This seems like the kind of thing Greg Mankiw might want to justify, rather than simply assert:

First, since most infrastructure is used locally, the proper level of spending is best determined by state and local governments rather than by the federal government. Earlier, I suggested that fiscal stimulus could be decentralized. Each state governor could be allowed to determine whether to take federal money as state aid or have it paid directly to his or her state’s citizens as tax relief. I still think that makes sense.

I can think of about ten different ways that this doesn’t stand up to scrutiny. First and most obviously, quality interstate transportation is economically important. Absent federal coordination of infrastructure spending, we would probably see sub-optimal investment in such transport. It doesn’t do a state much good to build a high-speed rail line or new freight capacity up to its border and no further.

I suspect that states and local governments will simply say that they’ll take as much infrastructure money as they can get; voters are likely to look more kindly on completed roads and railways than they are on another $300 check. It should also be clear that if the primary economic unit is the metropolitan area, then state and local governments might not use their infrastructure allocation particularly well. State capitals might not be politically able to spend money on infrastructure in rich metropolitan areas, even if that’s where the money would do the most good. And local governments often are too small to think big enough. Give the money to a bunch of suburban counties, and they’ll all get busy widening local roads, whether or not that generates the best outcome for the local economy.

It’s kind of weird to suggest something like this, which would only work if we had this idealized set of local governments. Federalism might make sense when you’re trying to establish policy competition, in order to see what works best. It makes no sense at all when building infrastructure networks, where decisions made in one location display significant spillovers at the local, state, and regional level.