Laserfiche WebLink
' recovery of selenium ins iked sam les (al ae and rotifer: 86.0%, 5E=3.65; fish: 100.0%, <br />P P g <br />5E=0.413). <br />Statistical analysis <br />The data generated by the ELS and acute tests represent concentration-response <br />relationships. For both data sets, it was of interest to determine if the observed responses were <br />related to the experimental treatments (i.e., selenium exposure or fish species). Consequently, <br />regression analysis was used to estimate statistical models that describe each data set, and to <br />determine if fitted lines had similar slopes and intercepts (Oris and Bailer 1997). <br />Survival data were analyzed using logistic regression. Proc Genmod (with options <br />link=logit, dist=binomial, dscale; SAS 1993) was used to describe the response of survival as a <br />function of the independent variables. The full regression model had the form <br />logit(P) = Ro + R~ x~ + Rz xz + Rs xl xz <br />where p =survival proportion, logit (p) =natural log [ p / (1 -p)], Ro =intercept, <br />[3,, R, =coefficients for linear terms of main effects, x, =dissolved selenium concentration, xz = 0 <br />for razorback sucker or 1 for flannelmouth sucker, and R3 =coefficient for interaction of main <br />effects. A nonsignificant coefficient for interaction suggests that the concentration-response <br />relationships for both species are parallel and may or may not have similar intercepts. A <br />nonsignificant interaction coefficient also suggests that including the term in the statistical model <br />increases complexity, but does not explain additional variation in the dependent variable. <br />Consequently, when the interaction coefficient was not significant, it was omitted from the <br />statistical model and the analysis was re-run. Interpretation of the reduced model is straight <br />10 <br />