TY - JOUR

T1 - Token swap test of significance for serial medical data bases

AU - Moore, G. William

AU - Hutchins, Grover M.

AU - Miller, Robert E.

N1 - Funding Information:
From the Departments of Pathology and Laboratory Medicine, The Johns Hopkins Medical Institutions, Baltimore, Maryland. This work was supported by National Institutes of Health Grant LM-03651 from the National Library of Medicine and Grants HL-17655 and HL-22963 from the National Heart, Lung and Blood Institute. Requests for reprints should be addressed to Dr. G. William Moore, Department of Pathology, The Johns Hopkins Hospital, Baltimore, Maryland 21205. Manuscript accepted February 25, 1985.

PY - 1986/2

Y1 - 1986/2

N2 - Established tests of statistical significance are based upon the concept that observed data are drawn randomly from a larger, perhaps infinite source population. The significance value, p, is the probability that the observations are drawn from a source population satisfying the null hypothesis; if p is small enough (less than 5 percent, 1 percent, etc.), then the null hypothesis is rejected. Serial medical data bases, such as a hospital clinic intake or autopsy case accessions, often do not have an identifiable source population from which they are randomly drawn. In an effort to make a reasonable interpretation of these less-than-ideal data, this report introduces a "token swap" test of significance, in which the usual paradigm of repeated drawing from a source population is replaced by a paradigm or misclassification within the observed data themselves. The token swap test consists of rearranging the data into a balanced distribution, and determining the disparity between the observed and the balanced distribution of data. In a two-by-two contingency table, patients are represented as "tokens" distributed into four "cells". Significance is determined by the proportion of "token swaps" that are able to transform the balanced table into the observed table. The token swap test was applied to three series of autopsy observations, and gave results roughly comparable to the corresponding (two-tail) chi-square and one-tail Fisher exact tests. The token swap test of significance may be a useful alternative to classic statistical tests when the limiting assumptions of a retrospective, serial medical data base are present.

AB - Established tests of statistical significance are based upon the concept that observed data are drawn randomly from a larger, perhaps infinite source population. The significance value, p, is the probability that the observations are drawn from a source population satisfying the null hypothesis; if p is small enough (less than 5 percent, 1 percent, etc.), then the null hypothesis is rejected. Serial medical data bases, such as a hospital clinic intake or autopsy case accessions, often do not have an identifiable source population from which they are randomly drawn. In an effort to make a reasonable interpretation of these less-than-ideal data, this report introduces a "token swap" test of significance, in which the usual paradigm of repeated drawing from a source population is replaced by a paradigm or misclassification within the observed data themselves. The token swap test consists of rearranging the data into a balanced distribution, and determining the disparity between the observed and the balanced distribution of data. In a two-by-two contingency table, patients are represented as "tokens" distributed into four "cells". Significance is determined by the proportion of "token swaps" that are able to transform the balanced table into the observed table. The token swap test was applied to three series of autopsy observations, and gave results roughly comparable to the corresponding (two-tail) chi-square and one-tail Fisher exact tests. The token swap test of significance may be a useful alternative to classic statistical tests when the limiting assumptions of a retrospective, serial medical data base are present.

UR - http://www.scopus.com/inward/record.url?scp=0022587053&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0022587053&partnerID=8YFLogxK

U2 - 10.1016/0002-9343(86)90007-0

DO - 10.1016/0002-9343(86)90007-0

M3 - Article

C2 - 3511687

AN - SCOPUS:0022587053

VL - 80

SP - 182

EP - 190

JO - American Journal of Medicine

JF - American Journal of Medicine

SN - 0002-9343

IS - 2

ER -