm
Video plus classification	Behavior, conflicts, compliance	Cost, privacy, calibration drift	Reporting only the clips that support the conclusion
Manual counts	Rider demographics and detail	Tiny sample, observer effects	Picking the count location and hour
App and GPS data	Route choice across a whole city	Severe sample bias	Treating it as representative of all riders

The Strava trap

Fitness app data is seductive. Cheap, citywide, shows routes rather than points, and it arrives as a nice heat map.

It also systematically over-represents exactly the riders a new protected lane is not built for. People who record rides on fitness apps skew toward faster, more confident, more recreational cycling. They were already willing to share a lane with traffic. The person the lane is supposed to unlock, a nervous rider going two miles to a grocery store on a heavy bike, is not logging that trip.

So app data is useful for one thing: seeing which corridors people route onto and away from. It's close to useless as a measure of whether a lane brought in new riders, because the new riders are invisible to it. When a city leans on it as primary evidence, the budget for counting equipment probably didn't survive the project.

Crash data is too slow, so cities cheat

Serious crashes at any single intersection are rare events. Detecting a real change in a rare event with statistical confidence takes years of data and often multiple sites pooled together. Political attention spans run about eighteen months.

So cities use surrogate safety measures, observable behaviors known to correlate with crash risk:

• Conflict events. Video analysis counting how often a bike and a vehicle come close enough that one of them has to react. More conflicts, more risk.
• Post-encroachment time. How many seconds separate two road users passing through the same point. Small numbers are bad.
• Vehicle speeds. Measured directly, and the single strongest predictor of whether a crash injures someone badly. A design that cuts the 85th percentile speed has almost certainly reduced harm even if crash counts haven't moved yet.
• Sidewalk riding. An excellent proxy for whether people feel safe in the road. It should drop after a good lane goes in. If it doesn't, the lane isn't doing its job.
• Wrong-way riding. Usually a sign the network is one-directional where demand isn't.

Surrogates are legitimate, and they're used across traffic safety research precisely because waiting for crashes is both slow and grotesque. But they're proxies, and a report presenting a drop in conflicts as a proven reduction in injuries is overclaiming.

Why crashes can rise while the lane gets safer

This one catches almost everyone, reporters included.

If a lane triples the number of people cycling, the raw count of bike-involved crashes can go up even as each individual rider becomes much less likely to be hit. Three times the exposure at half the per-trip risk gives you a 50% increase in incidents and a genuinely safer street.

So a raw crash count has the wrong denominator. You need crashes per rider, or per mile ridden. Cities that publish only the raw count after a successful lane are handing critics a free headline.

Whenever you see a crash number in a bike lane evaluation, look immediately for the exposure figure it's divided by. If there isn't one, the number can be pointed in whichever direction the writer prefers.

The flip side is also true and less often admitted: a lane on a street nobody was going to cycle down can show a beautiful safety record because almost nothing happens there at all. Low numbers aren't automatically success.

The business impact fight

Every project involving removed parking triggers the same argument, and the measurement here is unusually messy.

What businesses report is survey data, and merchant surveys during construction reliably show harm, because construction genuinely does cause harm. The more objective measure is sales tax receipts or transaction data for the corridor, compared against similar corridors that weren't touched. Two persistent findings from that kind of work, stated as tendencies rather than guarantees: merchants overestimate the share of customers arriving by car, and people on foot and bike tend to spend less per visit but visit more often.

What nobody measures well is the distribution. Total corridor sales can hold steady while one business that genuinely depended on loading access gets hurt. Aggregate numbers are true and unhelpful to the person they happened to.

Which points at the thing that matters most and gets studied least: loading. Not parking, loading. Remove the space where deliveries happened without replacing it and the failure shows up as double-parked trucks in the new lane, a design failure wearing a driver-behavior costume.

What drivers actually lose

Cities do measure this, usually with GPS-derived travel time data before and after, plus signal timing analysis. Typical honest finding: general traffic travel time changes by a small amount, sometimes worse in one peak direction, sometimes better because turning movements got organized. Total throughput often changes less than expected, because a lane's capacity is frequently limited by the intersection rather than the number of lanes between intersections.

That's the point drivers find hardest to believe and it's the most robust one here. Removing a mid-block travel lane doesn't necessarily reduce how many cars get through, because the bottleneck was the signal all along.

One real cost does get buried: buses. If a redesign puts them into a single mixed lane with turning traffic, bus riders pay for the bike lane. A good evaluation reports bus travel time and reliability separately. Plenty don't, and that omission is worth noticing.

The metric that predicts success and rarely gets published

Here's what I'd actually look at if I could only see one number: does this lane connect to anything?

A protected lane that runs eight blocks and dumps riders onto a five-lane arterial with no continuation will underperform, and no amount of good design within those eight blocks fixes it. The best predictor of whether people use bike infrastructure is whether it forms a continuous low-stress path between places they need to go.

Planners have a formal tool for this, usually called low-stress network analysis: score every street segment by how stressful it is to cycle on, then count how many destinations you can reach without exceeding a comfort threshold. A lane that closes a gap can hugely increase reachable destinations. A lane in the middle of nowhere changes almost nothing.

This analysis exists in most planning departments. It almost never appears in the public evaluation of an individual project, because it would sometimes show that a politically achievable lane wasn't a useful one.

Metrics that quietly disappear

Watch for these going missing between the proposal and the evaluation:

• Who's riding. Rider age and gender composition is one of the most informative measures of whether infrastructure lowered the barrier to entry. Often counted before, forgotten after.
• Bus performance, for the reason above.
• Winter counts. These vanish constantly, and they're the ones that reveal whether maintenance is real. A lane that doesn't get cleared isn't a lane in February.
• Maintenance and enforcement. Nobody publishes how often the lane was blocked. Everyone who rides it knows.

How to read a city's own evaluation

A short checklist that will get you most of the way:

1. Is there a control corridor? No control means no causal claim, however confident the language.
2. Are the before and after periods comparable? Same season, similar weather, no confound like a nearby closure.
3. Is crash data normalized by exposure? If not, ignore the crash section.
4. Are speeds reported at a percentile? Average speed hides the dangerous tail; you want the 85th.
5. Does it report anything that looks bad? An evaluation where every single metric improved is either a remarkable project or a filtered one.
6. Is the raw data available? Cities that publish the counts are more confident in them than cities that publish only conclusions.

And if you're going to a public meeting about a lane in your neighborhood, the question worth asking isn't whether ridership went up. It's this: what result would make you change the design? An agency with an answer is running an experiment. One without an answer is running a campaign.

Common questions

How long before you can tell if a bike lane worked?

Volume responses show up within weeks. Behavioral changes like reduced sidewalk riding show up almost immediately. Anything involving crash statistics realistically needs years, often pooled with similar sites to say anything defensible.

Is an empty bike lane evidence of failure?

Not by itself, for the same reason an empty road at 2am isn't. Bike traffic peaks harder than car traffic, so an off-peak photo is uninformative. It becomes a fair question if the lane is empty during peak hours months after opening, and at that point check what it connects to at each end.

Why do cities keep building lanes that don't connect?

Because the unit of political decision-making is a street, and the unit of usefulness is a network. Individual segments get approved or blocked by individual constituencies, so networks get built in disconnected pieces. It's a governance problem wearing a planning problem's clothes.

What's the single most useful number in these reports?

Vehicle speed. Quick to measure, hard to fudge, tightly linked to injury severity, and it captures whether the street's design actually changed rather than just its paint.