THE STATISTICAL ANALYSIS OF THE RESULTS OF SAMPLING AN ENVIRONMENT FOR A CONTAMINANT WHEN MOST SAMPLES CONTAIN AN UNDETECTABLE LEVEL

Abstract

Berry, G. (MRC Pneumoconiosis Unit, Llandough Hospital, Penarth, Glamorgan, UK) and N. E. Day. The statistical analysis of the results of sampling an environment for a contaminant when most samples contain an undetectable level. Am J Epidemiol 97: 160–166, 1973.—When an environment is sampled for levels of a contaminant the results may be unsuitable for a conventional type of analysis, because of the proportion of samples which contain an undetectable amount. The problem is illustrated by means of some data on aflatoxin levels in diets; out of 2432 samples, in all but 124 there was an undetectable level. These data were analyzed using the gamma distribution. The importance of ensuring that the assumptions made in the analysis are consistent with the data is stressed. In particular, it is shown that the log-normal distribution would have been inappropriate for these data.