[sacrosancty]

That's extremely context specific. You can imagine something like in social sciences where people have quite finely tuned but subjective personal awareness of human behavior and some data makes a coarse but objective measurement of it. It's hard for the data to be as "accurate" as the personal experience, so it can easily be wrong. But "it was colder when I was a kid" vs temperature measurements is the other way around.

[naniwaduni]

I mean, we're looking at an example where the anecdotes and data are fairly in agreement. But the point is that when you encounter a metric that's at odds with recurrent anecdotes, you should strongly consider that you aren't measuring the right thing. You can get incredibly accurate measurements of a doorframe's height but if people keep telling you they're having trouble getting packages in, it might just be that they don't fit the width.

[maximus-decimus]

On the flip side, if you change metrics until you get one that says what you want, do your metrics really provide any value?