Laserfiche WebLink
270 M. J. JENNINGS, L. S. FORE AND J. R. KARR <br />Ir <br />60 <br />48 <br />(D <br />0 <br />U <br />U) 36 <br />ILL <br />24 <br />12 <br />® Kent <br />® Nick <br />? VI Chic <br />0 Watt <br />Figure 3. Reservoir fish assemblage index (RFAI) scores for monthly supplemental collections based on nine metrics (defined in text), <br />excluding FHAI, PDIS and ABUN. Standard deviations were calculated from bootstrap estimates of the variance <br />reservoirs changed monthly. For example, the rank order for Kentucky changed from first to last from <br />September to October and last to first from November to December. <br />Resampling provided an estimate of the measurement error associated with each site-month combina- <br />tion. The measurement error, though rather underestimated, is not large relative to the variability observed <br />LO <br />Measurement Error <br />Sampling Error <br />c ? <br />0 <br />m co <br />0 <br />P <br />v <br />c <br />ft3 N <br />In <br />O <br />Figure 4. Comparison of variance estimates from bootstrap samples generated with autumn 1992 supplemental data from four <br />reservoirs-Chickamauga, Kentucky, Nickajack and Watts Bar-and four sample months-September to December. Measurement <br />errors are variance estimates from bootstrapping at the level of the individual. This procedure is based on 10 electrofishing runs and <br />provides an assessment of precision (measurement error resulting from construction of the index how data are used to generate an <br />RFAI score). Sampling errors are variance estimates from bootstrapping at the level of the electrofishing run. Ten runs sampled <br />with replacement from the 20 generated during autumn 1992 'supplemental' samples. This procedure provides an assessment of random <br />sample error caused by differences in location. Note that sample error is generally larger than measurement error <br />Sep Oct Nov Dec <br />I- CHIC -? ?- KENT -? [-- NICK -? [-- WATT