Common Errors in Statistics (and How to Avoid Them), 4th Edition

Preface

ONE OF THE VERY FIRST TIMES DR. GOOD served as a statistical consultant, he was asked to analyze the occurrence rate of leukemia cases in Hiroshima, Japan following World War II. On August 7, 1945 this city was the target site of the first atomic bomb dropped by the United States. Was the high incidence of leukemia cases among survivors the result of exposure to radiation from the atomic bomb? Was there a relationship between the number of leukemia cases and the number of survivors at certain distances from the atomic bomb’s epicenter?

To assist in the analysis, Dr. Good had an electric (not an electronic) calculator, reams of paper on which to write down intermediate results, and a prepublication copy of Scheffe’s Analysis of Variance. The work took several months and the results were somewhat inconclusive, mainly because he could never seem to get the same answer twice—a consequence of errors in transcription rather than the absence of any actual relationship between radiation and leukemia.

Today, of course, we have high-speed computers and prepackaged statistical routines to perform the necessary calculations. Yet, statistical software will no more make one a statistician than a scalpel will turn one into a neurosurgeon. Allowing these tools to do our thinking is a sure recipe for disaster.

Pressed by management or the need for funding, too many research workers have no choice but to go forward with data analysis despite having insufficient statistical training. Alas, ...