Methods Minute: Why Organizational Science Should Consider Bayesian Methods

Topic(s): Uncategorized

Topic: Research Methodology, Statistics
Publication: Organizational Research Methods
Article: The time has come: Bayesian methods for data analysis in the organizational sciences
Authors: Kruschke, J. K., Aguinis, H., and Joo, H.
Reviewer: Neil Morelli

If you’re like me you’ve probably heard grumbling about the limitations with traditional null hypothesis significance testing (NHST). Or, if you’re like me, you’ve probably grumbled about it yourself.

Kruschke, Aguinis, and Joo say you’re not alone and argue that organizational science should adopt Bayesian methods instead of NHST for data analysis. (Unfortunately, organizational science has yet to jump on the bandwagon; out of 10,000 articles published over a 10 year period, only 42 used Bayesian methods!)

If Bayesian methods are so great, how do they work? Kruschke et al. goes over a full demonstration using a multiple linear regression example that is definitely worth reviewing. But as a general overview, Bayesian methods are the mathematically correct way of “reallocating credibility” across the parameter values of a prior distribution using observed data. This reallocation results in a posterior distribution, which offers the most complete information allowable by the observed data.

To help explain, I’ll borrow an analogy given by the authors that Bayesian methods are kind of like Sherlock Holmes. To solve a case, the famous detective starts with known information (prior distribution) and deduces the most likely culprit (posterior distribution) using observation (available data). The evidence he observes make some suspects more credible than others.

Ok, but why use Bayesian methods? To the authors, Bayesian methods provide a vast improvement over the problematic traditional NHST methods. Here are few examples: Bayesian methods use previous information to improve inferences whereas NHST does not, and Bayesian methods are able to accept the null hypothesis, which drastically lowers the chances of a false positive. As an added bonus you also get a complete distribution of reasonable R2 values.

So, all the stat jargon aside, Bayesian methods should be considered because they provide the information researchers are often seeking in a fuller and more confident way, and can also be used with the data-analysis methods researchers are already familiar with. Bottom-line? According to the authors, Bayesian methods are not a silver bullet, but they do offer new opportunities for organizational science to develop more precise models and conclusions.

Kruschke, J. K., Aguinis, H., & Joo, H. (2012). The time has come: Bayesian methods for data analysis in the organizational sciences. Organizational Research Methods, 15(4), 722-752.

human resource management, organizational industrial psychology, organizational management




source for picture: