The standard for that evidence should be very different than the standard for reproducible physics, chemistry and biology experiments.
Other times it's just an extrapolation of a preexisting data set. Running this query produced this result.
Don’t forget the likely bias that you are talking about scientific results in the press, which is a negligible subset of scientific work and which is biased towards humans, and sociological studies in particular.
From you comment I doubt you’re seeing much work in insect embryology or slippage in gravel pile formation (I don’t run across that stuff either)
> The standard for that evidence should be very different than the standard for reproducible physics, chemistry and biology experiments.
I don't really see why we should treat evidence coming from surveys differently than evidence coming from physics, chemistry or biology experiments as long as surveys are well constructed, fulfill the criterion of validity and reliability and are reproducible. Yes, they are not 'hard facts' but often the only option to measure latent variables or fuzzy constructs. If they do this in a reproducible way I don't see the issue here.
On the other hand, 'hard sciences' such as neuroscience and medical science have a big problem with non-reproducible studies, even though they use more 'objective' measurement methods.