Nobody said it was a study on the general population, only on the UP wearers population.

OTOH, if their sample size is the same as the population size, their data is by definition statistically significant.

Not even that. If the sample size in general is extremely large (irrespective of its proportion to the population size), than the results are statistically significant.

A sample size in the thousands counts as "extremely large."

Surely the sample, however large, must follow the same distribution than the population distribution? If you sample 1000 insomniacs, how do you get anything significant?

My remark was just that, the sample bias has an impact on the significance of the results. I do not know if in actuality people using sleep monitors form a different distribution than the general population regarding sleep patterns, but it seems likely.

The title of the article mentions "Bay Area sleepers", so it would seem to me that they do form conclusions about the general population, not just Jawbone wearers.

I think that most people greatly overestimate the sample size needed to achieve statistical significance.