I do think data-driven par is the best method of setting par. Whenever I say par "should" have been this or that, my statement is backed up by a lot of scores from a lot of players at the target skill level.
But, this is just one of a few good methods that can also produce acceptable par, if they are applied right.
Whatever method is used, it should result in a par that is good for telling players how well they are doing during a round, compared to other rounds, and compared to other courses. It should do this by making the cost of a bogey about equal to the value of a birdie. To put it another way, after each hole we should be able to say whether the player moved toward a score that would win a big tournament, moved toward a score that would not cash in a major tournament, or held ground; simply looking at how they scored relative to par.
A few years ago, when it became apparent that the pars being assigned were not doing the job as described above. I set out to come up with a whole new definition of par that would work better. To do that, I needed to examine the existing definition. By examine, I mean I needed to apply it to a lot of tournaments to see how well it performed and where it was wrong. So, I encoded the actual definition into my data-driven method (as best I could). I then ran a bunch of calculations, and found out the current definition works just fine.
I'm fine with any method, as long as it produces pars that work well. See
here.