WMOD00464

• Title: The Management of Weather Resources - Volume II
• Prepared For: The Weather Modification Advisory Board
• Date: 6/30/1978
• Weather Modification - Doc Type: Report

A-6

2) giving past data 1/2 weight (per experimental unit or per block of agreed-upon size and
structure) .

3) analyzing past and future segments of the data together, with equal weight for equal
number of (similar) experimental units (unit weight on past data).

If those making decisions among such alternatives were routinely and keenly aware of just
where the breakeven points between these specific alternatives lie, we would expect much
better decisions as to how data before and after a change are to be combined. (If there are two
changes, the earliest group of data will usually get the product of the weights selected separately
for each changeJ

In any event, the analysis itself must take into account exactly when any changes were
made (something that can be automatic if the experiment is blocked, and blocks are not
allowed to run across changes in operation).

Such modifications should almost never be considered in a confirmatory phase that has
been going on, unless the decision is to give up the phase and start anew.

* parallel exploration *

An important question is the degree to which it is economical to have "parallel explora-
tion". Should we be EXPLORING the usefulness of seeding on days which we believe are less
suitable for seeding in the same experiment where we are trying to CONFIRM the usefulness
of seeding on days which we believe to be the most suitable?

If we are able to define, satisfactorily and objectively, not only "suitable of class A", but
also "suitable of class B" and "suitable of class C", etc., where we seek confirmation for Class
A days alone, but exploration for the days in the other classes, we may learn something of
whether our judgment in choosing class A was excellent, good, poor, or even just wrong -- and
we may learn, possibly a lot, about the consequences of seeding under conditions of classes B,
C, etc. There can be real gains here, but unless we are wary, there also can be real losses, both
in money and in confusion.

The economics deserve attention in each case, but it seems likely to be generally true that
the incremental costs of operating on additional days will, often, be small compared to the fixed
costs of having equipment, personnel, and data handling systems available. (Adequately organ-
ized data collection, oversight, management and auditing may be important here in holding
down the costs of handling additional days of dataJ

The two questions that need to be faced before deciding pro or con about parallel explora-
tion are:

a) What will be the dollar costs?

b) Can we keep the main questions adequately clear of the subsidiary ones, both in the
minds of the experimenters, and in the minds of those who read about their results?

In such a situation, the results of the confirmatory part might well deserve separate publi-
cation. Indeed, it would not, for example, seem to us to be going too far to publish, in consid-
erable detail and before the combined experiment began, the confirmatory phase's sharply
focused main question(s), perhaps even in the Federal Register.

In an exploratory phase, somewhat similar patterns -- a single well-thought-of definition of
"suitability of class A" accompanied by plausible or marginal definitions of "suitability of Class
B, ...Class C, etc." -- may often have a real place. An adequate advance record of what was
really "class A" will still be very important.