Along with my previous post on programming for social scientists, this is another must read paper (even if your branch of social science isn’t psychology since this problem exists throughout the social scientists).
Written by University of Texas-Austin Psychology Professor Tal Yarkoni and his post-doc data scientist Jacob Westfall and published in a top 10 psychology peer-reviewed journal, this article is full of practical advice with provocative implications for the quality of social science research to date.
For a practicing data scientist, most of the advice is obvious. But if you came to computational social science from the social science side of the house, much of it sounds like heresy. Unless you’ve already become suspicious of your field’s findings due to rampant withdrawals of papers, contradictory/not reproduced results, and tiny data sets.
According to this Google Scholar search, this article is the 3rd-most cited psychology paper since 2017, so apparently people are listening.
My next wish: a psychology department that has a basic machine learning course in it’s list of requirements and one less required statistics class. Or at least include a machine learning class in its electives to help balance out the plethora of statistics classes in their curriculum.