Experiment and bias: the case of parsimony in comparative cognition

Abstract

Comparative cognition is the interdisciplinary field of animal cognition and behavior studies, which includes comparative psychology and branches of ethology, biology, and neuroscience. My dissertation shows that the quasi-epistemic value of parsimony plays a problematic role in the experimental setting of comparative cognition. More specifically, I argue that an idiosyncratic interpretation of the statistical hypothesis-testing method, known as the Neyman-Pearson Method (NPM), embeds an Occamist parsimony preference into experimental methodology in comparative cognition, which results in an underattribution bias, or a bias in favor of allegedly simple cognitive ontologies. I trace this parsimony preference to the content of the null hypothesis within the NPM, and defend a strategy for modifying the NPM to guard against the underattribution bias. I recommend adopting an evidence-driven strategy for choosing the null hypothesis. Further, I suggest a role for non-empirical values, such as ethical concerns, in the weighting of Type I and Type II error-rates. I contend that statistical models are deeply embedded in experimental practice and are not value-free. These models provide an often overlooked door through which values, both epistemic and non-epistemic, can enter scientific research. Since statistical models generally, and the NPM in particular, play a role in a wide variety of scientific disciplines, this dissertation can also be seen as a case study illustrating the importance of attending to the choice a particular statistical model. This conclusion suggests that various philosophical investigations of scientific practice - from inquiry into the nature of scientific evidence to analysis of the role of values in science - would be greatly enriched by increased attention to experimental methodology, including the choice and interpretation of statistical models.