Ik word ziek van de statistiek, of: er van weten zonder er naar te handelen
Auteurs
Molenaar,W.
Trefwoorden:
J. Cornfield, J. W. Tukey's "two spans of the bridge of inference, " generalizations, random sampling, Bayesian methods
Samenvatting
A discussion of some well-known methodological issues leading to incorrect use of statistical inference by many social scientists. Frequent misuse of statistical techniques is partly due to a concentration in textbooks on technical issues instead of methodological foundations; such misuse should not be a reason for deleting statistics from social science curricula. The foundations & applications are far more controversial than the purely mathematical theory of statistics & probability. Introduced is the paradigm of "the two spans of the bridge of inference" (Cornfield, J., & Tukey, J. W. Annals of Mathematical Statistics, 1956, 27, 912) showing how a statistical generalization to the operational population (from which a sample was drawn) is often followed by a generalization to a wider target population. This second step is mathematical, but should be based on the substantive insight of the investigator or the reader. The pitfalls of applying results valid for simple random sampling to all kinds of cluster samples, systematic samples, or haphazard samples so frequently used in practice, are warned against. Opposed is statistical decision theory, in which two decisions & their losses can be symmetrically viewed, & hypothesis testing, in which the null hypothesis survives unless clearly contradicted. The issues of one-sided vs two-sided tests & of situations where it is desired to prove the null hypothesis are discussed. The confusion of "large enough for statistical significant" with "large enough to be of practical importance" is deplored. It is demonstrated how the use of confidence bounds can be informative on both significance & importance, & Bayesian methods are discussed. Overambitious projects in which too many variables are measured, unreliably or invalidly, by a failure to concentrate on a few well-designed problems, are warned against, & some threats to validity commonly met in sociological research are discussed. Modified AA.
