Statistical Literacy for the Humanities: Evaluating the Statistical Methods used in the Social Sciences
A hallmark of the social sciences is their reliance on statistical methods involving data often obtained in observational studies. Statistical literacy is presented as the study of statistics as evidence in non-statistical arguments. Four topics are presented. (1) Although social scientists are well-aware that "association is not causation", the inductive inferences involved may get lost in the deductive aspects of confidence intervals and statistical significance (p-values). The susceptibility of a weak association to the influence of a stronger confounder is reviewed. Using a simple graphical technique, the ability of a confounder to change an observed association and its' statistical significance is illustrated. (2) Social scientists may hold that the more factors one takes into account (the smaller the adjustments upon taking additional factors into account), the closer one is to the true association between the variables of interest. And thus, the closer one is to understanding the natures, causes, or consequences of things. Objections to this claim by Turner (Causality in Crisis) and by Lieberson (Making It Count) are reviewed. (3) Some of the newer methods in the social sciences are based on methods developed in an epidemiological-biological context. The applicability and utility of these methods in a social-cognitive context are examined. (4) Examples of how the failure to include a relevant predictor can greatly change the results of a study are presented. In conclusion, humanities majors should be statistically literate about the strengths and weaknesses of the statistical methods used by the social sciences.
Milo Schield (United States)
Director, W. M. Keck Statistical Literacy Project
Department of Business Administration
(30 min. Conference Paper, English)