Homogeneity of H2SD counts increase with cotton mixing - Consequences for reference cotton preparation

High-speed stickiness detector (H2SD) is a measuring instrument that counts sticky spots revealed by pressure and temperature. Standard cottons of known stickiness are needed for inter-laboratory tests and routine calibration of H2SD. An important issue is the variability of the measurements in a batch of standard cotton. The typical size of a batch is a bale. Two experiments were carried out to determine how much H2SD counts vary within batches of various sizes, and to what extent mixing the cotton reduces this variability. The first study was made on five batches of 2.4 kg of cottons, covering a wide range of stickiness caused either by Aphis gossypii or Bemisia tabaci; the second study was made on an entire bale, whose heavy stickiness was caused by Aphis. The small batches were opened 1, 2 or 3 times using a minicard without flats, and then optionally carded. The entire bale was mixed 1, 2 or 3 times with a beater and ultimately carded. There were unmixed controls for both experiments. The minimum achievable variability of a count is known from spatial distribution theory: if sticky spots were randomly scattered in a perfectly homogeneous cotton, the number of spots in a specimen would follow a Poisson distribution, with a variance equal to the bale mean. Thus, nor standard analysis of variance nor coefficient of variation are suitable to analyze counts: the first one requires a constant variance, and the second one requires the variance to be proportional to the square of the mean. A log-linear mixed model was used to analyze the data, in order to estimate different components of variability, namely between samples in a batch, and between specimen within a sample. In both experiments, the between sample variability was reduced to a negligible value after two mixing operations (whether opening or carding or beating). The within sample variability was less affected by the mixing operations and stabilized at 1.3 times that of the Poisson distribution after two or more mixings. We therefore recommend mixing a bale twice before using it as a standard for stickiness. The statistical distribution of the number of sticky spots in such a bale is yet to be found, in order to derive confidence intervals, repeatability and reproducibility. Meanwhile, one can provisionally enlarge Poisson confidence intervals by 14%. We wish to thank Cotton Inc. (USA) for supporting this project, and Cotona (Madagascar) for performing a part of the experiments. (Résumé d'auteur)