Experimental design, statistical analysis and modelling of dietary nutrient requirement studies for fish: a critical review

A limited survey of published reports on dietary nutrient requirement estim ates for fish (three journals, 46 papers) indicates that broken-line analys is and analysis of variance (ANOVA) are often used to estimate nutrient req uirements from dose-response data. The application of regression models usi ng published treatment mean values to re-evaluate estimates was possible us ing 33 of these reports. Re-evaluation suggests that the broken-line method and ANOVA frequently underestimate the requirement. Regression produced es timates that averaged approximately twice, but were up to five times the pu blished requirement. Additional problems that prevented re-evaluation or pr oduced errors in the original estimates were: failure to include nutrient l evels high enough to produce a maximum response, failure to space nutrient input levels closely enough to adequately model the dose-response relations hip, an apparent failure to screen data before analysis, and insufficient m odel diagnosis. Examples from the literature are presented to illustrate ho w design, method of analysis and the choice of model affect the requirement estimate. The effects of measurement frequency and the experiment duration on the resulting requirement estimate are discussed. A set of protocols is presented to help improve nutrient requirement estimates.