12.01.2013 Views

Q2 Z2,(Q2) Z2(Q2) - Institute for Water Resources - U.S. Army

Q2 Z2,(Q2) Z2(Q2) - Institute for Water Resources - U.S. Army

Q2 Z2,(Q2) Z2(Q2) - Institute for Water Resources - U.S. Army

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

These results are disappointing, but should have been expected.<br />

On the one hand, one might think that each X ij is normally distributed<br />

over the totai-sample:of the orie.ns óf the three modes. There are no<br />

a priori reasons to expect another distribution. There<strong>for</strong>e, one might,<br />

speculate that they are normally distributed as "pure" random variables<br />

are. But, if the total sample is normally .distributed, each separate<br />

,<br />

sub-Sample :cannot be so distributed'. This can be seen if one considers<br />

that—the X-'s j which make up the ith mode sub-sampl correspond largely<br />

to the X. 's in one of the tails of the total sample. distribution. Ii<br />

the extreme case where the's would be insignificant, the distribution<br />

of theX ij 's would correspond to a chunk of the <strong>for</strong>mer distribution and<br />

could not be normal. On the other hand, these observations are also<br />

conditionalby the actual geographic circumstances of the experiment,<br />

and some of the possible value ofX ij 's within their observed range are<br />

not at all represented in the sample. These distributions are essentially<br />

reflecting the geographic pattern of corn production.<br />

This situation raises a double question. First, if it is not<br />

possible to justify the hypothesis of normality - on empirical or theoretical<br />

-<br />

grounds - is it still possible to use it as a purely ad hoc device? This<br />

is the question of the robustness of the proposed statistical procedure.<br />

Second, is there not another method, a distribution-free method, which -<br />

could be used in place of the procedure with normality assuMption? ,<br />

As an answer to the first question, M.G. Kendall'nd A. Stuart<br />

propose to apply the discriminators, or classification rules, to each member<br />

of the samples on which they are based and to observe the errors in the<br />

samp1es. (1) This procedure has been applied to the non-weighed January sample,<br />

and the results reported in Table VIII.<br />

(1) M.G. Kendall and A. Stuart 'The Advanced Theorx_of Statistics(, Vol. 3,<br />

1966, pp. 324-.325.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!