Problems with P-Values As Illustrated Through Stories About Sex

PSYCHOLOGY COLLOQUIUM SERIES

Audience: Psychologists, psychologists-in-training, and the campus community

Problems with P-Values as Illustrated through Stories about Sex

Workshop Description

Most psychological research relies on p-values to make inferences about a general population based on a smaller sample of participants. Yet the p-value is also arguably the most widely misunderstood, misused, and abused statistical tool in existence. Psychology researchers are currently struggling with a so-called “reproducibility crisis,” in which a large portion of key published psychological findings cannot be replicated through independent research; problems with the p-value have likely played a large role in the crisis. To address these concerns, in March of this year the board of the American Statistical Society published its first consensus statement on p-values, highlighting common misunderstandings and abuses of the statistic. This talk will illustrate some of these problems through examples taken from recent published research.

Brief Biography

Dr. Nuzzo earned a bachelor’s degree in Industrial Engineering from the University of South Florida and a Ph.D. in Statistics from Stanford University, and completed a graduate program in science writing from the University of California, Santa Cruz. She has been a freelance science journalist since 2005 and a professor teaching statistics at Gallaudet since 2006. Her work has appeared in various publications, including Nature, New Scientist, and Reader’s Digest, Scientific American, and the Los Angeles Times. Her 2014 feature on p-values won the American Statistical Association’s Excellence in Statistics Reporting Award, and in 2015 she facilitated the American Statistical Association’s working group conference that produced the first p-values consensus statement.

Learning Objectives

As a result of attending this workshop, participants will:

1. Summarize the main conclusions of the 2016 ASA p-values statement.

2. Differentiate effect size from statistical significance.

3. Avoid “the winner’s curse” stemming from blind application of a bright-line p-value rule.

4. Informally incorporate prior probabilities in the interpretation of p-values.

5. Recognize p-hacking and its dangers.

6. Understand why more stringent p-value thresholds may be valuable.

Friday, October 14, 2016

Colloquium 1-3 pm HMB W220

DEPARTMENT OF PSYCHOLOGY | GALLAUDET UNIVERSITY