Seminar on History and Philosophy of Science

Abstract: Principles of survey design (e.g., stratification) and experimental design (e.g., blocking and randomization) are often justified by their ability to produce reliable or "valid" estimates of a quantity (e.g., a population mean or average-treatment effect). I offer an alternative defense of those principles: commonly-employed scientific methodologies, I claim, can be interpreted as solutions to various social choice and bargaining problems among experimenters and research subjects with diverse values and preferences. I focus on randomized designs (e.g. random sampling and randomization in experiments) in this talk. I first critique standard justifications of randomized designs, showing they either (a) fail to justify typical randomized designs that are widely used in practice or (b) rely on suspect decision-theoretic principles. I then turn to my alternative defense, which employs tools from social choice and cooperative bargaining theory. I conclude by situating my arguments in broader philosophical discussions about the legitimate role of values in science.