Academic journal article Psychonomic Bulletin & Review

Optional Stopping: No Problem for Bayesians

Academic journal article Psychonomic Bulletin & Review

Optional Stopping: No Problem for Bayesians

Article excerpt

Published online: 22 March 2014

© Psychonomic Society, Inc. 2014

Abstract Optional stopping refers to the practice of peeking at data and then, based on the results, deciding whether or not to continue an experiment. In the context of ordinary significance-testing analysis, optional stopping is discouraged, because it necessarily leads to increased type I error rates over nominal values. This article addresses whether optional stopping is problematic for Bayesian inference with Bayes factors. Statisticians who developed Bayesian methods thought not, but this wisdom has been challenged by recent simulation results of Yu, Sprenger, Thomas, and Dougherty (2013) and Sanborn and Hills (2013). In this article, I show through simulation that the interpretation of Bayesian quantities does not depend on the stopping rule. Researchers using Bayesian methods may employ optional stopping in their own research and may provide Bayesian analysis of secondary data regardless of the employed stopping rule. I emphasize here the proper interpretation of Bayesian quantities as measures of subjective belief on theoretical positions, the difference between frequentist and Bayesian interpretations, and the difficulty of using frequentist intuition to conceptualize the Bayesian approach.

Keywords Optional stopping · Bayesian testing · p-hacking · Statistics · Bayes factors

(ProQuest: ... denotes formula omitted.)

The field of psychology is experiencing a crisis of confidence, as many researchers believe published results are not as well supported as claimed (Carpenter, 2012; Roediger, 2012; Wagenmakers, Wetzels, Borsboom, van der Maas, & Kievit, 2012; Young, 2012). This crisis is comprised of publicized failures to replicate claimed effects, the publication of fantastic ESP claims, and the documentation of outright fraud. A common focus is now on identifying practices that violate the assumptions of our methods, and examples include peeking at the results to decide whether to collect more data (called optional stopping) and making additional inferential comparisons that were not considered before data collection. These questionable practices go under the moniker of p- hacking, and a remedy for the crisis is to avoid these bad practices (Simmons, Nelson, & Simonsohn, 2011).

An alternative viewpoint about the cause of the crisis is that the dominant inferential method, significance testing, is inap- propriate for scientific reasoning (Rouder, Morey, Verhagen, Province, & Wagenmakers, 2014). Many who are critical of significance testing recommend inference by Bayes factor as a replacement (Edwards, Lindman, & Savage, 1963;Gallistel, 2009; Myung & Pitt, 1997; Rouder et al., 2014; Rouder, Speckman, Sun, Morey, & Iverson, 2009; Sprenger et al., 2013; Wagenmakers, 2007). The Bayes factor comes from Bayesian analysis and results from using Bayes's rule to update beliefs about theoretical positions after observing ex- perimental data.

This article is about the wisdom of optional stopping, where the researcher collects some data, analyzes them, and on the basis of the outcome, decides to proceed with more data collection or not. Optional stopping is considered one of those bad p-hacking practices because it does affect conclusions from conventional significance tests. Yu, Sprenger, Thomas, and Dougherty (2013) have shown that common practices inflate both type I and type II error rates. Despite these results, there is a sense in which optional stopping seems like a smart thing to do. We seemingly should monitor our results as they come in, and we should end early when the results are clear and perhaps keep going when they are not. The critical ques- tion addressed here is whether optional stopping is problem- atic in the Bayesian context.

The answer to this question seems like it should be straight- forward, yet the literature is contradictory. On one hand, early Bayesian theorists stated that Bayesian quantities are interpretable under optional stopping. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.