A potential method is presented for combining data collected as part of the CCAMLR ecosystem monitoring programme (CEMP) into a single index for each of the predator, prey and environment parameters. The proposed method is based on the usual theory of multivariate statistics and takes into account the covariance between parameters. The power of the statistical procedure recently adopted by WG-EMM for identifying anomalies in CEMP parameters is examined by means of simulation tests. The power of the procedure to detect anomalies was found to fall to low levels once more than a few anomalous values have appeared in the data. An alternative procedure, using baseline mean and variance estimates was found to have consistently better statistical power regardless of the accumulation of anomalies. An approach to the further development of CEMP indices is outlined.
Abstract:
The properties of a method for detecting anomalous years in CCAMLR index series are discussed. In simple cases this method involves comparing a standardized residual with a critical value obtained on the assumption that the series being considered consists of random values from a constant normal distribution. This idea is extended to situations (a) where the series values are still normally distributed, but contain a linear trend and autocorrelation, and (b) where the series values are from some other constant distribution. For cases like (a) # is proposed that the standardized deviations from a fitted linear regression line are compared with a critical value that is obtained on the assumption that there is no autocorrelation. This test is shown to have good properties even when autocorrelation is present, at least according to one model For cases like (b) it is proposed that a Box-Cox transformation to normality is applied before testing standardized residuals. This test has good properties for data from a wide range of distributions. Some examples are given to illustrate the performance of the proposed methods under various conditions.
