The Minimum Significant Difference at the NOEC calculated with a non-parametric test.

Nelly van der Hoeven

ECOSTAT
Vondellaan 23
2332 AA Leiden
The Netherlands
NvdH @iecostat.nl

For many ecotoxicity tests, the statistic summarising the effect size is the so-called No Observed Effect Concentration (NOEC) (or ~Dose (NOED)), which is the highest test concentration (or dose) without a statistically significant result. The NOEC has many drawbacks ([1], [2]), one of these being that the NOEC will be larger if the experiment is performed with less replicates or under less controlled conditions with a larger within-replicate variance. Therefore, when the NOEC is used, it should be accompanied by some measure that indicates how much difference in response should be observed minimally for a response to be found significantly different from the response observed in the control treatment. ([1], [2]). The Minimum Significant Difference (MSD) is such a measure.

Many guidelines prescribe that the MSD or the confidence interval for the deviation from the response observed in the control treatment must be given ([3], [4]). In standard statistical software, the MSD can only be calculated for parametric statistical tests based on the normal probability distribution (e.g. Student's t-test and Dunnett's test). In the poster, a method will be presented to calculate the MSD if the non-parametric Wilcoxon rank-sum test is used to compare the observed responses at each of the concentrations with the responses observed in the control treatment.

[1] Chapman, P.F. et al., 1995. Asking the right questions: ecotoxicology and statistics. In: Report workshop Royal Holloway Un. London, UK
[2] van der Hoeven, N., 1997. How to measure no effect, part III: Statistical aspects of NOEC and ECx estimates. Environmetrics 8: 255-261.
[3] USEPA, 2000. EPA/833/R-00/003.
[4] OECD, 2006. ENV/JM/MONO(2006)18

In: Proceedings of the 30th Anniversery Meeting of the Netherlands Society of Toxicology, june 2009. p. 122