Biases in scientific research

Catalog description

This course explores how to avoid unintentional errors in research design and statistical analysis, i.e. unintentional statistical misconduct. Students explore examples of research results that failed to reproduce, papers that were retracted, or results that generally seemed suspect. Topics to be discussed include practice that are fairly widespread, but sometimes hard to detect, and even harder to correct in one’s own work: p-hacking, peeking, multiple comparisons, selection bias, sampling bias, hidden confounds, coding errors, outlier rejection, dependent samples, model miss-specification, post hoc hypothesizing, underpowered samples, non-homogeneous groups, subject bias, experimenter bias, unblinding, publication bias, etc. We are not interested in obvious scientific misconduct, such as data manipulation, plagiarism, paper mills, citation cartels, etc. which are already covered in a scientific ethics course. Students should have previously taken a graduate course in statistics, such as biostatistics.