312 



Fishery Bulletin 104(2) 



(2000). Spatial resolution to the particular ocean basins 

 in this stu(iy was defined by the information I was able 

 to obtain, as were the choice of species and life history 

 traits to analyze. Although other information for these 

 and other species may be currently available, I limited 

 my data to those found in primary peer-reviewed litera- 

 ture. When mean fecundity values were unavailable, the 

 mean was assumed to be the middle value of the fecun- 

 dity range given. Phylogenetic variance was controlled 

 because I strictly evaluated intraspecific comparisons. 



Initial data exploration was performed by visualizing 

 intraspecific gender-based pairwise comparisons by area 

 with dot plots (Fig. 1). Within each species and gender, 

 the outcome of each comparison (i.e., the value for a 

 particular trait in one area greater than, less than, or 

 equal to that of another area) was evaluated. An overall 

 relationship among areas for each life history trait was 

 then constructed as a composite of each pairwise result. 

 The purpose of this exercise was to visually explore the 

 data and assess whether any intraspecific life history 

 patterns by area were apparent. 



A GLM framework was then used to construct simple 

 models that quantitatively relate the effect of certain 

 factors (e.g., area, gender, or taxonomic level) to a re- 

 sponse variable — in this case to a particular life history 

 trait. The flexibility of the GLM framework also al- 

 lows one to consider non-normal response distributions 



while maintaining the advantages of linear regression 

 (Venables and Ripley, 2002) by means of a link func- 

 tion relating the response variable mean to the linear 

 predictors. I had no a priori knowledge of the variance 

 structure for each life history trait (some of which will 

 not have variance, given that they are maximum re- 

 corded values), so both lognormal (with an identity 

 link) and gamma (with a log link) distributions were 

 considered because of their appropriateness to continu- 

 ous and nonzero data. Akaike's information criterion 

 (AIC; Burnham and Anderson, 2002) was used to select 

 among models, with the lowest AIC value indicative of 

 the most appropriate model among all considered. Mod- 

 els explored included all combinations of the following 

 factors: area, gender, and the taxonomic levels of spe- 

 cies, genus, or family. Models that included the interac- 

 tion between gender and area and gender and taxonomic 

 level were also considered. Area effects and predictions 

 of life history values among models with the use of the 

 lognormal and gamma distributions were similar, but 

 models with gamma error structure resulted in the low- 

 est standard errors for area effects; thus a gamma error 

 structure was ultimately chosen for each model. 



Resultant model effects were used to compare ar- 

 ea effects and to predict species- and gender-specific 

 life history trait values for each area. The predicting 

 models were then applied to two species (S. acanth- 



