Experimental units--are both suitable?

You have observed x drivers on their actions at STOP intersection (stop or not stop) at n intersections. from these data you can obtain the proportion of drivers stoppting by pooling data at all intersections--option 1 (N=x), or you can obtain such estimate for each individual intersections (N=n) and then find the average of these--option 2. Can I use both to estimate the compliance rate? why ao why not? each of these represent what?--any idea I will appreciate