Fig. 2. Description of number of parameters (K) and AICc for various models testing for an effect of bromocriptine treatment on kidnapping behaviour in emperor penguins. Model 1 is the general starting model, with an effect of treatment and behavioural state on survival and sighting probabilities and an effect of treatment on transition probabilities. Model selection: an arrow starting from a model indicate the application of a constraint to this model. This constraint is detailed above the constrained model, which is at the end of the arrow. AICc values were used to determine which model provides the best description of the data based on the fewest parameters (AICc=AICc of the starting model–AICc of the constrained model). AICc values >2 indicate that the constrained model is preferable. AICc values <2 and >–2 indicate that models are fairly similar and the model including the fewest parameters was selected. AICc values <–2 indicate that the starting model is preferable. Filled and broken arrows indicate, respectively, that the constraint was selected or not. Each model was defined using the following notations: S, survival probability; P, sighting probability; ^ab transition probability from state a to b; 1, non-kidnapping state; 2, kidnapping state; ^12br/^12co, ¹² within the bromocriptine/control group; treatment/state: effect of treatment/behavioural state on a given parameter; prl: effect of initial level of PRL on a given parameter; ¹²_first¹²_2-7, effect of time since treatment on ¹², which can vary between the first day and the last days of the study period. In models testing for an effect of PRL levels or time since treatment on ¹² within one group (bromocriptine/control), ¹²_treatment was dissociated and we used this notation ^{12br 12co}. Models A and B are two nested models allowing us to test an effect of treatment on kidnapping behaviour during the first day of behavioural monitoring.