Michigan sends drivers past their street, around a U-turn and back again hundreds of millions of times a year. We put the Michigan left, and 17 rival designs, through nearly 70,000 simulated traffic runs to price the detour against the crashes it prevents.

James Lyons-Weiler, PhD · November 4, 2026

If you live in Michigan, you are of course familiar with the “Michigan left.” You want to turn left, but the sign forbids it. So you drive straight through the intersection, slide into a U-turn lane in the median a few hundred feet later, swing around, come back and turn right onto the street you wanted in the first place. Lifelong Michiganders do it without thinking. Everyone else does it while thinking very hard.

And the debates rage inside cars. The driver insists it’s faster. The passenger is certain it wastes gas. The relative visiting from Ohio wants to know why anyone would build a road that makes you pass your turn on purpose. If you are from out of state, where such turns are rare, the first encounter can feel like a navigation error built into the pavement. Drivers who stop at the crossover and then catch a red light coming back have an especially concrete complaint: every extra start burns fuel, every queue burns time, and the forbidden direct left begins to look like the obvious bargain.

Mock Page from Michigan’s Driver’s Ed Manual. This insanity should be given to all out-of-state drivers.

As an aside, it could be worse. There’s also the Massachusetts left, where the driver making a left from a two-lane road onto a three- or four-lane road is expected to pull out in front of left-oncoming traffic to block traffic to claim the right to transit across lanes. This act of bravery or terror, which is culturally enforced by impatient drivers behind you and their horn, is a less civilized tactic compared to the Michigan left, which is a codified, sanctioned maneuver hard-wired into some intersections.

Satirical Page from Masschusetts Driver’s Ed Manual. This insanity is actually the norm in Massachusetts. Don’t take it personally; every person in the queue also expects themselves to take on the right-bound traffic.

The Michigan left design is not new. In 1960, Michigan state traffic engineers tested a directional crossover on Eight Mile Road, 660 feet west of Livernois Avenue in Detroit, to keep left-turning traffic out of a busy intersection (Michigan Highways). By 2010, FHWA counted more than 425 miles of boulevards in Michigan, with more than 700 directional crossovers on the Michigan Department of Transportation’s highway system alone (FHWA, 2010). No one has published a current count that includes county and city roads, so we used a working figure of about 1,000 individual U-turn openings, tested across 700 to 1,500.

Subscribe now

How many Michigan lefts do drivers actually make? No agency counts them, so we built the estimate in the open. If each of those 1,000 crossovers serves about 1,500 intended left turns a day, Michigan drivers complete roughly 550 million Michigan-left maneuvers a year. We tested 200 million and 1.1 billion as low and high cases. For scale, Michigan’s roads carried about 98.7 billion vehicle-miles in 2025 (MDOT, 2025). At an assumed 12.5 miles per trip and one and a half left turns per trip, that is roughly 12 billion left turns a year. On those assumptions, Michigan lefts would be about one left turn in 20. These numbers are assumptions, not measurements.

Then we asked what those turns cost. Our online Traffic Sim Lab runs a standard signalized intersection and a Michigan-left intersection side by side, sending the identical stream of cars through both. In its default scenario, each of the four approaches carries 400 vehicles an hour, a quarter of them turning left and a quarter turning right. Twenty matched runs were scaled to 550 million maneuvers a year. On that basis, the Michigan left added about $172 million a year in extra fuel and travel time for all affected traffic, in 2025 dollars, valuing time at $20 an hour and gasoline at $3.50 a gallon. Most of that, $134 million, was time; $38 million was fuel.

The standard intersection (A) and the Michigan left (B) running side by side in the IPAK-EDU Traffic Sim Lab, with the same cars arriving at both. Gold vehicles intend to turn left; the dotted line traces one left-turning driver’s route, including the Michigan left’s U-turn. Screenshot from ipakedu.online.

Try it yourself: ipakedu.online members can run the simulator at https://ipakedu.online/traffic-sim/. Change the traffic, turn shares, lanes and signal settings, then see what the Michigan left costs and saves under your own assumptions.

The other side of the ledger is crashes. Using FHWA’s legacy crash rates (1.644 crashes per million entering vehicles for conventional intersections and 1.388 for median U-turn intersections), the same scaled scenario produced about 564 fewer crashes a year: 3,058 instead of 3,622. The change was not uniform. Opposing direct-left crashes fell from 430 to 81 and angle/turning crashes from 1,099 to 580, while rear-end crashes rose from 1,856 to 2,152.

Illustrative evaluation. One traffic pattern scaled to an assumed statewide volume; positive means the Michigan left costs more. Seed variation on the operating cost is about ±$17 million. Economic crash costs cover medical bills, lost productivity and property damage; comprehensive costs add monetized quality-adjusted life years. The comprehensive line values the head-on/object/overturn group at its economic cost because no comprehensive figure was available for it. At 200 million and 1.1 billion maneuvers, the economic-valuation net is +$48.7 million and +$267.6 million. Source: Traffic Sim Lab, ipakedu.online.

So which is it, an expensive detour or a bargain? Price crashes only by their economic losses, and the Michigan left costs Michigan about $134 million a year more than it saves. Each avoided crash would have to be worth about $305,000 in economic losses to break even. Price crashes the way FHWA does when it includes lost quality of life, and the verdict reverses to a modest net saving of roughly $20 million. That is smaller than the uncertainty in the inputs. The answer to the argument in the car depends less on the U-turn than on the price we put on an injury.

That single scenario also hides most of the interesting questions. Traffic is not always 400 cars an hour from every direction. The Michigan left is not the only alternative to a conventional left turn. A standard intersection can run its left turns in several different ways. So we held a playoff.

Share

The playoff

The people driving straight through a Michigan-left intersection may experience a different bargain from the driver making the U-turn. Removing direct left turns from the main intersection simplifies its signal: FHWA notes that much of the efficiency of a median U-turn intersection comes from its two-phase signal operation (FHWA, 2010). An arrangement that lengthens one driver’s route can improve the movement of everyone else. The comparison therefore has to follow the entire maneuver and account for everyone affected.

We designed an exploratory simulation tournament to examine that tradeoff. It began with Michigan lefts and conventional intersections and expanded to 18 designs in two separate comparison groups. Across screening, tuning, qualification and final evaluation, the study analyzed 69,896 traffic simulations. Its central finding is conditional but clear: a conventional intersection combining protected and permissive left turns had the lowest average modeled cost across six equally weighted traffic regimes. Full displaced left was close behind. Michigan left won one regime and encountered completion problems in two others.

How the playoff worked: six selection stages advanced tuned configurations of all 18 designs into 108 championship seats, followed by the six heat winners and the overall winner. Traffic Sim Lab v3.1, executed playoff protocol.

The leading difference was about 1.7 cents per entering vehicle. Several reasonable changes to the comparison reversed the order of the top two. That is the scale of the result we have to explain before anyone turns a simulation ranking into a recommendation for a road.

The familiar green light conceals an important distinction. Under a permissive signal, you may turn left during a green light after yielding to oncoming traffic and pedestrians. The signal permits the turn but gives you no exclusive right of way.

At a four-way intersection:

1. Your light turns green, and opposing traffic also has green.

2. Through traffic proceeds.

3. You wait for a safe gap, then turn left across the opposing lanes.

That is how most of the rest of the US understands left-hand turns.

A dedicated left-turn pocket keeps waiting vehicles out of the through lanes, but the pocket alone does not make the turn protected (FHWA, 2008, section 4.3).

Permissive operation works efficiently when gaps are plentiful; heavy opposing traffic can leave left-turners waiting through repeated cycles. The overall winner in our model was the combined arrangement, protected/permissive.

Our alternatives included those signal arrangements, split phasing, stop-controlled intersections, roundabouts, jughandles, quadrant roadways and restricted-crossing layouts. A displaced-left design moves the crossing of opposing traffic upstream, allowing different movements to proceed together at the central intersection. Conventional and diverging-diamond interchanges entered a separate group because their traffic movements differ from those of an ordinary four-way junction. Mixing them into a single ranking would obscure that difference.

We varied lane counts from one to four and included optional pockets where the geometry supported them. The main championship held through lanes at two per approach. Auxiliary lanes and pavement still differed among designs, so this was not an equal-construction-budget contest. Separate Michigan-left searches varied crossover distance, U-turn lanes and storage bays, signal control, timing and red-turn permissions.

Six traffic patterns tested different demands on those designs. Balanced flows ranged from 600 to 4,400 entering vehicles per hour. Two asymmetric cases sent most traffic along one road or the other, and a left-turn-heavy case assigned 40 percent of entering vehicles an intended left turn. The principal comparison assumed 35 mph, 10 percent trucks and no pedestrian arrivals. Right-on-red and U-turn left-on-red were disabled in the main championship and examined separately.

Each final configuration faced 100 fresh replications with a one-hour arrival observation window, after a warm-up period, followed by time for vehicles to finish. Competing designs received matched traffic demand. Settings were selected before the final evaluation, using different random seeds, so a configuration could not win merely because it had been tuned to the exact arrivals used to judge it.

The cost calculation valued lost vehicle time at $20 per hour and fuel at $3.50 per gallon, then added expected crash loss. These were stated modeling assumptions, expressed in 2024 dollars (the online lab figures above are in 2025 dollars). Lost time was measured against a common 1,200-meter trip at 35 mph, which meant the additional Michigan-left route remained chargeable. Fuel covered the full modeled trip, including idling and repeated acceleration. A driver making several component turns still counted as one entering vehicle.

That denominator matters. Our headline figures are costs per entering vehicle, including through and right-turning traffic. We also report time and fuel for drivers intending to turn left. Dividing the entire intersection’s cost by its number of left turns produces an allocation of network costs; it does not identify the extra cost caused by one additional left-turner. A claim about the “cost of a turn” must say which of these quantities it means.

Crashes required a second distinction. We modeled rear-end, opposing direct-left, angle/turning, sideswipe/merge and a residual head-on/object/overturn group. Published rates and type distributions supplied the expected frequencies. Michigan’s statewide 2024 crash tables supplied severity proxies, while FHWA supplied monetary values. These sources describe different populations and geographic boundaries; combining them is an assumption to investigate, not a substitute for observations at the intersections being modeled (Michigan Traffic Crash Facts, 2024).

The historical FHWA synthesis rates quoted above, 1.644 crashes per million entering vehicles for conventional intersections and 1.388 for median U-turn intersections, were provisional inputs here as well. They are not new statewide estimates produced by our simulation and applying them to our full modeled routes requires validation (FHWA, 2007, table 3).

The distinction between economic and comprehensive crash costs proved decisive here too. Economic costs cover losses such as property damage and lost productivity. Comprehensive costs also include monetized losses in quality-adjusted life years. FHWA’s 2025 update assigns a fatal crash approximately $2.24 million in economic costs and $15.99 million in comprehensive costs, in 2024 dollars. These are per-crash valuations classified by the most severe injury, rather than a payment to an individual. The comprehensive figure includes the economic component; the two should not be added together (FHWA, 2025, table 1).

We used expected crash losses for the rankings. If a short run happens to draw no crash, that does not make its underlying risk zero. An independent incident audit sampled 15,600 frozen traffic trials and drew 63 simulated incidents against 53.8 expected. Each affected trial was replayed with a temporary lane or junction closure. The draw included 36 rear-end incidents, 15 angle/turning incidents, seven opposing direct-left incidents, one sideswipe/merge incident and four in the residual group. These are generated events, not road observations.

The closure replays let us examine queue recovery. Their additional delays were reported separately from the main cost score because FHWA’s crash-cost framework already includes congestion costs. Adding both without reconciling their coverage would risk counting part of the same loss twice (Harmon, Bahar and Gross, 2018).

The leading results under comprehensive crash valuation were:

Equal weight to six traffic regimes; two through lanes per approach. Costs include lost time, fuel and expected comprehensive crash losses, in 2024 dollars. Source: IPAK-EDU Traffic Bake-off v3.1 research package (unpublished).

Full displaced left saved about ten seconds per entering vehicle relative to protected/permissive signaling yet finished second on total cost. Its assumed crash losses were higher. The main safety scenario used injury and property-damage factors from a 2020 displaced-left study catalogued by FHWA’s Crash Modification Factors Clearinghouse, which rates those factors two stars out of five. Transferring them to both our full and partial designs is consequential, and their applicability remains uncertain (Abdelrahman et al., 2020). A later North Carolina study of 16 converted intersections, using the empirical Bayes method that corrects for regression to the mean, found the opposite direction: about 14 percent fewer injury crashes and 29 percent fewer angle crashes after conversion (Cunningham et al., 2022). We have not yet rerun the championship with those factors. Given how narrow the lead was, they could change the winner.

Using economic crash costs alone moved full displaced left into first place at 45.8 cents per entering vehicle. Assigning the same safety risk to every design also made full displaced left the winner. So did weighting the six regimes by their traffic volumes instead of giving every regime equal weight. Even that volume weighting is only another scenario; we have not measured how frequently these conditions occur across Michigan.

The aggregate champion did not win any individual heat. Multilane roundabout won the quiet balanced case; permissive signaling won the busy balanced, arterial-heavy and left-turn-heavy cases; full displaced left won the overloaded case; and Michigan left won the crossroad-heavy case. Protected/permissive signaling earned first place through its average performance across the complete set. The choice of an aggregate score is itself part of the research question.

Michigan left forced us to examine a more basic problem before calculating an average: which vehicles had actually finished? Early short runs selected settings that developed queue problems under sustained demand. We added an endurance stage covering 408 candidates, each tested over 20 fresh full-hour replications, then froze new selections for a separate championship. The earlier results remain in the record.

In that championship, the selected two-lane Michigan left cleared every observed entrant in 99 of 100 busy trials and 37 of 100 overloaded trials. The rule required completion in every replication, so it received no overall six-regime monetary rank. Averaging only drivers who escaped the queue would reward a design for leaving its longest-waiting drivers out of the calculation. An unfinished simulation does not establish permanent gridlock on a real road: storage assumptions, control logic and the allowed clearance time all need examination.

On the four regimes it completed, Michigan left averaged 65.3 cents per entering vehicle and ranked sixth among designs eligible for that narrower comparison. Permissive signaling averaged 56.3 cents and protected/permissive signaling 59.0 cents. Full displaced left averaged 66.0 cents. These figures describe a subset selected by Michigan left’s successful completion; they cannot replace the six-regime ranking.

Michigan left’s winning crossroad-heavy configuration used two through lanes, crossovers 500 feet from the central junction, two U-turn lanes with separate bays, yield-controlled crossovers and an actuated central signal that responded to traffic demand. It cost 54.7 cents per entering vehicle and completed all 100 trials. Drivers intending to turn left nevertheless lost an average of 92.3 seconds relative to the common free-flow reference, with 81.5 cents in time and fuel costs. A network winner can still impose a substantial burden on a particular movement, and that burden is exactly what the driver in the car is complaining about.

The crossover corridor was fixed on the east-west axis. Rotating it was not part of this search, so the contrast between arterial-heavy and crossroad-heavy demand also identifies an orientation question for the next study.

A separate overloaded-traffic candidate with four through lanes, crossovers 800 feet away and signalized two-lane U-turn bays completed all 100 trials at 73.0 cents per entering vehicle. Its actuated controllers used minimum and maximum greens of seven and 35 seconds and a two-second passage setting governing how arriving vehicles could extend green. That is a useful candidate for further testing, but extra pavement had no construction, land or maintenance price in this tournament. A wider road can win an operating-cost comparison while losing a project-budget comparison.

The strongest limit on the safety conclusions lies inside the model. Crash hazards were assigned from external profiles; they did not emerge from vehicles actually colliding. In the main scenario, a longer queue did not automatically raise rear-end risk, and permitting turns on red did not automatically alter crash risk. Several designs and signal modes received the same baseline risk because a verified, applicable difference had not been established. Those equalities are placeholders. They can favor one contender over another, particularly when the operating-cost gap is small.

This means we have identified operating configurations under assumed safety conditions. We have not identified the safest Michigan-left configuration. Nor does a narrow interval across repeated traffic simulations resolve uncertainty in crash rates, driver behavior or intersection geometry. Increasing the number of runs improves our estimate of what this model does; validating its assumptions determines how much that estimate tells us about a road.

The next measurements are specific: observed turning volumes, queue lengths and travel times; crossover and central-signal coordination; crash locations and severity across the entire indirect route; pedestrian and cyclist movements; and the cost of land and construction. Those measurements would let us replace the assumptions most capable of changing the result. That is why the statewide figures at the top of this article are labeled illustrative. They show the scale of what is at stake and how the verdict flips with the price of an injury. They are not a forecast, and multiplying a difference of a few cents by an assumed statewide number of turns cannot make them one.

The source code, settings, failed trials, crash assumptions and results are preserved in the research package, and the simulator itself is open to members at ipakedu.online. Together they make it possible to inspect why a design won, change an assumption and determine whether it still wins. That is the standard this series will apply to its own conclusions. When a driver is sent past the desired street and back again, the public deserves an accounting that follows that driver through the entire trip, counts the people helped as well as those delayed, and makes the price assigned to injury visible.

Leave a comment

Subscribe now

Share

Sources and research files

1. Traffic Sim Lab (2026). Michigan left vs. standard intersection simulator, with annual cost and crash-cost report. IPAK-EDU, ipakedu.online. Statewide exposure assumptions and their derivation are in the lab’s annual-methods notes.

2. Traffic Bake-off v3.1 With Crashes (2026). Research package: methods, frozen configurations, seeds, failed trials, results and sensitivity analyses (REPORT.html, METHODS.txt).

3. Michigan Highways. The Michigan Left. History of the 1960 Eight Mile Road experiment, citing S. Lingeman’s 1996 history.

4. Federal Highway Administration (2010). Alternative Intersections/Interchanges: Informational Report, chapter 3: Median U-Turn Intersection. FHWA-HRT-09-060.

5. Michigan Department of Transportation (2025). Statewide roadway statistics (vehicle-miles traveled).

6. AAA Foundation for Traffic Safety (2026). American Driving Survey, 2024. Average trip length.

7. Federal Highway Administration (2008). Traffic Signal Timing Manual, chapter 4, section 4.3. Archived manual, cited for signal-mode definitions.

8. Michigan Traffic Crash Facts (2024). Statewide crash tables, crash type by injury severity, PDF page 5.

9. Federal Highway Administration (2007). Synthesis of the Median U-Turn Intersection Treatment, Safety, and Operational Benefits. FHWA-HRT-07-033, tables 3–4.

10. Federal Highway Administration (2025). Updated Crash Costs for Highway Safety Analysis. FHWA-SA-25-021, table 1, 2024-dollar crash-unit costs.

11. Harmon, T., Bahar, G., and Gross, F. (2018). Crash Costs for Highway Safety Analysis. FHWA-SA-17-071.

12. Abdelrahman, A., Abdel-Aty, M., Lee, J., Yue, L., Abdelrahman, A., and Al-Omari, M.M.A. (2020). “Evaluation of displaced left-turn intersections.” Transportation Engineering, 1, 100006. doi:10.1016/j.treng.2020.100006. Factors verified against the FHWA CMF Clearinghouse, study 630.

13. Cunningham, C., Lee, T., Saleem, T., and Srinivasan, R. (2022). Development of a Crash Modification Factor for Conversion of a Conventional Signalized Intersection to a CFI. North Carolina DOT, FHWA/NC/2020-29.

Appendix: the six traffic cases in detail

Each case is a one-hour traffic pattern applied identically to every design. All six share the same fixed conditions: two through lanes per approach, 35 mph, 10 percent trucks, 10 percent right turns, no pedestrians, and no turns on red. What changes is how much traffic arrives, how it divides between the two roads, and how many drivers want to turn left. In every case, east–west is the road that carries the Michigan-left crossovers.

In the table, E–W and N–S are vehicles per hour on each approach of that road, and Total is all vehicles entering per hour. Costs are per entering vehicle, in 2024 dollars. They include lost time at $20 an hour, fuel at $3.50 a gallon and expected comprehensive crash loss. “Lost time” is the average delay per entering vehicle compared with an uninterrupted 1,200-meter trip at 35 mph. A design is ranked in a case only if every observed vehicle finished in all 100 final trials. Interchanges are reported separately because their traffic movements differ from a four-way intersection.

1. Quiet balanced

What it is: light, even traffic: 150 vehicles an hour from each direction, 600 in total, with 15 percent turning left. Think of an off-peak hour, or a suburban crossing that never gets busy.

What happened: with plenty of gaps, almost everything worked, and the ranking was decided mostly by crash risk.

• Ranking: the multilane roundabout won at $0.385 (16.7 seconds lost), just ahead of the single-lane roundabout at $0.388 (17.3 seconds). Their assumed crash risk was about half that of the signalized designs, and that was the margin of victory.

• Fastest but not cheapest: the two-way stop was the fastest design, losing only 8.4 seconds per vehicle. It finished third at $0.445 because it carried the higher crash cost. If crashes are valued only by their economic losses, the two-way stop becomes the cheapest design in this case.

• Michigan left: fourth at $0.462 (22.1 seconds). Its left-turning drivers lost an average of 70.2 seconds each, against 15.9 seconds under a permissive signal.

• Others: protected/permissive, the overall champion, placed only eighth ($0.508). RCUT was last ($0.716, 50.6 seconds).

• Interchanges: diverging diamond $0.407; conventional diamond $0.553.

2. Busy balanced

What it is: heavy, even traffic: 600 vehicles an hour from each direction, 2,400 in total, with 20 percent turning left. Think of a busy suburban arterial crossing at rush hour.

What happened: signals took over, and the designs that depend on drivers finding gaps began to fail.

• Ranking: the permissive signal won at $0.580 (29.7 seconds lost). Next came the near-side jughandle ($0.630), protected/permissive ($0.635) and protected-only ($0.637). Both displaced-left designs placed sixth and seventh ($0.683 and $0.687), despite losing less time than most.

• Breakdowns: the stop-controlled designs collapsed. All-way stop cost $3.551 with 529 seconds lost; two-way stop cost $9.229 with about 25 minutes lost per vehicle. RCUT also struggled at $1.625 (197 seconds).

• Unranked: the Michigan left cleared every vehicle in 99 of 100 trials and so went unranked. Both standard roundabouts cleared every vehicle in 98 of 100 and were also unranked.

• Interchanges: diverging diamond $0.512; conventional diamond $0.720.

3. Overloaded balanced

What it is: 1,100 vehicles an hour from each direction, 4,400 in total, with 20 percent turning left. This is a stress test, beyond what an ordinary two-lane intersection is designed to carry.

What happened: this case separated the designs most sharply. Only designs that remove left turns from the main signal kept delay within reason.

• Ranking: full displaced left won at $0.801 (50.1 seconds), followed by partial displaced left ($0.847) and the quadrant roadway ($0.927). Protected/permissive held fourth at $1.030 (102 seconds).

• Collapse: after that, costs climbed steeply. Protected-only cost $1.762, split phasing $3.905 and the permissive signal $6.256, with almost 17 minutes lost per vehicle. The single-lane roundabout finished last of the ranked designs at $20.91, with nearly an hour lost per vehicle.

• Michigan left: it cleared every vehicle in only 37 of 100 trials, so it went unranked. The multilane roundabout managed 95 of 100. A separate four-lane Michigan-left candidate did complete all 100 trials ($0.730), but it used twice the road width.

• Interchanges: the ranking reversed here. Conventional diamond $2.312; diverging diamond $4.101.

4. Arterial-heavy

What it is: a busy main road crossing a quiet side street. East–west carries 900 vehicles an hour per approach and north–south 250, 2,300 in total, with 20 percent turning left. This is the textbook setting for a Michigan left: a heavy divided boulevard with its crossovers on the busy road.

What happened: the textbook setting did not favor the Michigan left in this model.

• Ranking: the permissive signal won at $0.575 (29.0 seconds), followed by protected/permissive ($0.613), protected-only ($0.622) and the near-side jughandle ($0.624).

• Michigan left: tenth of 15 at $0.720 (58.4 seconds). Its left-turning drivers lost almost two minutes each (119 seconds), because their U-turns had to be made into, and back across, the heavy boulevard traffic. RCUT was eleventh ($0.834).

• Roundabouts: they could not cope with the unbalanced flow. The single-lane roundabout cost $4.602 and the multilane $7.635, the latter with almost 21 minutes lost per vehicle.

• Interchanges: diverging diamond $0.544; conventional diamond $0.711.

5. Crossroad-heavy

What it is: the mirror image of case 4. The busy road runs north–south at 900 vehicles an hour per approach and the quiet road runs east–west at 250, 2,300 in total, with 20 percent turning left. The Michigan-left crossovers therefore sit on the quiet road.

What happened: this was the Michigan left’s best case and its only heat win.

• Ranking: the Michigan left won at $0.547 (34.5 seconds), ahead of protected/permissive ($0.613), protected-only ($0.623) and the permissive signal ($0.629). The winning setup used crossovers 500 feet from the junction, two U-turn lanes, yield-controlled crossovers and an actuated central signal.

• Left-turner burden: the win came with a large cost to left-turners, who lost 92.3 seconds each, against 38.0 seconds under protected/permissive.

• A plausible reading: the U-turns happened on the quiet road, where gaps were easy to find. Removing left turns from the main signal let the busy north–south road keep more green time. Comparing cases 4 and 5 suggests the orientation of the crossovers matters a great deal. This study did not test rotating them.

• Others: RCUT ($3.937) and both roundabouts ($4.606 and $7.412) finished last. The turbo roundabout exhibition entry cleared every vehicle in 99 of 100 trials.

• Interchanges: diverging diamond $0.486; conventional diamond $0.717.

6. Left-turn-heavy

What it is: 600 vehicles an hour per approach east–west and 400 north–south, 2,000 in total, with 40 percent turning left, twice the share in the other busy cases. Think of an intersection next to a shopping center, school or employer where most drivers turn.

What happened: surprisingly, the simplest signal won again.

• Ranking: the permissive signal won at $0.574 (28.3 seconds), followed by protected/permissive ($0.625), split phasing ($0.646) and protected-only ($0.670). With two through lanes and left-turn pockets, opposing traffic at these volumes still left enough gaps for permissive turns. That margin disappears in the overloaded case.

• Indirect designs: they paid for their detours when so many drivers turn. The Michigan left was tenth at $0.885 (82.8 seconds), with left-turners losing 132.5 seconds each; RCUT was eleventh ($0.942).

• Roundabouts: the single-lane roundabout cost $3.125. The multilane (93 of 100) and turbo (94 of 100) roundabouts did not clear every vehicle and went unranked.

• Interchanges: diverging diamond $0.480; conventional diamond $0.735.

What the six cases show together

• No universal winner: no design won more than three of the six cases, and the overall champion won none.

• Permissive signal: won three cases outright but collapsed under overload, finishing seventh overall.

• Protected/permissive: never won a case but was never far behind, which is why its six-case average was the lowest.

• Michigan left: depends heavily on which road carries the crossovers. It was first when they sat on the quiet road and tenth when they sat on the busy one. It also failed to clear every vehicle at busy and overloaded balanced volumes.

• Roundabouts: excellent at light volumes and among the worst at heavy or unbalanced volumes, at least in this model’s geometry.

• Displaced left: the clear choice when the intersection is overloaded.

 

IPAK-EDU is grateful to Popular Rationalism as this piece was originally published there and is included in this news feed with mutual agreement. Read More