I don't think anybody's trying to throw the baby out with the bathwater.

This is entirely serious. But most people have no idea how broad and deep the reach of the hard science integrity problem goes (at least in biology).

Not only that, but it's HARD for people to know, and there are with 100% certainty efforts to keep it that way.

BTW, social sciences get credit for the modern economy and for the known history of humankind, for modern farming, for today’s government and policy, for mental health and mental therapies, for modern advertising, for understanding of languages, and for modern corporate management, just to name a few.

One example: I did a pure-chemistry undergrad degree and a psychology-adjacent graduate degree. The "hard science" degree involved basically zero applied statistics of any kind, any statistics concepts that came up were emergent from lower level stuff even in e.g. stat mech courses, and I wasn't even aware that "Design of Experiments" was a thing. Whereas the "soft science" curricula was heavily focused on statistical rigor, DoE, layers upon layers of internal controls, and so forth; basically because there was no practical way to see an unambiguous effect. It certainly seemed "rigorous".

However the soft science stuff just has less predictive power despite the rigor. In broad sense it relates to human perception, capabilities of technology, model systems, semiotics/epistemology (maybe wrong word), prediction/confirmation, especially faith.

For example, at a macro scale on Earth, it is very easy to accurately predict at human scale how things work. Micro scale physical objects require a bit more bootstrapping. You can't see most molecules or atoms, so you have to come up with a way of inferring measurements through other processes which themselves have to be trustworthy. Etc.

At some point of course, one has to select some model as an axiom or matter of faith to make any progress. You can't model a dropping brick if you can't trust your timer or measuring tape, etc. So you stand on giants' shoulders and make a predictive model as an extension of the axiomatic one.

This starts getting really squishy when you're dealing with entire concepts that have no agreed upon definition whether quantitative or qualitative and try to make them into something onto which statistical rigor can be applied!!!!!

So with soft-science literature it is always extremely important to mentally substitute the details of a model's implementation, for the shorthand expression used to describe it which may overlap with a commonly used word.

For example, "We found that this drug candidate significantly reduced depression in mice" -> "We found that this drug candidate significantly increased the amount of time mice swim around before giving up when you chuck them into a tank of water, etc.

because the rigorous conclusion might not actually be practically meaningful if the axioms are practically unpredictive.

And even worse, going back to the "no agreed definition" thing, the definitions of psychological concepts are only defined in terms of these very vague experiments! It's like bootstrapping physics if you're a disembodied nothing in a simulation.

idk there's much more to say on this but I'm procrastinating at work so