WMOD00372

• Project Name: High Plains Cooperative Program
• Title: The Design of HIPLEX-1
• Date: 4/1/1979
• Weather Modification - Doc Type: Report

that the lowest tilt radar beam is at a level where the hyd~o-
meteors are predominantly liquid (see subsection VIII.H.).
Therefore, all primary response variables that are based on 'the
SWR-75 radar data, with the exception of TFE, will be stratified
according to whether or not the center of the lowest tilt radar
beam is at or below the +10 Oc level.

I

Physical partitioning of the experimental units will probably
occur in various a posteriori investigations. These investiga-
tions may ultimately lead to refinements of the hypotheses. ;

3. Avoidance of bias. - Efforts to avoid bias include double-
blind randomization, use of a placebo mechanism, and block
randomization.

.,

The double-blind approach is intended to keep (1) the aircrews and
(2) those directly involved in the construction of the experiment's
data base ignorant of the actual treatment applied. Two separate
randomizations will be used: one to select which seeding-agent
hopper on the seeding aircraft is "live" and the other to select
which hopper is activated during the treatment pass.

The placebo to be used in HIPLEX-1 involves a deactivated seeding
hopper (containing dry ice) and a noise source from the aft of the
aircraft to mask any sounds that might disclose to the crew
whether the "live" or deactivated hopper has been selected.' More
information is found in appendix D.

Block randomization of experimental events is used to insure that
about half of each class of clouds (A-I, A-2, and B) is seeded.
Separate sequences of random decisions will be applied to each
class. The field crews will not know the number of events in each
block.

4. Statistical techniques. - The primary inferences of the
HIPLEX-1 experiment will be based on permutation, rerandomization,
and nonparametric statistical techniques applied to univariate and
multivariate data sets. A recently developed class of multivariate
techniques that avoids the assumption of multivariate normality is
termed MRPP (multiresponse permutation procedures). MRPP appears
to be the best available tool for detecting the anticipated
effects of seeding on the multivariate measurements associated
with HIPLEX-1.

.-

It is likely that unanticipated questions will arise during the
course of the experiment. Unrestricted a posteriori analyses, as
appropriate, will be conducted to supplement a priori analyses.
These analyses will draw from suitable parametric, permutation,
and nonparametric techniques.

30