Coming from a machine learning background, I am often shocked at the quality of experiments in psychology and even economics.
I think the reason may be because in machine learning, you describe your algorithm and experiment exactly. If someone can use the same algorithm and devise an experiment to disprove your results, you look like a fool. With softer sciences, maybe people just say, "then you obviously didn't do it right!" and there is no code to point at.
I like to think that in medicine they know how to do things properly, since there is so much regulation and money on the line.
Studies of humans (medical, psychological, economic, etc) are attempting to discover unknown algorithms on unknown, extremely complex systems. The quality will always seem lower than engineering studies because scientists don't know what they don't know.
Example, quoted from the website:
""" The babies in the unjustified group spent more time looking at the adult's discordant expression and exhibited more checking behavior. Additionally, the babies' facial expressions only displayed empathy for the adult when a negative emotional response fit the scenario.
The takeaway:
Parents, take note: Your toddler can see through your "stiff upper lip" routine.
"""
The "takeaway" is completely unsupported. In no way does the experiment suggest that hiding pain is useless.
For one thing, "stiff upper lip" (aka "faked neutrality") is different from faked happiness.
For two, the experimental methodology only applies to cases where the baby knows that the experience's standard emotional response. That applies to hammer strikes on a finger, but not something like divorce or losing a job.
For three, "spent more time looking at the adult's discordant expression" shows that the babies learn from their parents reactions, as they expend effort to understand counterintuitive behavior. "Counterintuitive" is not the same as "fake" or "wrong".
So the "takeway" may in fact be the opposite of what the data suggests!