Laserfiche WebLink
seine sample area during the 1994 to 1996 period so density comparisons were not possible <br />between the two time periods. Statistical analyses were possible only for electrofishing CPUE <br />data in Lodore Canyon between the two periods because there were no 1994 to 1996 data for <br />Whirlpool Canyon. We plotted changes in individual species abundance in each Lodore Canyon <br />reach between the two periods, but did not conduct statistical analyses of reach or annual <br />differences within studies; we were mainly interested in broader scale abundance patterns <br />between the two periods. We also compared size-structures of selected taxa sampled in the two <br />periods in each of the four Lodore Canyon reaches for native fishes and a few potentially <br />problematic non-native species. <br />We then analyzed effects of season, reach, year, and turbidity on electrofishing and seine <br />data to document study area scale variation in species richness and fish abundance. We did not <br />include flow level as an analysis variable because it varied little during our sampling in the base <br />flow period. Water temperature was not included because these differences should be reflected <br />as the season class variable, warm in summer and cool in autumn. Statistical analyses were <br />conducted with a general linear model (GLM) approach (Proc Genmod, in SAS), using a <br />negative binomial data distribution for counts of fish in a sample after adjustment for effort (fish <br />captured per unit effort), and a log link. A main advantage of this approach over traditional <br />analysis of variance (ANOVA) applications that assume normal distributions is that the negative <br />binomial data distribution (number of fish captured per sampling effort) allows for a large <br />number of zero observations in the data set without transformations (or back-transformations). <br />Normality assumptions using transformations for zero-rich fishery data are rarely met in <br />traditional ANOVA applications and, as a result of inflated variance estimates, significance tests <br />among groups are usually conservative (pers. comm., G. White, Colorado State University). <br />This approach also allowed inclusion of categorical variables such as turbidity in analyses to <br />examine catch rate variability. Significance tests for this approach used a chi-square test rather <br />than the more familiar F-test used in most ANOVA applications. Differences among capture <br />rates were tested with a least squares means procedure. We then plotted density data for native <br />or abundant non-native species captured in seine samples to delineate variation at reach, season, <br />12