The log-linear modelling of starfish activity comprised two stages. Firstly a series of models were fitted which included factors representing size categories of starfish, type of substratum, time of day, depth at which the starfish was observed, and reef identity. The objective of this analysis was to establish which of the factors affected location and starfish activity, and to demonstrate that the variation of these effects between reefs was small compared to the overall effects. The data for this analysis were restricted to the populations from the six reefs on which the largest numbers of observations were made (Table 1). Of these six reefs, five included both day and night surveys. Animal size was coded into three groups: ,25 cm, 25–40 cm and .40 cm. Substratum was coded as hard corals or other. Time of day was coded as day or night, and depth was coded as ,3 or $3 m. The reef populations were represented by a 6-level factor. A hierarchal log-linear analysis using backward elimination was used to derive a minimal model
(Bishop et al., 1975).