Don Zimmerman's

Statistical Devil's Dictionary

with apologies to Ambrose Bierce

Mean—a statistic used by researchers who want nothing to do with the bulk of the data.

Variance—a measure of the degree to which investigators with a fetish about the mean ignore uninteresting and possibly important outlying measures.

Median—a complex statistic favored by computer programmers who enjoy nontrivial programming challenges.

Standard deviation—a measure devised by clinical psychologists to describe those quantitatively-oriented colleagues regarded as deviant and obsessive about statistics.

Percentile—a measure of the degree to which well-motivated examinees, after taking a test, can justifiably look down upon their lesser peers.

Decile—a percentile for people who have difficulty keeping more than ten things in working memory.

Normal distribution—a distribution free from abnormalities, inconsistencies, distortions, or features of any kind that could compromise efficient data analysis.

Probability of a Type I error—the probability that an investigator will announce an important discovery to the world.

Probability of a Type II error—the probability that an investigator will continue performing experiments until a Type I error occurs.

Nonparametric test—a test devised by the Albanian statistician Imrovic Nonparametric in order to provoke the Englishman Ronald Fisher.

ANOVA—(also spelled "A Nova")—the feminine counterpart of "A Novus" (also spelled "A Novice")—A female statistician new to the discipline.

Population—the grand total of every last bugger and his brother who possibly could be measured.

Sample—a rag-tag, bob-tailed bunch of atypical misfits who have volunteered to participate in an experiment.

Null hypothesis—an assumption about the nonexistence of even the slightest difference between two quantities, made so that discovery of a slight one will be an occasion for rejoicing.

Alternative hypothesis—an hypothesis waiting and ready to be presented in case the manuscript with the first hypothesis is rejected for publication.

Sample size—the number of observations needed to make a study appear, at least superficially, to be well-designed.

Skewness—a noticeable imbalance in the shape of a distribution resulting from failure to obtain data that conforms to normality.

Kurtosis—a disease that causes populations to accumulate too few central exemplars and too many outlying ones (or vice versa) and that in addition often produces headaches among statisticians.

Control group—the unsung, forgotten participants in an experiment who have to get along with old methods or sugar pills and are never allowed to get in on the real action.

Bootstrap—a tool employed by young, modern statisticians to intimidate older statisticians who are more set in their ways and lack familiarity with recent computer-intensive methods.

Jackknife—a truly vicious intimidation tool, occasionally wielded by investigators who fail to get results with the bootstrap.

Reliability—a measure of the extent to which repetition of an experiment results in the same errors that occurred the first time.

Validity—a measure of the extent to which the results of a procedure conform to the preconceptions, beliefs, and expectations of the theorist planning and executing the procedure.

Factor analysis—an elegant mathematical technique for replacing a large number of interesting variables with a small number of hypothetical, unobservable, and uninteresting ones.

Standard error—the commonplace and predictable extent to which authors of statistics textbooks copy and perpetuate the discrediited beliefs and mistakes of earlier authors.

This dictionary is still under construction