There are academic studies of pair programming. If I remember, it does obviously decrease software defects, but almost everyone that wants pair programming ignores the fact that you are now paying twice as much for the same software. The software is not 2x better. Obviously a programmer's salary is so minuscule to the business managers don't even factor this into their thinking. Usually they want pair programming for political reasons, i.e. no single person can ask for a raise due to being the most knowledgeable.

Fully agree. I would be really interested in seeing such a study.

There is no need of studies to see that pair programming is a great way to pass experience from a more to a less experienced software engineer. It might not be the only one way to tutor a junior, not even the best, but works.

Hopefully you don't work in medicine or in any discipline where lives are at stake.

The irony of this comment is that letting people die via A/B guessing is exactly how modern medicine got to where is today.

You give one group a "cure", the other group a placebo. Whichever group survives or does better is the winner.