Orientation

NOEC, ECx and more - the statistical evaluation of biotests.

Introductory course on statistical methods in ecotoxicology, especially for single species tests.

Target group

The course is suitable for anyone who deals with evaluating biotests in their daily work, whether through the direct, practical evaluation of data sets in the laboratory or a subsequent test of the results within the framework of a authorization  procedure.

Pre-requirements and aim of the course

At the start of the course, the theoretical foundations of statistical tests are presented by way of simple relations so that all the participants, irrespective of their previous knowledge, are starting from the same base. If you already have some knowledge in this area, then the course offers you a good opportunity to consolidate your knowledge. If up to this point you‘ve had little opportunity to become involved in this area, then the course will give you an effective introduction and a brief overview.
The aim is to convey the essential basic principles of all statistical tests and to put the participants in a position to choose appropriate statistical tests, to carry these out professionally and have the expertise to evaluate the results.   

Concept

The course deals with the statistical evaluation of single species tests. It includes an introduction to the foundations of statistical tests. Special attention is paid to the scaling of data as the choice of an appropriate test depends to a large extent on the underlying type of variables (e.g. quantal or metric).  Basic terms such as sampling distribution, laterality, level of significance etc. are explained and typical statistical variables are defined. Building upon these basic principles, a general schema for selecting and carrying out statistical tests is presented.

Various tests for biotest evaluations are discussed. At the same time, emphasis remains on the practical application of the tests, that is to say, specific questions such as “when pairwise, when multiple tests“, outlier problems, which test for for which question, test selection and performance.

A further main topic is data modelling (for example dose-response functions). The possibility of adapting functions is addressed, during which standard models for dose-effect functions (for example  probit, Weibull and logit models; multiple parameter dose-effect functions) and the methods for their adjustment (linear and non-linear regression) are presented and discussed.

The theory becomes clear and is consolidated by way of practical exercises, during which the participants use the ToxRat Professional software. If desired, actual in-house data sets can be discussed with the course participants.

Contents

The following contents are included in a two-day training course.
We will gladly take different parts from the course and put them together to form a shorter programme. No matter which aspects you require, the theory is always illustrated by specific examples.

• Data Scales
• Type of Endpoints in Ecotoxicity Tests
• Probability and Probability Distributions

• Introduction into Sampling Theory
  (Central Limit Theorem)
  What is a Statistical Test? What is a Confidence Interval?
• Hypothesis Testing
• Introductory Example – z-test and t-test
• Comparison with a Standard

• Outlier tests (Dixon/Grubbs test)
• Pre-testing
• Normal distribution (Shapiro-Wilks, Kolmogorrov-Smirnov test, Range-to-standard- deviation test)
• Variance homogeneity (Cochran, Levene, Bartlett)
• Comparison with a standard (t, Wilcoxon-, c² test; one sample tests)
• Pair-wise comparisons (t test, Wilcoxon-Mann-Whitney rank sum test, c²-fourfold test, Fisher exact binomial test)
• Limit test; testing the solvent control against the blanc control
• General detection of an effect
   (ANOVA, Kruskal-Wallis test, c² contingency table, exact contingency table)
• Determination of a toxic threshold (NOEL/LOEL) (multiple tests)
• Sample size (replication) and power of tests

• Comparison of the ECx and NOEC
• Dose/response relationships for quantal responses
  • Probit-, Logit- and Weibull functions, fitted by max. likelihood regression
  • ECx calculation
• Dose/response relationships for quantitative responses
  • Modified Probit-, Logit- and Weibull functions, fitted by max. likelihood regression
  • Linear models, threshold models, nested nonlinear models
  • ECx calculation
  • Calculation of confidence limits