For the 54 taxa selected for the seasonal effects analysis, ad- 

 ditional analysis was necessary to maximize data utility. A bias 

 factor existed if a particular species occurred over a limited part 

 of a permanent transect, and its density was calculated by dividing 

 total abundance by n, the total number of quadrat samples taken in 

 the permanent transect. This provided a value for mean density 

 over the entire island; however, this would be justified only for 

 species ubiquitously distributed [i.e., over the entire length of the 

 transect). The distribution of only one species, the starfish, 

 Patiria miniata, approaches this (see Figs. 17 to 20). A better 

 approach would be to divide total abundance by the niiraber (n] of 

 quadrats where the species may reasonably be expected to occur, and 

 express mean density with reference to the parts of the island over 

 which the species actually occurs (or those associations of which it 

 is a member). Mean densities of each species may be more meaning- 

 fully compared to resolve seasonal differences using this approach. 

 Briefly, the mechanics of this data processing operation involved 

 scanning the raw data tables to bracket the upper and lower occur- 

 rence limits for each species and then logging onto computer keypunch 

 forms the frequency of every density value observed (including zero 

 density values for quadrats lacking a given species, but falling 

 within its range of occurrence) . 



Before the data were subjected to parametric statistical analy- 

 sis, it was necessary to perform data transformations to normalize 

 the data. For species whose densities were recorded as percentage 

 coverage, the values were transformed to angles through the use of 

 the arcsine transformation {6 = arcsin Jp~, where p is a proportion) 

 This transformation rendered a distribution of percentages or pro- 

 portions more nearly normal by stretching out both tails of the dis- 

 tribution and compressing the middle values (Sokal and Rohlf, 1969). 

 Numerical densities were subjected to the square root transformation. 

 Because zero values were frequent in the data, the computer was pro- 

 gramed to add 0.5 to all values before data t ransformation. The 

 transformation was then of the form ^Y + J^ (Sokal and Rohlf, 

 1969). 



The actual calculations of the means used all the raw data for 

 variances to be calculated for each of the 54 taxa examined. 

 Seasonal means (data for all four sides lumped) were first tested for 

 significant differences by performing an F test (variance ratio test) 

 to determine whether variances for two seasons under comparison were 

 equal. If the F test was nonsignificant (variances probably equal), 

 the following student's t test for differences between seasonal 

 means was applied (Sokal and Rohlf, 1969) : 



