Bayesian analysis improves experimental studies about temporal patterning of aggression in fish

Author(s)

Noleto-Filho, Eurico Mesquita ^{[1, 2]} ; dos Santos Gauy, Ana Carolina ^{[1, 2]} ; Pennino, Maria Grazia ^{[3, 4, 5]} ; Goncalves-de-Freitas, Eliane ^{[1, 2]}

Affiliation

^[1] Univ Estadual Paulista Julio Mesquita Filho UNESP, Zool & Bot Dept, R Cristovelo Colombo 2265, BR-15054000 Sao Jose Do Rio Preto, SP - Brazil

^[2] Sao Paulo State Univ CAUNESP, Aquaculture Ctr, Sao Paulo, SP - Brazil

^[3] Univ Fed Rio Grande do Norte UFRN, FEME, Dept Ecol, Natal, RN - Brazil

^[4] Univ Valencia, SMEG, Dept Estadist & Invest Operat, C Dr Matter 50, E-46100 Valencia - Spain

^[5] Ctr Oceanog Murcia, Inst Espanol Oceanog, C Varadero 1 San Pedro del Pinatar, Murcia 30740 - Spain

• Document type: Journal article
• Source: Behavioural Processes; v. 145, p. 18-26, DEC 2017.
• Web of Science Citations: 2

Abstract

This study aims to describe a Bayesian Hierarchical Linear Model (HLM) approach for longitudinal designs in fish's experimental aggressive behavior studies as an alternative to classical methods In particular, we discuss the advantages of Bayesian analysis in dealing with combined variables, non-statistically significant results and required sample size using an experiment of angelfish (Pterophyllwn scalare) species as case study. Groups of 3 individuals were subjected to daily observations recorded for 10 min during 5 days. The frequencies of attacks, displays and the total attacks (attacks + displays) of each record were modeled using Monte Carlo Markov chains. In addition, a Bayesian HLM was performed for measuring the rate of increase/decrease of the aggressive behavior during the time and to assess the probability of difference among days. Results highlighted that using the combined variable of total attacks could lead to biased conclusions as displays and attacks showed an opposite pattern in the experiment. Moreover, depending of the study, this difference in pattern can happen more clearly or more subtly. Subtle changes cannot be detected when p-values are implemented. On the contrary, Bayesian methods provide a clear description of the changes even when patterns are subtle. Additionally, results showed that the number of replicates (15 or 11) invariant the study conclusions as well that using a small sample size could be more evident within the overlapping days, that includes the social rank stability. Therefore, Bayesian analysis seems to be a richer and an adequate statistical approach for fish's aggressive behavior longitudinal designs. (AU)

• FAPESP's process: 13/09021-0 - Effect of aquaria water renewal on the aggressive behavior and recovering of the dominance hierarchy in Acará-Bandeira
• Grantee: Ana Carolina dos Santos Gauy
• Support Opportunities: Scholarships in Brazil - Scientific Initiation