Laserfiche WebLink
Figure 2. Scatter plot of annual target-total versus control-total rainfall, with <br />corresponding LAD regression line. <br />Next residual Euclidean distances from the multidimensional line are computed <br />for each of the 53 annual points, and classified into two groups – corresponding <br />respectively to the “historical” years 1950-1975 and the NDCMP years 1976-2002. An <br />MRPP test is then applied to those two groups of residuals to determine whether any <br />difference between the groups is greater than would be likely to occur if the groups of <br />years had been established by random assignment from the set of 53. If the P value (the <br />probability of a test statistic as small as, or smaller than, that actually occurring) under <br />random permutations of the assignments is less than 0.05, or perhaps 0.10, the difference <br />between the two groups of years can be inferred to result from the seeding operations. <br />The P value for this primary analysis turns out to be 0.322 – i.e. the probability of <br />a test statistic as small as, or smaller than, that actually observed is 32.2%. This result <br />cannot be considered significant by any of the usual measures, so the major result of this <br />analysis is that no significant indication of any effect of the NDCMP seeding on the <br />rainfall in the target area can be identified. <br />5. Additional exploratory analyses <br />The accumulated database permits a variety of additional exploratory analyses. At <br />the outset it should be understood that (1) the absence of any significant indication of a <br />primary effect on the seasonal rainfall makes it unlikely that any significant effect will be <br />8 <br />