METHODS OF GROWTH EVALUATION OF JUVENILES OF CENTROPOMUS UNDECIMALIS IN EXPERIMENTAL CONDITION

Dietary requirement studies for fish culture often consider broken-line analysis or analysis of variance to estimate nutrient requirements in dose - response data. Lack of replication and temporal pseudoreplication (sensu Hurlbert, 1984) are two commonly found caveats to interpret data from experimental designs dealing with fish growth since experimental units are consecutively used over time in the analysis, reducing the independence assumptions. Furthermore, research resources often limit the number of experimental units (in these cases "a tank of fish") and hypothesis on the origin of the differences cannot be tests since random effects cannot be statistically eliminated by testing over the variance of experimental units. This is particularly true when sample size is limited and it is difficult to identify individual fish and their removal from each experimental unit, after measurements have being taken, is impossible. The purpose of this work is to present the results of a regression approach to assess dose-response analysis considering an experimental design to evaluate the growth of juvenile Snook, Centropomus undecimalis, reared with isoenergetic diets formulated with different levels of protein. Our computational approach requires sampling without replacement of growth data in order to obtain sample descriptors that are nearly independent in two consecutive times; hence assuring that the covariance between the two is minimum. This simplifies computations and interpretation of results ensuring that simple random samples are considered. Furthermore, this approach allow to deal with the fact that samples at the beginning of the temporal series usually have different variability values from those obtained at the end of the experiment due to processes related with the individual chronology of feeding which leads to the fact that few individuals growth faster than the rest, making unrealistic the assumption that variances within groups are equal. Resampling techniques allow the use of the regression model by using random subsets that fulfill the statistical assumptions.

Hurlbert, S.H. (1984) Pseudoreplication and the design of ecological field experiments. Ecological Monographs 54(2): 187-211.

Key words: Growth assessment; regression; pseudoreplication; resampling.