HOW TO MEASURE NO EFFECT. PART III: STATISTICAL ASPECTS OF NOEC, ECx AND NEC ESTIMATES

Abstract

One of the principal aims of ecotoxicity tests is to determine the concentration level below which the test chemical will have no or at most a negligible effect on the test parameter. Nowadays, the NOEC (no observed effect concentration) is normally used as an estimate for this concentration. The NOEC has, however, several major drawbacks as a summary statistic: (i) it is based on wrong usage of hypothesis testing—the acceptance of a null hypothesis (no difference); (ii) the estimate depends on the accuracy of the experimental test—when the sample error is small, i.e. the test is performed accurately, the NOEC will be lower; (iii) the estimate depends on the sample size—the larger the sample size, the lower the NOEC will be; (iv) the NOEC is a test concentration; (v) the NOEC depends on the chosen significance level. An alternative is proposed, the ECx, i.e. the concentration causing an effect of x per cent: 1. To estimate the ECx a concentration–response model is needed. The value of x in the ECx estimation should be chosen so that the ECx estimate is not too model dependent. 2. In NOEC estimation a certain deviation from the control is accepted. For an ECx estimation to give comparable protection with an NOEC, the x must be chosen somewhat smaller than the maximum non-significant deviation in the case where the test is performed according to normal guideline procedures and the variance in the control is the mean variance for such tests. Another alternative is parametric NEC (no effect concentration) estimation. The NEC is the threshold concentration below which the test chemical will not induce an effect. (a) All NEC estimates will be model dependent. Therefore, they are not appropriate in standard ecotoxicity tests where data will seldom be sufficient for model verification. © 1997 John Wiley & Sons, Ltd.