Perhaps your question should be, which will provide the most precise location....
I have been using street addressing for years. And as I type this I am about to deploy address points.
When you geocode against a range of addresses on a linear feature, you get an interpolated location. I often use the example of a segment of Main St that has a range of 100-200 on the even side and 101-199 on the odd side. Address 150 will be matched in the middle of the block on the even side of the street. Accurate, but not especially precise. Maybe the actual house is on the other side of the street and down at the corner?
I mentioned that I am about to deploy address points. Most of these have been generated via the parcel centroid and then moved towards the the frontage street. Accurate and precise! Some of them have been GPSd. Accurate with precision on steroids!
Once you start going to street name and then postal you lose all sense of precision, and you accuracy takes a major hit too.
But....
What are you using the data for? I dispatch 9-1-1 calls. I need as accurate and precise data I can afford. (Oh, did I mention that really good data aren't cheap?) Maybe if you are mapping an outbreak of measles, postal/zip code is all you need. Accurate in the sense that you can get an idea where the sick kids are, but not very precise by any means. (In fact, most health related data are only available at the postal/zip level to assure privacy.)
You asked specifically about street level geocoding: lets go back to house 150 Main St. You will be matching only to the Main St part of the address.Which means if you have an address of 1500 Main St, the results will be the same. Right? I don't use that type of matching because its really of no value to me. Again, is all about how you need to use the data.
That should just about do it....
