Flynn, Michael F. (1986, ASQC) STAT-A-MATRIX, Inc., Edison, NJ
Many samples conducted by modern businesses are wastes of time, money and resources. They do not accomplish their objectives (for they ony seem to). However, this failure is often inadvertent. Many people do not know how to make their data meaningless and so must rely on chance to do it for them. This paper seeks to rectify that by pointing out several tried-and-true ways that anyone can use.The best place to build in problems is in the initial planning and execution of the sample. Once the numbers themselves are bad or inappropriate, it doesn't matter how sophisticated the statistical analysis is. The next best place is in the interpretation of the results. The intervening number-crunching is simple enough to teach to machines. We can classify these non-numerical errors loosely as: a) planning errors, b) selection errors, c) measurement errors, and d) interpretation errors. Included among them are Kimball's "Errors of the Third Kind" and the dread "Rabbit's Foot Fallacy." None of them can be corrected by inflating the sample size. In fact, the proportion of non-sampling errors will increase with the sample size.Hegel the philosopher once said that if we do not learn the errors of the past we are condemned to repeat them. This is intolerable. We should be able to create our own, new errors.