Laserfiche WebLink
<br />18 <br /> <br />TECHNIQUES OF WATER-RESOURCES INVESTIGATIONS <br /> <br />. <br /> <br />The table presented by Dixon and Massey <br />(1957, table A-5, p. 384) gives values for one- <br />half of the distribution. For 95-percent limits <br />there would be 0.025 in each tail, and the value <br />is taken in the colunm headed to.,," Notice that <br /> <br />The confidence limits are: <br /> <br />for infinite degrees of freedom the t and normal <br />distributions are the same. In the normal <br />distribution, 1.96u on each side of the meau <br />includes 95 percent of the items. In table A-5, <br />1.96 is listed under to..,.. <br /> <br />b,- (t".O.05) (8,,) <13,<b,+ (t",o.oo) (8,,) <br />1.08851- (2.03) (0.0344) <1l2<1.08851 + (2.03) (0.0344) <br />1.0187<13,<1.1583 <br />-0.03938- (2.03) (0.0394)<.8,< -0.03938+ (2.03) (0.0394) <br />-0.1194<.8,< +0.0406 <br />1.42885- (2.03) (0.1234) <13,< 1.42885 + (2.03) (0.1234) <br />1.1783<.8,< 1.6793 <br /> <br />.8 is considered the true slope. Therefore the <br />confidence limits give the range within which <br />.8 lies with 95-percent probability. In this ex- <br />ample the limits of Il2 include zero. This indi- <br />cates that .8, is not significantly different from <br />zero at 95-percent level and that the parameter <br />A should be eliminated from the regression. <br /> <br />Regressions having various numbers of <br />independent variables <br /> <br />Examples have been given of computations <br />for regressions having one and three inde- <br />pendent variables, and the normal equations <br />for a regression of two independent variables <br />have also been given. The method of solution <br />involving two variables is similar to that for <br />three independent variables, but is much <br />shorter. <br />Normal equations for regressions of four or <br />more independent variables have been given <br />by Ezekiel and Fox (1959, 1'.181-183). Because <br />computation of such regressions on a desk <br />calculator is very time consuming, digital com- <br />puters are being used. <br /> <br />Use of digital computers <br /> <br />Programs for regression computations are <br />available for most computers, and regressions <br />of more than two independent variables should <br />ordinarily be made by digital computer rather <br />than on a desk calculator. Simple regressions <br />and regressions of two independent variables <br />may be made quite rapidly on a desk calcu- <br /> <br />- <br /> <br />lator; use of a desk calculator for computations <br />of these sizes may be advantageous. <br />Regression programs for digital computers <br />vary but usually require listing of the data in <br />floating decimal notation. These values are <br />then punched on cards which are entered in <br />the computer. Results are printed by the com- <br />puter. A wide variety of options as to output <br />is available. Detailed instructions for prepara- <br />tion of data and instructions to the computer <br />should be obtained for the particular computer <br />and program to be used. <br />Although no knowledge of regression analysis <br />is necessary for preparing data for a computer <br />program, some experience is needed to appraise <br />the results. Opportunities for errors to be intro- <br />duced into the process exist in the listin'g of <br />the data and in its transferral to cards. <br />Questionable results may also be obtained if <br />too few significant figures are carried through <br />the computations. Only a person who has made <br />regression computations the hard way can ade- <br />quately judge whether the results of a regres- <br />sion analysis by digital computer (or any other <br />method) are correct. The availability of digital <br />computers has permitted ready computation <br />of regressions using many variables, which has <br />sometimes led to substitution of the computer <br />for the analyst's brain. The problem should be <br />solved by the analyst; the computer does the <br />arithmetic. <br /> <br />. <br /> <br />Application of the regression method <br /> <br />An analytical problem to be solved by <br />regression involves (1) selection of factors which <br /> <br />. <br />