The only distinction I care about is exactness and non-exactness.
Is the research based upon formulating a theory that is capable of forecassting not-yet observed events a nonexistent, exact margin or error, and are the conditions then re-created to see if the forecast is within the margin of error that the instruments that measure it have?
Some say biology is “hard”, and some say it is “soft”; some say many parts of cosmology are “hard” but they certainly aren't “exact”.
In exact science there are typically multiple ways to derive the same answer within one theory, and they all result into the exact same result.
And it starts with that the terms are colloquial and bereft o an actually hard definition.
“colloquial”, “roughly”, ”perceived”. — these are not the terms that definitions are made of.
The point is that there is no actual hard distinction between “hard science” and “soft science” but there is a hard distinction between an exact theory, and an inexact theory.
There is no exact science, physics isn't an exact science either.
You need softer definitions like "Hard" and "Soft" precisely because there are no exact results in any science we have. And it is fine to use soft definitions for these things since categorising scientific fields doesn't need to be a science.
A hard science is where replication is expected to never fail, if replication fails once then the theory is thrown out. Applying that to social science studies would seem ridiculous, no social scientist would want that, so they want their field to be soft.
> There is no exact science, physics isn't an exact science either.
Which is why I said “exact theory”.
> You need softer definitions like "Hard" and "Soft" precisely because there are no exact results in any science we have. And it is fine to use soft definitions for these things since categorising scientific fields doesn't need to be a science.
There are exact results every day, but those are not delimited cleanly by “fields”: a theory is exact or it is not.
> A hard science is where replication is expected to never fail, if replication fails once then the theory is thrown out. Applying that to social science studies would seem ridiculous, no social scientist would want that, so they want their field to be soft.
And this criterion is never mentioned at any point in the Wikipedia article linked.
It also seems a useless definition as replication can always fail due to flukes, and the confidence numbers chosen for replication, typically within 0.05, are very arbitrarily chosen.
Whether it is “replicated” or not is a rather arbitrary delimitation of an arbitrarily picked number, and 5% is certainly not improbably low to begin with.
‘Hard science’ is just a title people have been using to distinguish physical sciences from social sciences. Don’t get too hung up on worrying what ‘hard’ means or thinking that it is some kind of difficulty or value judgement; it’s not. It may have been at some point, but today it’s just a category and nothing more. This WP entry is much more clear than the above one, IMO: https://simple.wikipedia.org/wiki/Hard_science