Paul GrahamVerified account

@paulg @isallmaroon So what are the numbers for other sciences?

@paulg This is seriously alarming and compels us to figure out new and better ways to perform social experiments for better reliability.

@isallmaroon @paulg So far so good for the Social, then :-) … Now, what I fear are numbers related to computer science.

@isallmaroon Yeah. IIRC a recent episode of @EconTalker was also about the value of replication attempts. Doubt any field does enough.

@zedshaw @paulg this is my favorite: http://blog.dilbert.com/post/109880240641/sciences-biggest-fail …

@paulg Fair, but overfitting and data manipulation is not limited to social psych. Doesn't basically all startup advice have the same issue?

@paulg @knewter I think that comes out of our @OSFramework

@paulg @curiouscat_com This is, I think, mis-stated. Best not to think of research as true or false. It doesn't work like that.

@paulg @johngapper as a psych grad, doesn't surprise me. Esp the ' psychology only tells us what a 1st year psych grad would do' bit!

@paulg as Thiel said, if a discipline has "science" or "logos" as part of its title, it is dead givaway it is neither