Laserfiche WebLink
Mr. Wayne Sowards ~ 2 • 20 September 1982 <br />(1) They are not significantly different for any year. This means that <br />the two areas are alike and you can just use the reference area and <br />forget about the pre-mine area. <br />(2) If some are not significant and some are significant, then the <br />composite p* value would be the following: <br />m <br />p* = PCXZm > -2 Ellnpj] <br />j <br />that is the probability of an x2 distribution with 2m d.f. (where <br />m = number of p's Cyears7 you have calculated) to be greater than <br />the number of <br />m <br />-2 E lnpj <br />j =1 <br />where pj is the p value for year j (that is the exact probability <br />value for the t-test on year j). Then if p* is, let's say, greater <br />than 0.05 (or whatever level you are working with), you accept the <br />hypothesis that the two areas are similar and proceed as in (1). <br />(3) If for all the years the t-test indicates that the two areas are <br />not similar or p* is less than 0.05, for example, then you are <br />going to have to use another procedure like the regression you are <br />proposing. However, this procedure has an important draw back; you <br />cannot predict beyond the range of X values you use to build the <br />model (i.e., if the cover of the reference area during the five or <br />so years of the data ranges from 10-20% and you hit a year in the <br />future with more than 20% or less than 10%, then you cannot use <br />your regression line because the model you built was specifically <br />fitted for ranges between 10-20). <br />This summarizes our thinking at the present time and hope that it will <br />be of some help. If you want to talk further, please give me a call. <br />Sincerely, <br />~G~ <br />Edward F. Redente <br />Assistant Professor <br />EFR/sjd <br />