Thomas Lumley • 6/20/2015

A much-needed gap

The article argues against the common misuse of the Shapiro-Wilk normality test on data from large, multi-stage surveys (e.g., NHANES). It details why this is problematic: normality is rarely the key question, test power is overly sensitive to sample size, sampling weights are ignored, and the null hypothesis is ill-defined for such complex data.

0 comments

#data analysis #statistics #Shapiro Wilk Test