Making sense of nonsense: Using regression analysis to deal with highly variable data collected from a yellowtail kingfish (Seriola lalandi) digestibility experiment

When conducting digestibility experiments with fish many researchers encounter problems that result in erroneous digestibility coefficients. Erroneous digestibility coefficients result from variability in the raw data which is amplified by the formulae used to calculate them. Variation can stem from natural differences in the digestibility of the same diet between individuals or groups of fish. It can also creep insidiously into studies due to poor preparation and mixing of ingredients or problems with the collection of faecal material or the veracity of analytical results. Therefore, it is fairly common to expect variability in data collected from digestibility trials with fish. But what can be done about it? In this paper, we present an experiment with yellowtail kingfish (Seriola lalandi) that was done to determine whether the digestibility of extruded wheat (EW) was affected by its inclusion rate (10, 20, 30 or 40% diet(-1)). The experiment, based on the indicator-ratio method, was conducted without incident, but the raw data on the nutrient and marker concentration of faecal samples was variable. We dealt with this problem by using linear regression to estimate more reliable analytical values for faecal samples. These values were used to recalculate logical digestibility coefficients for EW. Using this approach, we show that dry matter (approximate to 57%-40%), carbohydrate (approximate to 48%-27%) and gross energy (57-42%) digestibility of EW decline as its inclusion rate rises, whereas the digestibility of protein (approximate to 84%) and the digestibility of fat (approximate to 94%), remain reasonably constant. We validated the regression approach applied to yellowtail kingfish by examining published data from a similar digestibility experiment on Australian snapper Pagrus auratus. A regression approach was useful in reducing the variability in our raw data because the design of our experiment approximated a dose-response relationship. Designing digestibility experiments using a dose-response approach may prove useful in overcoming the inherent variability often encountered in these types of experiments.