Laserfiche WebLink
<br />evaluations that \vere conducted previously into perspective, and to assess the applicability of <br />this statistical approach for this purpose. Second. an cvaluation of all the seeding programs was <br />conducted using ratio statistics (Gabriel, 1999), a statistical approach that is inherently more <br />prccise and more robust than the historical regression approach. Gabriel reported that ratio <br />statistics produces results that approximate those from re-randomintion for sample sizes of 100 or <br />more: however, experience by this author has shown that comparable results arc obtained for even <br />smaller samplc sizes. Ratio statistics were emphasized since it is computationally easier to apply <br />than re-randomil.ation and it providcs additional, useful infonnation about confidence intervals. <br />Thc main emphasis in the presentation ofthc results is on confidcncc intcrvals because they <br />infer a range within which thc true elTect lies whereas null hypothesis significance tests infer only <br />whether there is any effect at all (Gabriel, 2002; Nicholls, 2001), It should be noted, ho\vcver, that <br />statements of signilicance are implicit to contidencc intcrval statements. Saying, for example, that <br />therc is 90% contidence that the truc effect of seeding lics bctwecn a Single Ratio (SR) of SR901.O <br />and SR'l()1H is tantamount to saying that the confidence interval result is sih'1liticant at a 2-sided <br />levcl of significancc of 0,10 (l-sidt.x1 level of significance of 0,05), 'olUS, a 90% contidence <br />interval that includes an SR value of 1.0 indicates that the experimcntal rcsult is not signiticant at <br />a 2-sided Icvcl ofsigniticance of 0,10 (I-sided Icvel of significance of 0,05), <br />This is an exploratory study and, as such. it involvcs consideration of a number of <br />hypothcses/analyses. A fter applying the BonfelToni method (see, e,g" Gabriel, 2000) to partition <br />the usual significancc Icvel of 0.05 among aUthc specified tcsts, it is unlikely that any test could <br />satisfy the resuhing level of signiticance, On the other hand. with a largc number of tests, a fe\v <br />might yield significant results purely by chance. Thercfore, it is emphasized that the results of <br />the evaluations in this study arc used only as measurcs of the strenglh of lhe suggcsted sceding <br />elTect. Strictly speaking, the suggested cll'ects that are indicated must bc confimled through new. a <br />priori cxpcriments specitically designed to establish their validity. Ncvertheless, a suggcsted ctTect <br />of sceding might bc of considerable valuc to a watcr manager if, for example, it was known with <br />80% conlidencc that the rcsult of the seeding operation was positive. <br /> <br />4. Entlualion of Kings Ri\'er by the historical regression method <br />A rcgrcssion equation that predicts thc stTeamllow at the targct station, KGF, as a <br />function of the streamflow at the control station, MDP, was derived for thc 33.water year <br />historical pcriod (1922-1954) prior to the start of opcrational secding. The predictor equation for <br />thc water year. October-to-Septembcr, is <br />KGF = 3.832176 * ;VUW -160272.2 Acre-Feet (AF) <br />The correlation coetlicient is 0.936 (see Table 2), This is comparable to the corrclation <br />coetlicient of 0.947 betwcen thesc 2 stations obtaincd by lIendcrson (1966) for thc 25-year base <br />period of 1926-1950, When llcndcrson (1966) added a control station from the Kcrn River in his <br />evaluation of the tirst 10 years of the Kings River seeding program. hc obtaincd a multiple <br />regression cquation having a correlation coellicicnt of 0.989 with a smaller standard elTor of the <br />estimate than the regression cquation based on MDP alone. When IIcndcrson (2003b) cvaluatcd <br />the Kings River seeding program ovcr the first 47 years of the program, he could not usc the <br />Kcrn Rivcr control station because it had bccome a targct of operational seeding that bcgan in <br />1978, so he used Cottonwood Creck as a second control station. <br />Assuming that the relationship bctween the target and control station is valid during the <br />operational period. thc rcgression equation for thc watcr year Octobcr-to-Septcmber was used to <br />predict the streamtlow at thc target station that would have occurred in the absence of seeding for <br /> <br />36 <br />