50,000 miles on cargo bikes

November 20, 2025

Sometime this year I crossed 50,000 miles of commuting to work and running errands on cargo bikes since 2006, starting first on a Specialized hybrid extended with an Xtracycle FreeRadical attachment, then upgraded to a Big Dummy, and then, also, an Xtracycle Edgerunner.  Until earlier this year, none of these had e-assist, now the Big Dummy does (750W Tongsheng TSDZ8) which makes it even more wonderful.

One other thing I noticed was several new, articulate advocates for e-biking and e-cargo biking.  I wish I was as good at it as there are, though I guess for some of them it is their part or full-time job and that is not me.  There was one thing that initially rubbed me a little wrong, where you would hear some of them say things like “I’ve been e-biking for a year and it changed my life” and talking about distances in the mere four digits.  My snap reaction was something between “you have no idea” and “oh my sweet summer child”, but the second snap reaction was that this is the wrong reaction, these people are better at advocacy than I am, and I did see wonderful changes in just a year, things were medically visible in less than six months, and also that “better in a year or less” is something that is more appealing to people, telling them “50,000 miles” seems unreal and unattainable so why bother, so really, this is fine.

But I wish I were a better advocate.  I did what I could; several people did test rides on my bike, I’ve helped more than one person get started with winter cycling.  I’ve got this blog, for what it’s worth.  If someone needs a pointer to a paper or a video to illustrate a particular point, chances are well better than zero that I can provide that.

But if I may — for someone who has started riding (e-)cargo bikes a lot — what might the long term hold?

Most important, there can be a long term.  I started one year, I got stronger, that’s nice, and pretty soon I wanted to see what else I could carry, and what else I could do, and most of it worked, so I got more confidence, and carried those things, too.  One of the first big lessons was learning that I could pretty quickly (within a few hundred yards) adapt to a terrible, wobbly, unbalanced load, and just ride.  Early on I learned how to tow a bicycle with a cargo bike, that’s darn useful.

Somewhere along the way I got tired of urban driving and parking, because stuck in traffic, ugh, who needs that?  And parking, yuck, it takes so much time and the spaces are so tight and hope you don’t accidentally ding another car with your door or vice versa, and it costs money, and if you read the signs wrong you might get a ticket, or worse, towed.  So that’s more reasons to bike, more reasons to drive less, and so I biked even more.

Over a few years, I didn’t get stronger exactly, but I got tougher.  An extra long ride, or a ride carrying heavy stuff, or rides with more hills in them, didn’t tire me out as much.  This may actually only have been getting very good at reading how I was feeling, whether I was exceeding my limits, and pacing myself, but it meant that I could “go” for a long time.  This also helped climbing stairs to view points in in Yellowstone and shoveling snow, which were both helpful.  And all of this is happening as I am aging — it should have been going in the opposite direction, but it didn’t.

My balance got better.  It was not bad to start with, I could already ride several of my bikes no-hands, but it got better.  I could ride no-hands at lower speeds, I could do controlled turns no-hands, I could ride no-hands with loads on the bike or towing a trailer.  Maybe you can’t ride no-hands so this won’t happen for you, but generally, my balance got a lot better.

My bike handling got better.  Nowadays I can confidently ride through skinny places, or on skinny lanes (like a Dutch curb, a little wider than a Boston granite curb, I’m not Danny Macaskill yet).  I’m better at sharp turns and maneuvering in tight spaces and making room for people walking.  All of this adds confidence and removes obstacles, so, I ride my bike, yet more, why wouldn’t I?  Some of it’s kinda fun.

Physically, all sorts of good things.  I’m more flexible, still. All the measured medical numbers are better, some a little, some a lot.  I’ve got extra circulation in my legs, not the horse veins you see on people who race, but more.  My doctor remarked on it, said I almost certainly have the same around the heart, and “that’s a good thing, especially if you get a heart attack”.  I didn’t lose much weight, so don’t count on that, don’t be sad when it doesn’t happen.  Maybe if I could find the time and patience to exercise the rest of me as much as I exercise my legs, that’s probably a good plan for when I retire.

Out of the blue, mentioned on a random Global Cycling Network video, I came across this paper on cycling in elderly Japanese (paywalled, sorry), a 10 year study where the subjects were (average) 74 at the start.  35.7% were cycling some amount at the start.   So odds are good I’ll still be on my bike 10 years from now.

Understanding traffic

September 1, 2025

I bike to work and to lots of other places because I don’t like traffic (and the Boston area has traffic) and I don’t like looking for parking.  Lots of other people choose to drive despite the traffic, but almost every single person in traffic complains about traffic.  I don’t want to just go with the glib “you ARE traffic”, but instead, maybe help people understand what causes traffic congestion and why, because if we don’t understand the causes, we’re unlikely to come up with solutions or mitigations that work.,

Slightly updated after feedback.

Throughput versus speed. Road speed is the how quickly a driver can get from point A to point B if nothing gets in their way.  Road throughput is about how many cars can travel past a line across the road; if that rate is exceeded, then drivers will experience “something in their way”, namely, other drivers.  If a route is congested, the maximum possible speed for the cars traveling on it is irrelevant, because other cars are in the way.  Suppose A and B are just a mile apart on a straight road, and the speed limit is 60mph.  Ideally, that trip takes a minute.  Suppose there’s 90 cars on that road in front of the driver.  Standard throughput estimates say that it will now take 3 minutes to cover that distance, or an effective speed of 20mph.

Limited throughput.  Cars are not an efficient way of packaging people for transportation.  They take up a lot of space on the road, and the constraints of reaction time and hedging against bad luck mean that realistically, the throughput for a single lane is at most 1800 cars per hours, or one car every two seconds.  If you see a single lane controlled by a light that is green for 45 seconds, that means 22ish cars will pass in that light cycle.  Car throughput is maximized at around 30-35mph; slower than that and the length of the cars does not pass quickly enough, faster than that and a combination of both the superlinear increase in stopping distance and even just the safe following distance rule guarantee that cars are generally separated by more than 2 seconds.  For comparison, the Red Line, working properly and in a single direction, carries double that only in seats, so-called policy capacity is 6 times that, “crush capacity” is almost 10 times that.  Bicycles manage double that in a bike lane, and in a full-sized lane bikes (specifically, kids on bikes popping wheelies) can manage at least 5.5 times that flow.  It is also very important to remember that the throughput of an actual traffic jam is extremely low; parked cars are not traffic flow.

Non-linearity. Returning to the 1-mile A-to-B example with the 60mph speed limit, the interesting number is 30 cars per mile.  At that density, in a minute, one car passes B every two seconds; if there were any more cars on the road, it would take longer than 60 seconds for all of them to clear, so the last car cannot be traveling at 60 miles per hour.  Below the threshold rate, there is no congestion — adding 10% to traffic adds zero to delay, but above it, there is — adding 10% to traffic density adds (at least) 10% to delay.  This is especially notable at traffic lights; if, during each cycle, more cars appear at a light than can pass on the green, then the line of cars at the light will grow longer with each cycle.

The tightest bottleneck.  If a straight stretch of road has 4 intersections with stop lights for cross traffic, and one of those lights is green for 20 seconds for the straight road and green for 40 seconds for the cross traffic, then the end-to-end throughput of that road (ignoring turns on/off for the sake of simplicity) is 1/3 of its hourly capacity, or 600 cars per hour.  Widening the road won’t fix that intersection, except that in practice, the intersection will also have lanes added and that will increase throughput; not the wider road, which is mostly wasted space. And fixing that one intersection will only improve the road to the throughput of the next-worst bottleneck.  Cars are especially susceptible to bottlenecks because they use space so inefficiently, but even in efficient systems, there will still be a tightest bottleneck.  On the London Underground, it is often the escalators to and from the platforms.  However, the throughput for a single escalator (2 people wide, both standing) is 151 people per minute, or 5 times the capacity of a 10-foot-wide car lane for cars.  This is also why Elon Musk’s proposal for his Boring Company to build tunnels for automobiles in cities was so stupid, because no matter how good the tunnels were, he proposed to access them with CAR ELEVATORS, which have a ludicrously low throughput.  (Intersection width correction from Hacker News)

The marginal driver’s preferences.  If there are enough people interested in driving that they can exceed the capacity of the road, what determines the traffic level?  The answer is that it is the “marginal driver(s)” — those whose desire/need to make that trip at that time exceeds the delay and awfulness of sitting in that traffic.  People vary in how much they hate traffic, and they vary in how much they are able/willing to adjust their travel to avoid it.  If they choose to avoid it, they do not add to it, but if they drive, they make it worse. This leads to a traffic level that is not entirely stable or predictable; people use recent experience to predict what the traffic today or tomorrow will be like, and plan their travel accordingly.  But some drivers will plan to drive no matter how bad the traffic is, and they set the level for everyone else.

Latent demand and induced demand.  Just as there are marginal drivers, there are marginal not-drivers.  These are people for whom the traffic hassle, or traffic delay, are just too much, and they either choose another mode, or travel earlier or later, or just don’t travel.  If traffic was not quite so bad, they would drive.  These people are latent demand, and they exist for all traffic modes — if the trip were shorter and the road was nicer, perhaps they would walk, or bike. If there was better, safer, bike parking, they might start biking to the subway. Or if the bus came more often, or if they could be assured they would find a seat, perhaps they would take the bus.

Through traffic is fluid. For some cities and towns (like the one I live in), most of the traffic is driving through the town, not to it or from it.  Through traffic always seeks the best route from start to finish, and is not particularly committed to a particular route.  If half of our town’s local traffic disappeared (perhaps everyone decided to carpool, or work from home, or ride a bike), the traffic level would not fall by half, because the reduced congestion would create a more attractive route for people trying to get to Cambridge; more of them would chose our town as part of their route.  Or, if a road was blocked for long-term construction, through traffic would adjust, and seek other routes to avoid the mostly blocked one, and traffic would not be as bad as a static analysis of road use might suggest.  You can also see this sometimes on a commute home; if some other large route is blocked, all the alternate routes pick up the spillover traffic, “where did all these people come from?”

Flow competition and Braess’ Paradox. Because different routes can share the same road segments, there can be “competition” in their flows; increased traffic from A to B can create jams for people traveling from C to D, sometimes without even affecting the people traveling from A to B.  An easy example of this is when the routes from A to B  and C to D share an interstate, but the A-to-B route enters the interstate first.  A-to-B traffic on the interstate may be flowing freely, but still be heavy enough that it is difficult to enter the interstate.  That will cause backups on the entrance ramp, perhaps extending onto the feeder roads.  The same phenomenon occurs if the A-to-B traffic is on a through road, and C-to-D must enter from a side street with no light to pause the cross traffic.  In some cases two different routes for the same source and destination can compete with each other in not-entirely-intuitive ways, such that removing a road segment, would block the competition, and actually improve overall flow. This is Braess’ Paradox. 

These are all reasons why, when faced with terrible traffic, “just add more lanes” is not likely to solve the problem.  The backup is usually caused not by lanes, but by intersections, and even if the bottleneck at one intersection is fixed, there is another intersection down the road that might be only slightly less constricted, and now it binds.  If, after fixing all the intersections, flow is improved, people who were making do with something other than driving at the peak rush, will show up to consume the new capacity.  Long-term, housing and employment patterns will change to take advantage of actual new capacity. And even if all this works and the latent/induced demand is not too much, delivering this improved flow “downstream” may back up some other route through competition, or worse, back its own self up because of Braess’ Paradox.

What might work better?  Attractive alternatives.  The harder it is to choose not to drive, the worse traffic will generally be, if people have no choice they will put up with terrible traffic.  If the alternatives are nice, they’ll be more easily deterred by traffic, and thus traffic will not be as bad.  What might these alternatives be?

• Better (faster, frequent) transit.  Transit should also be well-connected; it should meet bike paths, where systems or modes meet, transfers should not involve gratuitous delay (I have a decades-old memory of rolling into a train station, on schedule, and as the train braked to a halt, watching the bus I wished I could catch, depart the station without me, also on schedule).
• Alternate modes (buses, bike routes, subways, trains) should actually connect people to their destinations.  Silicon Valley has a lovely commuter rail line than runs through the centers of many cities in Santa Clara Valley and along the San Francisco Peninsula, but many big employers (e.g., Intel, Oracle, Meta, Google, Tesla, Dell) are located well away from the city centers, so any rail commute requires something else (shuttle bus, bicycle) for the last mile.
• Transit-in-traffic should receive priority; buses should have reserved lanes so they are not delayed by car traffic (why take the bus if it is no faster than a car?), buses and trolleys should have signal priority so that they wait as little as possible at red lights.
• For shorter trips, safe and relatively direct places to ride bicycles, and abundant protected (from theft and weather) places to park bicycles.
• For some people, the option to use small e-things or e-bikes, which means, providing places to charge for people who need to charge.
• In general, any easily-addressed impediment to using a non-car mode, should be addressed.  Why don’t bus stops have seats?  Why aren’t those seats protected from sun/rain/snow?  Housing, office, and business zoning often includes a car-parking requirement — why not a bicycle-parking requirement?

A connected network for whom?

July 7, 2025

About a year ago, I helped move some young relatives in and out of Boston, one leaving Chinatown, the other moving in on September 1 (locally, the day when most of the students renting in the area move in; it causes predictably awful traffic and parking problems).  Because of the parking problems, I did most of this help using a bicycle trailer with a 300lb capacity.  This generally worked well, because even a large bicycle with a trailer is actually small; it parks on the sidewalk and takes less width than a newspaper vending machine or a mailbox, and, very importantly, the bicycle could park directly in front of the pickup or drop-off, and did not require carrying stuff from some more distant parking space.  We also used cars to help with one move, but overall it seems that a young person’s stuff would fit in about 4 bicycle+trailer loads.  In both cases, more than would fit in a single car load, given the cars available (try renting a truck on September 1 in Boston, good luck).

The one glitch in this plan came from routing.  The bike is longer than normal, the trailer has a 64″ long bed and is a little wider than the bike.  Route planning software (Google Maps, Pointz) does not model bicycle trailers, and Google Maps appears to only model a “conventional” bicycle.  Pointz at one time used a more fine-grained model (“cargo bike”, “e-assist”) but appears not to now.  In one case I got routed via pedestrian foot bridges onto and off of the Esplanade along the Charles River.  This is bad for a couple of reasons.  First, the bridges declare “walk your bike” which is just about impossible with a loaded trailer; riding the bike, in a low gear and pushing only with legs, it’s not hard to handle the bridge grade, but walking the bike, pushing yourself with legs but then pushing the bike with your hands, that is hard work and also complicates steering the bike.  Steering the bike is important, because the long bike, towing a trailer, will just barely fit around the curves — you must approximately scrape the bike around the outside wall of the walkway to ensure that the trailer clears the inside.  If you’re walking the bike, at every curve, you must be walking on the inside, not the outside.  I, rebel that I am (and also very experienced at balancing and controlling a bike at slower-than-walking speeds), did not walk my bike, and instead just rode slowly and carefully, checking for no oncoming traffic before negotiating turns.  But this was not good, and the next time around I took care to avoid this route, instead either taking the Harvard or Longfellow bridges directly, which worked okay, even with the slopes on the Longfellow.

The larger problem here is that the bicycling “network”, is inadequate both in its extent, and in its vision for what someone might do on a bicycle, and if someone strays very far from the blinkered vision of its designers, the inadequacies are all the more glaring.  I have strength, skill, experience, and an extraordinarily capable not-e bicycle, so I can negotiate these gaps, but given this under-designed network, that makes me a rare bird.  For most people, the problem is not so much the bicycle’s ability to do things, as it is the network’s ability to do things.  Most people can actually haul trailers and handle cargo bikes just fine with not much practice at all, and once you ride regularly, the ramp up in skill is surprising and delightful.  But if the network doesn’t support that for people just starting out, then they won’t start, and it won’t happen.

Complicating things further, if someone is using an e-bike, they must factor in legal restrictions on where they’re supposed to use it, and how well those laws are enforced (it varies — twice I’ve been stopped riding around Fresh Pond, asked about the “motor” on my bicycle, but that has never happened on the Minuteman Bikeway) and their appetite for breaking bicycle traffic law. And there are some people on bikes who care about these details; they ask about the law, and they take care to follow it.  Nowadays most cargo bikes come with e-assist and many manufacturers are only selling e-assist cargo bikes, so for someone on a newish cargo bike who cares about the law, many links in the existing network disappear.

(Before anyone goes off on a rant about scofflaw cyclists, yesterday I drove home from NY to MA, I can read my speedometer, we drivers clearly do not actually respect traffic laws, so just stow that rant, okay? The social pressure to break speed limits in a car on the interstate is pretty intense, whereas the social sanction for those sketchy bicycle scofflaws is well-known, and lots of normal people don’t want to be one of “those cyclists”).

The e-bike problem is getting worse, because apart from problems with the legally-defined network, people are using e-bikes, and some of them are using e-bikes for delivery, and they have a monetary incentive to ride as quickly as they can.  Even at legal e-bike speeds (for MA, 20mph) the protected bike lanes recently added in places like Cambridge are not exactly designed for that, and given existing, slower, bicycle traffic, the problems only get worse.  Again, bicycles have uses not anticipated by the route designers.

We’ve also not done a good job with easy access to bicycle parking / loading zones (is there enough of it? is it roll-up accessible? PARKING IS PART OF THE NETWORK) and as a result we also end up with delivery e-bikes traveling on sidewalks for the last few yards of their trip, or parking their e-bikes in whatever random parts of the street or bike lane look unused enough to allow a stop for “just a minute”.  We have this very same problem with people using cars for quick pickup and drop-off; because we decided to prioritize long-term car storage in streets over loading zones, there’s never enough legally-defined and convenient loading zones, and so all of that gig-economy car pickup and dropoff ends up being improvised, often in bike lanes, sometimes in fire lanes, etc.  The human behavior is exactly the same, but there’s even less legal accommodation for quick-stop bicycle commerce than there is for quick-stop auto commerce, even though the space required for bicycle parking is much smaller and cheaper.

The (un)availability of bike share also affects the network.  Bikes can break down, someone who gets around by bike needs access to a spare bike (or spends money on Lyft/Uber till the bike is repaired).  Bike share is a great source of spare bikes, but that means that bike share needs to be widely available; if it’s not, that degrades the network.  Some towns are in the bike share network, some are not.

What should we do about this?  I have opinions, and they are:

• Where existing links are inadequate for bicycle use, we should improve them, perhaps including replacement or adding redundant routes.  There are apparently some nearly-pre-fab bridge systems that have a relatively low design+install cost, we should consider these.  Because of the cost of upgrading after the fact, we should overbuild these.  We want more biking, we want more bicycle deliveries, these should make that be easy.  Redundant routes are fine, because the older ones are designed and built for pedestrian traffic, will tend to not attract bikes if these alternatives are installed, and redundancy will de facto provide separation for anyone walking who doesn’t want to share a bridge with a bike.
• We need an oversupply of bike parking and bike loading zones. Parking is part of the network. Do we want more or fewer bicycle deliveries?  Look at trucks, the noise they make, the pollution they cause, the wear-and-tear on roads, the risk to pedestrians and cyclists, we cannot possibly prefer a truck delivery when a bicycle delivery is possible.  If we want something to happen, we should make it easy, and bike parking is cheap compared to car parking.  It’s boneheaded not to have an oversupply.  Furthermore, when/where it is widely used, we should consider quality; an awning to keep the weather off parking costs more, but is better.
• We should legalize use of type I and type II e-bikes wherever normal bikes are allowed.  *This includes the sidewalk, of course with the same low-speed-and-yielding restrictions*. Poof, the network extends further (because e-bikes have more range) and missing links for e-bikes are restored.  The (20mph) assist limit matters, I am not at all enthusiastic about type III e-bikes and I am also okay with the (18mph) assist limits imposed on BlueBikes, even the new 15mph BlueBike limit in NYC is okay. The same-as-bikes and no-assist-past-20mph rules are linked: 20mph is a speed I could manage for over an hour on open roads when I was a kid, but Boston, Cambridge, Somerville, and NYC are NOT open roads and have a lot of pedestrians in them, 20mph is already very dubious in mixed traffic, faster than that in mixed traffic is a really bad idea.  (The assist limit in Europe is 15mph, it’s not a terrible choice.) 
• Throttle e-bikes are fine and necessary.  There are disability-related reasons to allow this, but also, to the extent that a human is just a couple-hundred-watt motor, what on earth is the difference?  People using e-bikes for delivery cover a huge number of miles in a day, they will get tired, if nothing else there will be actual wear and tear on their clothing from constant pedaling (I ride a mere 12 miles per day and it’s an issue).  If there’s a network, this will let some people make full use of it, and get cars and trucks off of roads.
• BUT: Because of pedestrian safety/comfort concerns, and because even 20mph is a little high in crowded places, enforcement, probably semi-automated, of bicycle speed limits for e-bikes is fine. More people biking, more people biking on ebikes, there will be more conflicts, and we need to control these. I imagine something involving multiple radar and cameras, a “your speed” display, and an occasional cop sitting in the shade some distance down the path pulling people over who had been recorded speeding.  The camera stream and enforcement record should be randomly sampled to check for bias, and then all that data should be destroyed, by law, with an audit trail (why am I so detailed here? Because privacy and bias are real problems, even if I do think most people who raise them are just looking for ways to thwart enforcement so they can speed more freely.)  
• AND: I don’t have a lot of faith in education, but the default education should prioritize NOT CRASHING INTO PEDESTRIANS.  The current how-to-ride education does not, overemphasizes “the law”, and generates all sorts of judgey-blamey behavior from some cyclists not unlike what many drivers do, because “rules are for everyone” with no regard to the differences in risk contributed by the different modes.  People walking are legally allowed to dress as ninjas, are legally allowed to wear noise-cancelling headphones, are legally allowed to have a phone in their face, they are legally allowed to glance up the road and see that a car has the ability to stop before a crosswalk and then step into that crosswalk and expect that driver to stop.  Given how enthusiastically both cyclists and *especially* drivers break traffic laws, it is absurd to expect pedestrians to do anything more than conform to the bare minimum legal requirements.  Bicycles are not cars, cyclists are far better at perceiving their environment and communicating with pedestrians, have a better reaction time, and take up far less space, the default should be “I can avoid that pedestrian no matter what they do”, and if that’s not true, that’s a bicycle speed problem, not an inattentive pedestrian problem.

Nice “interaction” video / “why protected bike lanes”

July 1, 2025

I managed to collect a video recently that might help explain “why protected bike lanes?”

TLDR: a thing that bike lanes help with, is simplifying safety even when they cannot completely eliminate risk. This is one of the reasons they’re less stressful.

One problem with a lot of bicycle safety discussions is that people tend to focus on Just One Cause for each crash, when I am pretty sure (from observing, thinking, and watching, some of this for years, there are books on some of this too) that we don’t get actual bad crashes without 2 or 3 things happening at the same time. If it’s just one thing, we’re kind of expecting one thing, our roads are kind of designed for mitigating one thing, when just one thing happens, we almost always cope.  And also, “just one thing” is happening continually, for varying values of “thing”.

What this video shows is 2 things happening one right after the other in a way where the first could distract from the second and cause someone on a bike to then make a mistake, and I think it’s instructive to try viewing it, full screen, at real-world speed, and see if you can figure out all the things that happened and when they happened from the first viewing. Then try it at half speed, then feel free to read the description of what I remember and what I was able to pull from the video (at the video itself, or the bottom of this post).

(Regarding my reaction — it was about what I would hope for from someone with 50,000 miles of experience on local streets in the last 19 years and who had already ridden that stretch of Broadway in that direction 2000 times. That amount of experience should not be a requirement for feeling safe on our streets.)

One goal of protected bike lanes is to help simplify car-to-bike interactions and reduce the incidence of these double-fault interactions. There are still risks remaining, but certain risks are well-reduced (dooring), and oncoming traffic swerving into your face is pretty much eliminated. The need to handle two things at once is very reduced. This is not unlike my favorable experience using rotaries in left-side-driving countries; they rely heavily on rotaries in Australia and New Zealand, but you only need to watch One Thing to enter the rotary, and that is vastly easier than watching two things. A German colleague reports a similar experience driving in Netherlands (vs Germany, or the US) — driving in the Netherlands is apparently not that stressful because each intersection has been designed so that you only need to watch for one thing, not many.

What I recall, what the video shows.

Overall, the video shows an incoming car passing a parking car at a pretty good speed, swerving into my lane while I am in it.  This draws my attention for a moment, then I am looking at the car in front of me that is signaling left, then it isn’t (I specifically remember the missing “blink”), then it is signaling right, all while I am approaching, and very soon after the first right-signal blink, I brake hard, then swerve a little, and pass slowly at a wide clearance.  The car doesn’t move (much), and I look back as I ride away and it still hasn’t moved.

More detailed, after adding the speed and acceleration data to the video and viewing it frame by frame:

The blinker is on and then off again for about 1/3 second each (21 frames at 60fps). The last left blink occurs 3 seconds in, and then 56 frames later the right signal blinks instead (that is, after the blink stops, 21 frames elapse, the next left blink does not occur — and I remember noticing this missing blink — and then 35 frames (0.6s) later the right signal blinks). I remember that the car had my full attention because of that missing left blink, but looking at the amount of time I had to think that, it was only 0.6s, so, really, “full attention”?

After the first frame in which the right signal is lit, 24 frames (0.4s) elapse, and the Y-axis decleration rises to 0.5g and the speed is 16.7mph. This could be noise, not my reaction. Note that the camera is not level so the y-axis is measuring some gravity, and the camera is on my head, so my movement also matters, and also we don’t know the exact algorithm the camera is using to estimate speed at a frame-by-frame level (what it measures, is acceleration, and rotation, cross-checked with less-fine GPS data).

28: 4 frames later the y-axis deceleration is 0.6g, speed is 16.7mph.

32: 4 frames later the y-axis deceleration is 0.7g, speed is 17.2mph.

36: 4 frames later the y-axis deceleration is 0.8g, speed is 17.0mph.

40: 4 frames later the y-axis deceleration is 0.9g, speed is 16.7mph.

46: 6 frames (1/10 s) later, speed is 15.6mph.

52: 6 frames later, y-decel = 1g, is speed is 14.9mph.

58: 6 frames later, speed is 13.7mph.

60: 2 frames later, y-decel = 0.9g, 13.3mph, 

66: 12.5mph

72: 11.5mph

78: 10.9mph

84: 9.9mph

90: 9.1mph

96: 7.9mph (1g in one frame)

102: 7.1mph, y-deceleration falls to 0.8g, also turning.

108: 6.2mph, y-decel falls to 0.7g

114: 5.1mph, 0.5g

From frame 40 to 102, (62 frames, so a hair over one second) the deceleration was 0.423g. (16.7 – 7.1) = 9.6mph, times 60/62 = 9.29mph/second, times 5280/3600 = 13.63 fps/s, divided by 32.2 = 0.423g.  That’s well beyond the rear wheel skid braking deceleration (0.25g on a normal bike) and 85% of the customary estimate for flipping a normal bicycle over its front wheel (0.5g).  This was on a longtail cargo bike, a heavy bike loaded low and behind the rider so it should have a little more braking headroom before flipping, but because I also ride normal bikes, harder braking is a risky habit. My prepared reaction time, from first instant of right blink to initiating a reaction, is about half a second, the brake lever is clearly being activated in a linear fashion from frames 28 to 40.  For a 65-year-old man, don’t expect better than this.

(Bicycle) Speed and safety

June 14, 2025

For years I’ve heard people talking about how motor-assisted bicycles “are safer” because they can “match traffic speed”, but I think this is mistaken.  Here’s three reasons why.

For me the most compelling reason comes from a 2007 paper by Laurie F. Beck, Ann M. Dellinger, and Mary E. O’Neil in the American Journal of Epidemiology, “Motor Vehicle Crash Injury Rates by Mode of Travel, United States: Using Exposure-Based Methods to Quantify Differences“.  It’s paywalled, but the main results are tables of fatal and injury crash rates per trip.  Per one billion trips, the fatal injury rates for cars, motorcycles, walking, biking, and buses are 92, 5366, 137, 210, and 4.  The corresponding nonfatal injury rates are 8030, 103366, 2155, 14612, and 1608.

In the statistics by gender, the most dangerous thing women do (excluding motorcycling) is walking, with 80 deaths per billion trips, and the safest thing men do (excluding riding the bus) is driving, at 124 deaths per billion trips. What’s interesting to me about this is that buses are really safe, motorcycles are really unsafe, and despite our perception that biking is unusually unsafe, actually biking, walking, and driving risks are in roughly the same ballpark, and other factors (like gender, like rental bike share) also matter quite a bit.

But, returning to “keeping up with traffic”. That is definitely something that motorcycles can do, and they’re not very safe. Maybe keeping up with traffic is safer up to a point, but there must be a point at which additional speed is not safer, because the per-trip fatality rate on a motorcycle is 25 times higher than it is on a bicycle, and how does it get there? A factor of 25 is a lot of increase, too. So what is that point? Is it 40mph? Is it 30mph? Is it 20mph? This makes me wary of “keeping up with traffic” because somehow other factors intrude to make motorcycle riding 25 times riskier than biking, and bicycles and motorcycles are otherwise somewhat similar.

The second reason I’m wary of “bicycle speed to keep up with traffic” is that in Europe, the speed limit for e-assistance is 15mph (25kph), not 20mph like it is in the US for the more bicycle-like e-assist (type 1 and type 2; type 3 is 28mph, not legal in all states, and often comes with additional restrictions on where the bikes can be used).  The Europeans do this because they think it is safer than 20mph in important ways.  This raises the question, are the Europeans better at safety than we are?  And, oops, yes they are, their record on road safety is far superior to the US.  They kinda are the road safety experts, and they think 15mph is safer than 20mph for e-assisted bicycles.

The third reason comes from the relationship between car speed and fatality rates for cars crashing into pedestrians, focusing on speeds below 50kph (31mph). The risk of death increases from around 1% at 15mph, to about 8% at 50kph. This is not exactly bicycle crashes, but it illustrates how quickly risk rises for crash speeds in the range contemplated for e-bikes. If “keeping up with traffic” means 28mph instead of 15mph, then, to be safer, the crash rate must fall by more than 4/5 because the crash lethality increases by a factor of 5.

edit/ps: one issue with this reasoning is how much crash energy depends on car speed and how much it depends on bicycle speed, and how that’s distributed for the different crash types.  For a plain overtaking crash, it depends quite a lot on the car and somewhat on the bike, with “crash” including both the impact with the car and finally coming to rest.  For most of the rest—right hooks, dooring, failure to yield, loss of control from road conditions (potholes, slots)—the crash energy depends mostly on the bicycle speed.  There’s an obvious urban/rural skew here, with overtaking relatively more common outside of dense areas, and less so within dense urban areas.  Stroads are worst-of-both-worlds, of course.

There are other reasons to be wary of high speed on a bicycle, less quantified, but based on years of experience and observation, biking at various speeds and sometimes crashing, and other times nervously watching drivers and wondering if they see me (or not).  The bad crashes involve cars.  Drivers aren’t expecting bikes to move very fast; even 20mph seems to surprise them.  Higher speeds also give whoever is on the bike less time to react.  Both of these things make crashes more likely, and because the bicycle is moving faster, the crash is more violent.  Edit/ps again there is a rural/urban overtaking/not skew here.

I think it’s also helpful, for people who don’t actually measure their speed, to quantify how fast people actually ride bikes.  To a usual observer, a bicycle going 20mph looks like it is going “very fast”—if they weren’t observed together, anyone watching a car and a bike pass by at 20mph, would judge that the bike was “fast” and the car was “slow” (and that the bike was distinctly faster than the car).  Most people bicycle commuting, at least in Cambridge, are traveling below 20mph.  E-assisted blue bikes around here are limited to 18mph, and they’re pretty quick compared to most people commuting.  At the same time, someone who is determined to be fast on a bicycle and trains for that, will probably become fit enough that they can travel at 20mph for perhaps an hour (I managed this as a kid).  Other lucky or especially determined people will be fit enough that they can maintain 25mph for an hour (as a kid, I knew a few people who could do this).  Elite athletes can do better, but there’s none of them on the morning commute.  Sprinting is a thing, but it’s not that common, most people are just riding along.

Cybertruck vs Pinto safety

February 27, 2025

I wrote this up because I saw some articles that were obviously the result of someone feeding some bullshit “debunking” the Pinto vs Cybertruck catch-fire-and-kill-you comparison to some credulous journalists.  There are tools for making sense of this, instead of flapping our hands and going “math is hard!”

Given an event that occurs 8.5 times per million trials (the rate of car fire deaths per Ford Pinto), how likely is it to see 4 or more events in 35000 trials (the rate of car fire deaths in Tesla CyberTrucks, excluding that guy in Las Vegas)?  That is, if the Tesla Cybertruck is only as dangerous as the Ford Pinto, a 40-some-year-old car famous for catching fire in crashes, how likely is it that we would have seen this many car fire deaths already?

To solve this problem, we can model the situation using a Poisson distribution, which is appropriate for counting the number of events occurring in a fixed interval (in this case, 35,000 trials) when the events happen at a constant average rate and independently of one another.

Step 1: Determine the expected number of events (λ)

The event occurs 8.5 times per million trials (1,000,000 trials). We need to scale this rate to 35,000 trials.

λ = rate per trial * number of trials

The rate per trial is 8.5/1,000,000 = 0.0000085, and the number of trials is 35,000.

λ = 0.0000085 *35,000 = 0.2975

So, the expected number of events (λ) in 35,000 trials is 0.2975.

Step 2: Poisson probability formula

The Poisson probability of observing exactly k events when the expected number is λ is given by:

P(k) =e-λ * λk / k!

We want the probability of seeing 4 or more events (P(k ≥ 4)), which is:
P(k ≥ 4) = 1 – P(k = 0) – P(k = 1) – P(k = 2) – P(k = 3)

Step 3: Calculate probabilities for 

k = 0, 1, 2, 3

Using λ = 0.2975:

P(k = 0) = e-0.2975 * 0.29750/0! = e-0.2975 * 1 = e-0.2975 ≈ 0.7426

P(k = 1) = e-0.2975 * 0.29751/1! = e-0.2975 * 0.2975 ≈ 0.7426 * 0.2975 ≈ 0.2209

P(k = 2) = e-0.2975 * 0.29752/2! = e-0.2975 * 0.08850625/2 ≈ 0.7426 * 0.044253125 ≈ 0.0329

P(k = 3) = e-0.2975 * 0.29753/3! = e-0.2975 * 0.0263506/6 ≈ 0.7426 * 0.00439177 ≈ 0.0033

Summing these probabilities:

P(k = 0) + P(k = 1) + P(k = 2) + P(k = 3) ≈ 0.7426 + 0.2209 + 0.0329 + 0.0033 = 0.9997

Step 4: Calculate P(k ≥ 4)

P(k ≥ 4) = 1 – P(k < 4) = 1 – 0.9997 = 0.0003

Or in words, if the Tesla Cybertruck were only as safe as a Ford Pinto, it is is very unlikely that we would observe this many (4) car fire deaths so soon.

But how risky is the CyberTruck, compared to the Pinto?  To get this answer, one can integrate over the relative risk of cybertruck to pinto, notice where that integral reaches 0.025 (2.5%) and 0.975 (97.5%); 95% of the area is between those two points. Hacking this up in a Go program, I get a range of likely Cybertruck fire-death-risks, relative to the Ford Pinto, between 3.8x and 9.8x.

Feel free to check my work, I am not actually good at this (and caught a serious gaffe earlier).  But I also think this is generally kind to the Cybertruck, because the average Pinto spent far more time on the road than the Cyrbertruck has been in production.

Car trouble

December 31, 2024

Cars are useful, but cars have problems.  Some of this is obvious but we don’t want to notice it, so we don’t, some of this is obvious but we treat it like some sort of weird we-have-no-alternatives dilemma (it’s not).  Or, we make lame excuses. This is all pretty objective stuff, the main reason we ignore it is because when we drive we receive the benefits, and the costs are mostly somebody else’s problem.

So, car trouble.  

First things first: cars kill a lot of people in this country.  Killing people is bad, right?  Is anybody going to seriously argue that a few ten-thousand crash deaths are totally worth it because I like to get to work 5 minutes faster every morning? And yes, many of the crash deaths are caused by “bad drivers” — but somehow, we seem unable to identify these bad drivers before they kill, and we don’t keep them off the road.  Many car crashes are also caused by plain old human error, amplified to lethal scale because we don’t care to restrain the size, speed, or use of cars “too much”.  I know of crashes caused by a stray insect flying into a car, by an unruly dog, and in one case a rollover in front of an elementary school caused by who-knows-what.  Human error is eventually guaranteed; it is in our nature to cut corners until we discover (sometimes painfully) which corners matter.  This is hardly inevitable; among the countries in the OECD annual road safety report, the United States has unusually many road crash deaths per capita.  The lame excuse that intrudes right about here is “yes, but we drive a lot here”.  CONGRATULATIONS, on recognizing that excess exposure to risk amplifies the risk, DO YOU THINK PERHAPS WE SHOULD DO SOMETHING ABOUT THAT?  And even so, it turns out that measured per-mile, the risks are still high.  So perhaps we could do something about that, too.

Some of this comes from penny-pinching unsafe vehicle design.  In quite a few other countries, large trucks are required to both have overrun side guards, and better-visibility cab designs.  We don’t do that here because the trucking industry prefers to kill people.  They don’t say that they prefer to kill people, no, that would be terrible, what they actually prefer is to not do the things that would kill fewer people.  That’s totally different!

Second, automotive pollution causes harm in ways that are both large and still being discovered.  The visible, obvious crash deaths are actually the smaller number; particulate and nitrogen oxide pollution from cars and trucks cause over 50,000 early deaths annually. We cannot point to a particular death and prove it was caused by pollution so nobody will be hauled into court for “pollution murder”, but we can measure it in a population, the deaths are there.  Good drivers do this too, though diesel engines are a particular offender. Particulate pollution, besides causing early deaths, is also correlated with other ills, like lower birth weights, higher incidence of dementia, and even increases in crime. Noise pollution also causes measurable harm, at minimum, increased blood pressure.

But also, tires. Recently, we learned that car tires are the largest source of ocean microplastic pollution (*). Only a few years ago, we learned that a chemical in tire rubber kills Pacific Coho salmon.  It kills some other fish, too, who knows how long the list actually is, we’re still figuring these things out.  We’ll assume it’s not bad for humans till we learn otherwise, I guess.  

Third, cars appropriate and ruin public space.  Cars are loud; next to a busy road, it’s difficult to listen to quiet music, difficult to talk on a cell phone.  Cars take up too much room, both for travel and for storage.  On a road, the lane used for parking cars, could have been used for a streetcar, or as a bus lane, or as a bike lane, or to allow a wider sidewalk, or to accommodate street vendors, or for extra restaurant, bar, or cafe seating in good weather.  Even half a mile from a road, road noise intrudes; sitting in my childhood home, on 5 acres with trees and a wild pond and at least half a mile from any major road, there’s road noise, layered over the birds, frogs, and random insects.  

Auto and truck exhaust are also a large component of US greenhouse gas emissions, and it’s not falling very rapidly.  The oil needed to power them also pollutes in its supply chain; sometimes there are oil spills, old wells are often improperly capped, fracking contaminates surface and ground water, and even when everything is going well, about  30% of ocean shipping (with its own fuel consumption and pollution issues) is oil tankers.

Astonishingly, cars and trucks don’t even pay the full costs of their road use.  It is a relatively popular myth among drivers that they “pay for the roads” therefore they should have special rights to use the roads, but instead, for every dollars’ worth of road repair and construction, tolls, excise, and gas taxes pay only about 65 cents.  The rest comes from other taxes that everyone pays.

An economically efficient gas tax that considered all of the external costs of driving would be much larger.  Consider just early deaths from pollution (estimated at 53,000 deaths annually) and $7.5M for each life lost gives an annual cost of about $400B.  At annual gasoline and diesel consumption of about 200 billion gallons per year, that means the nuisance tax for just those pollution deaths (not counting any other disability, not counting global warming) would be 2 dollars per gallon.  This is a little over-simplified because the harm from pollution depends on how many people are exposed to it — for example, rural air pollution is less costly than urban air pollution because the population density is lower — but if the taxes in rural areas are lower, then the taxes in urban areas should be higher. There’s other ways to collect it; rather than a gas tax, it might be rolled into a congestion tax instead, so that the tax is specific to bringing a car into a place filled with people.  And this is just the tax for one harm caused by driving; there are others, and the taxes accumulate.

(*) The cited figure is 78%.  There’s a reddit thread that claims to debunk it, that thread is flaky. This report claims “78% of microplastics from 4 modeled sources”; they don’t include all sources but the limit of the debunk is “they didn’t measure everything” with no evidence that the rest is large.  IN PARTICULAR, that report used an expensive definition of the word “microplastic” which might lead you to believe that the plastic fragments are measured in micrometers (like PM10 and PM2.5), in fact, for ocean pollution purposes, “micro” is “5 MILLI meters or less”.  Nonetheless, even considered as the total mass of all ocean plastic, tire actual-micro-and-nano particles are estimated to make up 9% of that total; and because of their tiny size, are much more easily incorporated into the food web, and have much more larger surface area for chemical release.

Possible change to light timing at Garden, Huron, and Sherman. (I write letters…)

October 27, 2024

(to 311@cambridgema.gov)

Possible change to light timing at Garden, Huron, and Sherman.

I bike on Garden to and from work every day, I get plenty of chances to study this light’s behavior. Right now it has a sensor trigger for Garden traffic, and a button-requested all-ways pedestrian signal. The default timing when traffic is present is pedestrian traffic, Garden, Sherman, Huron from the south, and then Huron both directions. It’s relatively common for people bicycling (usually on Garden) to proceed during the pedestrian signal, usually slowly and avoiding pedestrians, but apparently not always; I’ve read some complaints on social media from people who live near there of bikes not yielding to them in the crosswalk. It is, on the other hand, slightly safer for bicycles to cross the intersection when no cars are in motion, so there’s a bit of a tradeoff there, even though the safer bicycle behavior is technically illegal.

I don’t know all the constraints on the light timing so the following suggestions might not actually be useful, but if possible, these might help with the bicycle-pedestrian conflicts.

Possible change #1: rearrange the light timing so that the light for Garden goes green before the pedestrian signal, not after.  That is, pedestrian, Sherman, Huron-from-the-south, Huron, Garden.

With the current timing bikes accumulate during the Sherman and Huron phases, so if there are bikes, there will usually be some waiting when the pedestrian phase begins, and some of those will run it, and some of the runners will not yield to pedestrians. With the rearranged timing, those bikes will go during the Garden phase and clear the intersection before the pedestrian signal. Some who arrive during the pedestrian cycle may still run their red, but there will be many fewer, and also arriving later in the phase after pedestrians were already in the crosswalks.

Possible change #2: legitimize the running, but try to control it. Five seconds after the pedestrian cycle begins, change the red bicycle signal to flashing red, for stop-then-go. Possibly also include a “yield to pedestrians” sign, possibly include a red yield triangle in the bike lane.

The hope here is that daily cyclists would learn they could cross legally after a short delay, and start to wait for the delay. This is somewhat speculative since, at least from what I see, most cyclists running the light are already doing it carefully and the ones who are not and create conflicts with pedestrians either don’t care or are clueless, and might not respond to this nudge. On the other hand, this will not make things worse, and will provide a small time and safety savings to careful cyclists, and legitimize careful cycling.

These two changes could be combined, though from a “does this work?” point of view that would muddy the waters if this was regarded as an experiment. I think the first choice would work better and cost less to implement.

I write letters…. (311requests@somervillema.gov)

October 26, 2024

The new bike light on Highland at Davis Square is hard to see and makes the intersection configuration more confusing; there is a risk that someone on a bike will not notice it, see the main light go green, and proceed into the path of a right-turning vehicle. The light is located to the right side of the intersection on the top of a post above other signs, not where someone would normally be looking as they approach the intersection, and not where they would notice it after stopping, if they didn’t know to look.

I have attached a photo, taken while waiting at the stop line (you can see a piece of it at the extreme lower right of the photo) to show how it might be missed. Someone waiting further forward (as suggested by the green paint for the advanced stop area on the left) would just not see it.

I think it would be better if the light were placed directly in the line-of-sight for people on bikes waiting at the light, preferably right at the line. The light at Inman for someone on a bike waiting to cross Beacon and continue on Cambridge towards Harvard is a good example; it is at face height, right there. That might require engineering and cost money, but I think it would be better. I attached a photo of that light, too.

I write letters… (311@Cambridge.gov)

October 25, 2024

In words, there is a parking-proteected bike lane on the north side of the street at 1575 Cambridge (and Dana), and it approaches a crosswalk. When the lane was installed (there’s a picture in Google Maps, I attached it) there was a small utility box just before (east of) the crosswalk, shorter than a person (though not shorter than a child).

Since then the utility box has been replaced, and the new one is larger, and it is about as tall as a person. It is now more than adequate to hide a pedestrian who wants to cross, and someone in the bike lane would not see them until perhaps 1.4 seconds before reaching the crosswalk. The attached pictures showing proposed changes were taken 1.4 seconds before reaching the crosswalk at 14mph, I have GPS-annotated video. Some people travel faster than 14mph.

I think this is a hazard waiting for just the wrong timing to result in a crash, and of course, everyone will blame the “irresponsible bicyclist”, instead of whoever designed this blind intersection. If nothing else, this email will put something in the records for later discovery.

This is the second time I’ve been caught like this, I have tried to train myself to swerve away from what I cannot see but humans make mistakes. This intersection should be fixed.

Proposed fix #1 is to move the crosswalk so that it starts before the box and is slightly diagonal to the corner. I think this would require moving a car parking space one slot down the road.

Proposed, lesser-but-easier, fix #2 is to shift the bike lane away from the curb, at minimum with paint, to give a clue that it would be a good idea to leave some room. This will work less well because not everyone will follow the shift and it’s not that much extra room.

I am not sure warning signage would be a net help; there’s already trees and signs and those boxes and the parked cars give poor visibility of people crossing from the Dana side.

I think it would also be helpful to review whatever processes allowed this hazard to be created.