The hierarchal log-linear analysis resulted in a model where the highest order interaction terms were depth ? size ? reef, activity ? substratum ? reef, activity ? time of day ? size ? reef, activity ? depth ? reef, and the sampling constraints of time of day ? depth ? reef and size ? reef (Table 4). Though not significant the effects activity ? time of day ? size ? reef were retained to test the effect of activity ? time of day ? size. The overall 2 fit of the model (x 5332.6, P50.006) was moderate, possibly indicating some (270) over-dispersion, though by fitting a model with so many parameters relative to data points, this may also reflect over-fitting. The significance of the highest order terms representing variations of other factors between reefs were tested against the meansquare deviance (Table 4)