Experimental design and statistical analyses of fish growth studies

摘要: article i nfo Every year, numerous studies are published that compare the effects of different factors on growth aquaculture fish. However, comparatively little attention has been given to experimental designs these — in how many rearing units should each treatment be replicated, fish tank (n) and data analysed. The reliability results increases with increased replication n. In reality, however, design must strike a balance between limited resources statistical analysis. A survey recent publications Aquaculture suggests, most (83%) apply triplicates an average 26 (range: 4 100). minimum difference can reliably detected analyses is determined by number replications treatment, n, variance treatments applied. present study, we accumulated information species estimate detectable assist researchers designing experiments effectively. These suggest similar for and, therefore, same (level n) suitable (MDD) mean body-mass groups typical study (triplicates, 25 variance) 80% power (less than 20% chance Type II error) around 26% grand mean. Increasing n from 100 will reduce MDD 19% mean, while further increase have lesser effect. quadruplicates or sextuplicates (with as 100), 16% 12% respectively. under 10% only possible when experiment selected within narrow size range variance. Simulations were performed, where samples (experiments) repeatedly drawn artificial populations identical distribution commonly used studies. Two had dose-dependent responses one population showed no response treatment. resulting was analysed mixed model ANOVA fitting either polynomials asymptotic models data. Contrary earlier suggestions, critical (minimum generate maximum response) estimated approached more closely did non-linear models.