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



327 



known only in a few simple cases. By analogy to the 

 results in bivariate linear regression, we suspect that 

 coefficients of our regression models are biased towards 

 zero. The estimated correlations with individual ex- 

 planatory variates, including contaminants, may also 

 be biased towards zero because of the smearing of the 

 relationship by errors in the predictors. 



Further complication is caused by the strong corre- 

 lations between potential explanatory variables reflect- 



ing climate and those reflecting contaminant loadings 

 (Prager & MacCall 1990, Table 6.1). A few examples 

 are the correlations between annual dredging remov- 

 als and total rainfall in San Diego (r=-0.32), PCBs 

 loadings and July-August SST at San Luis Wharf 

 (r=0.63), and lead loadings and May-June upwelling 

 at 39°N, 122°W (r=0.79). This collinearity makes it 

 nearly impossible to distinguish, by statistical means, 

 the effects of climate from those of contaminants. 



