Herman Rubin wrote: > The problem is that you are trying to teach statistics > without the background needed to UNDERSTAND it. Without an > understanding of probability concepts, NOT how to calculate > probabilities in simple problems, no statistical procedure > "makes sense". Means, medians, variances, significance > tests, confidence intervals, etc., are dei ex machinae. > > The worst of all consists of transformations to normality. > This destroys all structure, and in the real world, nothing > is normal. Correlations are much harder to do inference > with than covariances; statistics is not a collection of > religious mantras, but is the problem of decision making > under uncertainty. It must be modeled as such.
Over the years, Prof Rubin has made this statement repeatedly. I believe it. I see how those without the understanding he speaks of have great difficulty integrating concepts and how they fail to see connections that such understanding makes obvious. There is another reality we must face as educators. There are more data sets than trained statisticians to analyze them, while fewer people are choosing statistics as a career. During WWII, the need for medical doctors was so great that three year programs were developed to keep that profession staffed. Without being overly dramatic, statistics is in a similar situation. Prof Rubin reminds us of the dangers of shortcuts, but not all shortcuts are need be dangerous. I believe many of these introductory courses are attempts to get people to the front lines ASAP. They will necessarily be inadequate in some respects. What is needed is a systematic study and development of ways to integrate the two approaches to produce, if not the next generation of statisticians, then a new generation of analysts who might work in close association with a statistician. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
