I've discussed how transit use can be a feedback loop. Like any product, if people use it and like it they'll buy more, and tell their friends. This drives up demand, which can allow the provider to raise fares and finance expansion. Public transit operators, with their social missions, often have difficulty raising fares, but high demand can provide political support for more subsidies, allowing expansion of the system.
The feedback loop can also operate to the detriment of transit. If people use and like the highways, they'll support highway subsidies instead of transit subsidies, and it's the highway system that will expand.
This has the potential to work in transit's favor, since it's more efficient at providing access to what people need, even though the cost of labor in driving is borne by the traveler. If we assume that the subsidy dollars are divided up in proportion to population, this actually benefits transit users because they get more bang for the buck.
Let's say that in the year 2000 there are 99 drivers and 1 transit user (the famous "bus with just one rider"). They each pay $10 in taxes, of which $990 is spent on roads and $10 on transit. But if each transit dollar is twice as efficient as a road dollar, that's like spending $980 on roads and $20 on transit. That means that transit improves, and let's say that in 2001 one of the 99 drivers decides to switch to transit. So the next year there will be $980 for roads and $20 for transit. But that $20 is worth $40, which attracts another two former drivers.
Eventually it reaches the tipping point, and soon enough there's just one lonely driver!
You may object to the assumptions, but the math is sound. As long as (a) subsidies are distributed based on usage, (b) transit is more efficient at providing access than roads are, and (c) people patronize systems depending on the access that they provide. This is true whether transit is twice as efficient at providing access or 101% as efficient, as long as it's more efficient.
Now, obviously we haven't seen these kind of numbers. That means that either the shift is starting very slowly, or one of the three assumptions is inaccurate.