I think traffic lights depend on the skill of the person setting them up. They’ve got a bunch of sensors (car waiting over magnetic sensor, visual sensor detects something other than road in magic area, pedestrian button pressed, time of day, timer since last transition, emergency vehicle override) as input and different intersection states as output.

Someone could program a cycle to just ignore all of the sensors and run through the various states on a timer. Or they could make a more complex cycle loop that lengthens the main state at night, switches sooner if a sensor is triggered, and tries to be smart about it. Or you could go for an even more complex statistical model that not only takes sensor states into account but tries to predict those sensors for even more accuracy.

My guess is that there’s great variance in both the skills of the person doing the programming and their managers and politicians/administrators calling the shots. And that variance in skill includes ones who don’t bother trying anything more complex than the defaults as well as those who do but aren’t good at it (eg: I’ve noticed that some intersections make their main cycle longer at night and don’t cycle unless someone is waiting but make them wait longer than they would during the day when night time means that interrupting the main cycle affects fewer people so they make people wait on the side streets for no good reason).

And there’s also the question of persistence. If they need to be reprogrammed any time the power goes out, they might just go with the easier route if they can’t restore backups.