Prager and MacCall. Contaminant and climate effects on spawning of three pelagic fishes 



319 



Table 5 



Summary results of regression models of logarithm of recruits per spawner for three 

 coastal pelagic fish stocks off southern California. Predictor variables were parent stock 



size IP) and either principal components I W„ ; = 1 10 1 of climate data I models denoted ) 



or principal components ( C,, i = l 10) of climate data and data on contaminant loadings 



i models denoted V). Each model also included an intercept. Variable selection used an all- 

 subsets algorithm with the C. statistic; thus models may include variables not significant 

 at P<0.05. and all P values are considered nominal. 



Variables included in model (with sign of df of 



coefficient! and nominal probabilities oft- model, 



statistics for H„: coefficient = 0. error 



F Nominal 



statistic Prob > F 



R- 



Northern anchovy Engraulis mordax 



-P. 0.025; -W„ 0.13 2,15 



• -P, 0.17; -C 5 , 0.087; -C ; , 0.13 3,14 



Pacific sardine Sardinops sagax 



-W-„ 0.030 1, 15 



• -P. 0.0004; -C„ 0.0001; -C 5 , 0.002; 



+C 10 , 0.002 4, 12 



Chub mackerel Scomber japonicus 



-P. 0.0001; -W, 0.015; +W„ 0.021; 



+W 6 , 0.008; -VV 7 , 0.023; -W„„ 0.09 6, 23 



• -P. 0.0003; +C„ 0.094; -C 4 , 0.004 3,26 



4.20 

 4.14 



5.77 

 14.4 



0.036 

 0.027 



0.36 

 0.47 



0.030 0.28 



0.0002 0.47 



6.74 0.0003 0.64 



8.11 0.0006 0.48 



higher R 2 and nominal significance level* than the 

 model including only climate effects, it is impossible to 

 attribute this difference to the effects of contaminants 

 on recruitment. Given the small sample size and rela- 

 tively small difference in fit, it seems more logical to 

 attribute it to chance. 



The two models of anchovy spawning success can be 

 compared by plotting the correlations of the two corre- 

 sponding explanatory effects with the original explana- 

 tory variables (Fig. 4). For some variables (e.g., those 

 related to rainfall), correlations are similar between 

 models; however, many other variables fall into the 

 second and fourth quadrants of the plane, meaning 

 that they are positively associated with spawning suc- 

 cess in one model and negatively associated in the 

 other. For example, the two models assign opposite 

 signs to the influence of most upwelling variables. 

 Based on this inconsistent pattern of correlations 

 (Fig. 4) and the lack of nominally significant param- 

 eters (except for the stock-size parameters) in Table 5, 

 we conclude that these models identify neither climate- 

 nor contaminant-related variability in the spawning 

 success of this stock. 



Indeed, the major determinants of anchovy recruit- 

 ment strength remain to be discovered. Smith (1985) 

 speculated that recruitment might be controlled in late- 

 larval stages through "plasticity of the interaction be- 

 tween growth rate and survival." Peterman & Bradford 

 (1987) detected a decrease in larval survival associ- 

 ated with episodes of high wind speed; however, 

 Peterman et al. ( 1988 ) did not detect significant corre- 



*To emphasize the impossibility of determining true significance 

 levels, reported significance levels are denoted "nominal." 



R Rainfall 



O Sea Level 



O Sea Surface Temperature 



U Upwelling 

 • Contaminants 



-0.8 -0 4 0.0 0.4 0.8 



Correlation of "Combined" Effect with Variables 



Figure 4 



Comparison of two regression models of logarithm of spawning 

 success (recruits/spawnerl of northern anchovy Engraulis 

 mordax stock off southern California. Models include effects of 

 stock size and environment. Coordinates of a point are the 

 correlations of the models' explanatory effects (see text! with 

 an explanatory variable. For legibility, only category of vari- 

 able (point) is indicated. Vertical axis: correlations with model 

 estimated on climate data. Horizontal axis: correlations with 

 model estimated on combined climate and contaminant data. 

 Points in first and third quadrants of plane indicate agree- 

 ment between models as to a variable's effect; points in other 

 quadrants indicate disagreement. Plot suggests that neither 

 model offers a clear explanation of anchovy spawning success. 



