[mjburgess]

A counter-point, in a certain sense: when the conclusions of scientific papers (in these softer science fields), contradict common sense, they tend to be unreproducible; the ones which don't, are.

The problem with studying humans is, roughly, the central limit theorem doesnt work: properties of biological and social systems do not have well-behaved statistics. So all this t-test pseudoscience can be a great misdirection, and common sense more reliable.

In the case where effect sizes are small and the data generating process "chaotic", assumptions of the opposite can be more dangerous than giving up on science and adopting "circumstantial humility". (Consider eg., that common sense is very weakly correlated across its practicioners, but "science" forces often pathological correlations on how people are treated -- which can signficantly mangify the harm).

[crazygringo]

> when the conclusions of scientific papers (in these softer science fields), contradict common sense, they tend to be unreproducible; the ones which don't, are.

Citation needed?

I don't know what would lead to that conclusion. And it would seem to run counter to the entire history of the field of psychology, for example.

[kanjus]

Can't find the citation, but remember gwern mentioning a study in one of his posts on replication that found that unintuitive findings tend to be both less replicable and more cited than intuitive ones.

Psychology is the field that is most hit with replication failures and has a slew of unintuitive results that turn out to be malpractice.

[crazygringo]

Psychology is also the field with a slew of unintuitive results that have been repeatedly replicated as correct. And what is "intuitive" anyways? What was extremely non-intuitive a century ago is common sense today.

So that's why I question the assertion. You're right that there are tons of replication failures, but whether intuition correlates with replicability way doesn't seem relevant. Especially when the point of so much research is to look for currently "non-intuitive" things, so of course that's where more replication issues might exist. It doesn't mean you should stop researching in that direction.