Main content area

Logit models for evaluating spawning performance of channel catfish, Ictalurus punctatus (Rafinesque)

Quintero, Herbert E., Abebe, Asheber, Davis, Donald Allen
Aquaculture research 2007 v.38 no.6 pp. 635-643
diet, linear models, fecundity, feeding frequency, logit analysis, spawning, hatching, Ictalurus punctatus, fish, normal distribution, case studies, breeding stock
Broodstock evaluations are often measured by variables such as spawning success, fecundity, fertilization and hatching rates, usually expressed as percentage values. Outcomes are generally analysed as continuous random variables, assuming that they follow a normal distribution. Ordinary linear regression models (e.g. analysis of variance) as well as χ² analysis are typically applied. However, these models may not be the most appropriate as a number of test criteria may not be met. For example, spawning success outcomes are inherently discrete and non-negative data and hence their distribution is not likely to be normal. As these models may not be the most appropriate, a case study using logit analysis as an alternative method for the evaluation of this type of data is presented by considering the response as binary data (spawned versus did not spawn). An exact version of logit analysis was performed due to the sparseness of the data. The results demonstrate that appropriate statistical models provide better insight into the cause-effect relationships that exist between control variables and the dependent variable (likelihood of spawning in this case). As would be expected, each strain of fish responded somewhat differently to the test variables. Changing the protein level of the diet from 32% to 42% or increasing the feeding frequency from three to six times per week either did not influence spawning or negatively affected spawning respectively. Additionally, older fish performed better than younger fish and the early spawning period was better than the later spawning period, regardless of strain. These responses, however, were only detected using logit analysis, which is a more sensitive test and would thus be recommended for this type of data.