As an industry we seem to think hard and make proclamations about programing. Where we should be thinking hard and making hypothesis and testing them.

Turns out some people have been doing this and some people put it into a nice book. They conducted systematic reviews of the literature around many software development questions.

The book (I know about anyway) is:

Making Software: What Really Works, and Why We Believe It

and Chapter 17 address pair programing.

It has been a while since I read it, and it is a little thick to jump in and read it to make sure I'm 100% accurate ATM. But the summary I took away for my work environment is that, you get the same productivity with 2 devs working by themselves vs 1 pair, with a lower defect rate. I'd like to quote things directly but I'm not investing that much time right now.

Pair programing IMHO is skill that needs to be learned, and it is a different skill depending on the type of pair you have at the time and what you're trying to do