Assessing the Effectiveness of Normal Probability Plots

• Activity Number: 699
• Type: Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Education
• Author(s): John Walker*+
• Companies: Cal Poly San Luis Obispo
• Keywords: normal probability plot ; normality ; assumptions ; robustness

Abstract

Normal data is an assumption for many statistical methods. Many statistical methods are robust to non-normality, but not for small samples. Unfortunately, formal tests of normality have low power for small samples. We teach normal probability plots as a graphical method of assessing normality, but it can also be difficult to judge normality from these plots when the sample size is small. How useful are they, and how much experience must one have to use them effectively? We conduct a study in which professors and students judge a series of normal probability plots of randomly generated data from both normal and non-normal populations. The results for each group are compared to results from formal tests of normality.