Let's redefine it with this language:
"When statistics are used as PR, data gathering is a joke, state and police competence is cratering, and direct experience and observation tells you things are getting worse, things are getting worse".
The problem is the claim doesn't distinguish biased sampling from a biased model. The former is far more likely.
Just ask yourself which is more likely: "my experience is abnormal" or "everyone is experiencing the same thing I am"? In your friend group the latter might be more likely because you're similar locally and culturally. But across the state? Across the country? Across the globe? No, you'd be crazy to think experiences are typical. There's way too many variables at play and even if we were clones we should expect differences.
So no, it doesn't make sense
We can redefine this with more statistical language. The claim would then be "When the likelihood of a sample is low, the sample is likely correct." Which is nonsense.
It is worth noting when you continuously are getting low likelihood samples, but the usual conclusion is that you have biased sampling. Maybe the model is bad, but in some sense that's not so different.
Let's go back to normal language. Anecdotally, a HN user might think a $100k/yr salary is not very much. But the data suggests that it is. Is the anecdote right?! No, we are just biased because we're comparing incomes in tech and often around the Bay area.
I think it is no surprise that one of the richest men in the world is out of touch. He is, by definition, a statistical anomaly. ¯\_(ツ)_/¯