Road diet · operational demonstration
On a four-lane undivided street, a driver waiting to turn left into a driveway or parking space stops in a travel lane. Everyone behind them either stops too or has to weave around them. Add a center turn lane and that vehicle steps out of the way entirely, and traffic flows smoothly and predictably. This toy simulation illustrates that difference.
Safety is the reason to do a road diet. The FHWA recognizes 4-to-3 conversions as a Proven Safety Countermeasure: narrower crossings, lower speeds, fewer conflict points. Everything below is about traffic operations instead, because “it will cause gridlock” is the objection a diet has to survive. It is the rebuttal, not the reason.
An illustration of one mechanism, not a traffic study — no calibration to observed counts, no pedestrians, parking or side streets, one idealized block. Order of magnitude, not forecast. Full method and caveats below.
Watch it
Both roads carry identical demand — the same arrival times, the same drivers, the same turns. The only difference is where a left-turning vehicle waits.
Why it is safer
The three-lane road is monotonous: the same trip, every trip, whatever the traffic is doing. That is not just a nicer drive — it is the same property that makes it the safer road. A driver who is never surprised by a stopped car in a live travel lane never has to make the sudden decision that a late brake or a swerve into the next lane is, and those are precisely the maneuvers behind the rear-end and sideswipe crashes the conversion removes.
The four-lane road's variability and its crash record are two readings of one thing. Everything below measures the variability, because that is what a simulation can measure. The crash record is what it is evidence for.
Traffic is released from a signal at each end, which is why it arrives in platoons rather than trickling in. Each signal shows its indication only on the half of the road it releases, and nothing is held on the way out, so the block is measured on its own.
Every replay starts from the same place. The simulation opens with an empty street and the signal on red; traffic stacks up at the stop bar for one full red, and the clock starts at the instant that first green turns. So the block is clear at 0:00 in every sample, at every volume, on both roads — three minutes, two full signal cycles, released at the same moment in both directions and on both roads.
A vehicle that is going to turn left is amber for its whole trip, from the stop bar to the driveway, so you can see which car is about to cause the blockage before it does, and where it sits in the platoon. The traffic sample control switches between twelve independent hours — the figures further down average all twelve, but the replay always shows the one you choose.
At 700 through vehicles per hour, each direction
Averages over twelve independent simulated hours, so no single unlucky hour drives a conclusion. 700 vehicles per hour going straight in each direction, plus the 10% that turn left — about 800 vehicles per hour per direction, roughly 1,600 two-way, with the signals giving the mainline 60% of the cycle. The replay above is separate and deliberately fixed: it always shows whichever of the twelve sample hours you pick, and never reshuffles.
The 700 figure is not arbitrary. On the corridor this was built for, the busiest intersections — Alden, Hommocks and Richbell — carry roughly 700 through vehicles in their PM peak. So the 700 column is the realistic busy case, and the 400 and 550 columns are the rest of the day.
The single busiest movement anywhere on the corridor is 831 vehicles, southbound in the PM peak at Delancey Avenue. That is what the 850 column stands in for — deliberately a little above the worst observed number, so the comparison does not rest on the corridor's quietest assumptions.
Delancey is worth naming for a second reason: NYSDOT has identified the Delancey/Orienta intersection as problematic, and it is one of the reasons the department is exploring a roundabout there.
The first two tiles are, to any driver, a tie. The same number of vehicles get through, and the road diet's one-second edge on the average trip is nothing — if you stood at the curb with a stopwatch and timed a random car you would not be able to tell the two roads apart. Capacity is not the trade — at least not at a signal timing this street could plausibly have.
What differs is the spread. Take the gap between the typical trip and the unlucky one, and watch what happens as the road fills:
How much worse the 95th-percentile trip is than the average trip, in seconds.
The three-lane road gets more predictable as it fills — its spread barely moves, 2.2 seconds to 1.9. The four-lane road's more than triples, from 4.9 to 17.1, because the busier the street the harder it is to get around somebody waiting to turn. Both roads are carrying the same traffic at the same average speed the whole way. One of them delivers that average to almost everybody; the other delivers it on average while handing one driver in six something much worse.
That is the honest shape of this trade, and it is a better argument than a speed claim, because it is the delay people actually notice. Nobody complains about a trip that takes the time it usually takes.
Traffic is released from a signal but not held at the far end, so these figures are pure block traversal with no signal delay folded in. That is the cleaner measurement, and it is conservative: on a real corridor the driver who loses thirteen seconds mid-block is the one who arrives at the next signal just after it turns red, and pays for it again. This page does not count that.
Across the volume range
At light volumes the two cross-sections are nearly identical — a left-turner blocks a lane, but there is room to get around and time to catch the green. As volume rises, escaping gets harder and the green gets tighter, and the four-lane road's worst trips come apart.
At 850 veh/h the three-lane travel times are measured on the vehicles that got through, not on the ones still queued upstream — read that column with the served-volume row.
What the diet needs from the signals
Everything above assumes the arterial gets 60% of the cycle — 54 seconds of green out of 90 — an ordinary split for a street like this one. That assumption is doing real work, and it is the one to check before promising anything, because it matters far more than the cycle length does.
A single through lane discharging at a two-second saturation headway carries about 1,800 veh/h while it has green. Multiply by the split and you get the ceiling: roughly 1,080 veh/h at 60% green, 1,170 at 65%, but only about 900 at half the cycle. The four-lane road, with two lanes at the stop bar, has twice that headroom and never notices.
Holding demand at 850 through vehicles per hour and varying only the green split:
It is set by how much green the arterial gets, and arterials are routinely given more than half the cycle. At the 60% assumed here the single through lane serves 858 vehicles an hour — the whole demand, and everything the four-lane road manages — with travel times two seconds better. It is only at half the cycle that the ceiling starts to bite.
The model uses 60%. Running the same sweep at the more generous 65% is the check worth having, because someone will ask whether the result depends on a favorable split. Both cross-sections serve their full demand at either setting, 850 included, with a residual queue of under ten vehicles behind the stop bar. Everyone's trip is about a third of a second shorter at 65%, on both roads. What changes is the spread:
| Gap between the unlucky trip and the average trip | 400 | 550 | 700 | 850 |
|---|---|---|---|---|
| 4 lanes, 60% green (this model) | +4.9 s | +9.3 s | +13.1 s | +17.1 s |
| 4 lanes, 65% green | +3.9 s | +8.6 s | +12.1 s | +15.8 s |
| 3 lanes, 60% green (this model) | +2.2 s | +2.2 s | +2.1 s | +1.9 s |
| 3 lanes, 65% green | +2.3 s | +2.3 s | +2.3 s | +2.1 s |
Going the other way, to 65%, would shrink the four-lane road's spread by about a second at every volume and grow the three-lane road's slightly. A tighter green packs the platoon more densely, which gives a blocked driver less room to get around a left-turner — and gives a road with nothing blocking it nothing to notice. So the case does not depend on a generous split: between 60% and 65% the diet is the more reliable road either way, and by more at the 60% this model actually assumes.
Holding the split at 60% and shortening the cycle from 90 s to 60 s cuts the four-lane road's spread by about a quarter and the three-lane road's by rather less, so it narrows the gap between them. Across this range of volumes both cross-sections serve the same traffic either way; only the spread moves:
| Gap between the unlucky trip and the average trip | 400 | 550 | 700 | 850 |
|---|---|---|---|---|
| 4 lanes, 90 s cycle (this model) | +4.9 s | +9.3 s | +13.1 s | +17.1 s |
| 4 lanes, 60 s cycle | +3.1 s | +7.2 s | +10.2 s | +13.7 s |
| 3 lanes, 90 s cycle (this model) | +2.2 s | +2.2 s | +2.1 s | +1.9 s |
| 3 lanes, 60 s cycle | +1.9 s | +1.7 s | +1.7 s | +1.6 s |
The reason is platoon size. A 90-second cycle releases a bigger burst after a longer red, so more drivers arrive behind any one left-turner at once and the queue behind it is longer; a 60-second cycle spreads the same hourly volume into smaller, more frequent groups, which leaves a blocked driver more gaps to escape into. Shortening the cycle to 60 s would drop drivers held up behind a left turn from 11.0% to 7.2% at 700 veh/h, and from 17.3% to 12.3% at 850.
Retiming to a shorter cycle is a real partial fix for the four-lane road's problem, and someone will point that out. It does not close the gap — at 700 veh/h the four-lane road would still run +10.2 s against +1.7 s — and it buys nothing on the safety side, which is the main case. It also has to survive the pedestrian crossing and side-street clearance times a 60-second cycle leaves, which is a question for the corridor's own timing plan rather than for this model.
This is still the question to bring to the corridor's actual signal timing. A single through lane discharging at a two-second headway carries 1,800 veh/h × the split: 1,170 at 65%, 1,080 at 60%, 900 at 50%. Demand at the 850 level is about 945 veh/h including the turners — under the first two, over the third. That is why 50% is where it breaks and 60% is not: at half the cycle the single through lane serves only about 754 of the 850, and the queue upstream never comes down.
Method
1,000 ft of straight roadway at 35 mph between two signals, with three unsignalized driveways per direction at 250 / 500 / 750 ft, offset side to side as real driveways are. No pedestrians, no parking, no buses — a toy, deliberately.
Vehicles arrive at the upstream signal at random, queue on red, and discharge on green at a two-second saturation headway per lane — so they enter the block in platoons, not a trickle. 10% turn left, split evenly across that direction's three driveways. Both directions carry the same volume, because a left-turner's wait depends on the oncoming stream.
90-second cycle, 54 s of effective green for the mainline (60%), both signals running together. Each releases one direction into the block; neither holds traffic on the way out, so travel time is pure block traversal, and the two directions are symmetric.
Free-flow speed is 35 mph, the same on both cross-sections — the model gives the road diet no speed-reduction credit, even though real conversions usually deliver one. Nobody achieves it from a standing start: the block averages 27.1 mph on four lanes and 28.2 on three at 700 veh/h, against 19.5 s for a car that never left 35.
A lane change counts as forced when the driver is in the inside lane, is being slowed, and there is a left-turner within 200 ft ahead in that lane that has dropped below 13 mph. The turner need not be the immediate leader: once one stops in a travel lane the whole queue behind it is stuck, and the fourth driver back is as stuck as the first.
Intelligent Driver Model car-following, MOBIL lane changing, HCM critical-gap acceptance for the left turns. A lane change takes three seconds, during part of which the vehicle intrudes into both lanes and holds up followers in each — weaving costs the traffic around it, not just the driver doing it.
Left-turners move into the center lane 200 ft ahead of their driveway. If an opposing left-turner already occupies that stretch of it they cannot, and they hold up the through lane exactly as they would on four lanes — that penalty is in the numbers, and it is why the three-lane "held up" column is not a perfect zero at every volume.
Every figure averages twelve independent simulated hours, and all twelve are stored in this page — the traffic sample control replays whichever you pick. Doubling from six hours to twelve moved every headline number by less than a second: average travel time by at most 0.08 s, and the four-lane spread at 700 from 11.1 to 11.7 s. Single hours vary far more than that — the four-lane spread at 700 ranges from 8.8 to 14.9 s across the twelve, which is exactly why nothing here rests on one of them.
Operations, not safety — the 19–47% crash reduction at the top comes from field studies of real conversions, not from simulation, though the weaving and left-turn blockage counted here are the mechanism behind a good share of it. It also assumes the signalized approaches keep their capacity; a diet that narrows the intersection approach itself would add delay this does not show.