Table 1.

Log-linear models used to test patterns of habitat use (adapted from Munday 2000). Hierarchical models were tested sequentially until there was no further improvement in the fit of the model to the data. Two models were considered as alternative conditional models (3a and 3b) in the progression from model 2 → 4

Model	Factors included	Hypothesis tested
1	Site × year	Coral use is proportional to availability
2	Coral + site × year	Corals used disproportionately to availability and the pattern uniform among locations and years
3a	Coral × year + site × year	Corals used disproportionately to availability, but the pattern changes between years
3b	Coral × site + site × year	Corals used disproportionately to availability, but the pattern changes between locations
4	Coral × year + Coral × site + site × year	Corals used disproportionately to availability, but the pattern changes between locations and between years