WMOD00478

• Title: Field Experimentation in Weather Modification
• Date: 3/1/1979
• Weather Modification - Doc Type: Report

......,. .~ 'H~ "'-.-.""'-~"""'" .,. '."

. Gabriel: Field Experimentation in Weather Modification

out. (Braham's reference to Kruskal and Brownlee's in-
volvement is a case in point.)
This discussion of methodology and the role of the
statistician suggests that in both stage (2) of supervising
the performance of an experiment and in stage (3) of
doing the principal confirmatory analysis, it is desirable
that the.statistician not be too closely involved with the
experiment. This is not so in the stages of (1) design and
(4) exploration, where close association of statistician
and meteorologist is essential to successful planning and
exploration.
How, then, can one reconcile the requirements for close
involvement of a statistician in design and exploration
with the necessary distance a statistician must maintain
to supervise the performance of the experiment and its
confirmatory analysis? Clearly this can be done best by
involving different statisticians at the different stages:
the experimental team may hire or co-opt one statistician
to work in close association on all stages from design to
exploration. At the same time, a different statistician can
be retained as an independent consultant whose expertise
and authority provide the statistical quality control for
the performance and confirmatory analysis of the experi-
ment. It is difficult for one person to function effectively
in both roles.
In practice there may be slight variations on this
duality of statistical roles. Of course there is no objection
to the consultant statistician's learning much about cloud
physics and seeding, but this is not as essential as it is for
the statistician who works' on the experimental team.
What does not fit into this scheme of statistical-
meteorological association is exploratory activity on the
part of statisticians, acting independently of meteoro-
. logists. Such activity can be justified only if it is presumed
that the meteorological component of cloud seeding re-
search is negligible compared to the mathematical-
statistical. There do exist fields in which research work is
more statistical than substantive-demography is a
prime case in point, and some other social sciences seem
to be following suit. I believe this is a very unhappy and
unproductive state of affairs, and I can hardly wish this
on meteorology. A cautionary example of how far such
things might go was given by the fruitless analysis of
rainfall data of the Swiss hail prevention experiment in
terms of the personal characteristics of the weather fore-
casters on duty.
Most statisticians will be quite content with their role
in assisting meteorologists and participating with them
in what is, after all, meteorologicat and not statistical
research. One would not generally expect statisticians to
run their own analyses of meteorological data and pro-
pound physical theories on the effects of seeding. A sta-
tistician's eminence in his own field should not give par-
ticular weight to his excursions into meteorology, any
more than a meteorologist's innovations in statistics
would be appreciated merely because he has done good
work in meteorology. It would be unfortunate if meteo-
rologists' respect for statistics led them to lend undue

'-~

h',.'l",".At.ioi.>,,,."

~.. :';'~~_"::':~"~~':""""__:-~_.~~A....-..:::.~_.:...:.>':..::.~

83

authority to the ideas on cloud physics propounded by
st:ll.tisticians, however eminent, just as it would have been
unfortunate if R.A. Fisher's doubts about the connection
between smoking and cancer had been taken too seriously.
Weather modification leans heavily on statistics be-
es,use of the enormous variability of the weather. At
times, it seems almost as though weather modification has
become a statistical, rather than a meteorological, subject
of study. That is unfortunate because ultimately what is
at stake are not the methodological innovations inspired
by rainmaking experimentation, but the substantive
meteorological findings. The statistical contribution
should be to provide good designs and methods of
analysis, rather than to make independent interpretations
of the findings.
Braham mentions some matters on which he feels that
meteorologists have received conflicting and unclear
Sil~ls from statisticians. These issues mostly lie on the
borderline of statisti"cs and scientific method in general,
a region in which statisticians are no more at ease than
most other scientists. Statisticians are comfortable with
analyzing data and deriving methods and designs under
stated conditions. We have good mathematical statistics
tools that allow us to be clear and unequivocal in stating',
for example, that an experiment must run for at least 4-5
YElars to have a chance of showing significant seeding
effects. But we equivocate, disagree, and perhaps
obfuscate when it comes to more fundamental questions
of the role of statistics in the scientific process. There is
little agreement or clarity on these issues within the
profession-the Bayesians' attempts to formalize these
matters are considered by most statisticians to be too
restrictive for most real applications-a.nd so every sta-
tistician speaks very much from personal experience and
gut feelings. That may be valuable, but it cannot carry
the authority of a well-established scientific discipline,
and it certainly should not be treated with undue respect
by meteorologists.
Meteorologists can ask statisticians for evaluations of
designs, estimates, tests, and probabilities. They can,
and should, insist on being told what assumptions have
been made for these ev~luations, what evidence there is
tOi justify these assumptions, and how crucial they are;
i.e., how nonrobust the calculations. On this, they are
entitled to authoritative, clear, and reasonably unequiv-
ocal answers. To illustrate, when some statisticians
oppose the use of cross-over designs, they are usually
explicit about fearing the effects of contamination of one
area by the other area's seeding, and they can be asked
1,01 cite the evidence on which they base these fears. The
cloud seeder can then make a meteorological judgment of
whether the evidence for contamination is indeed con-
vincing. If so, the statistician's advice should be taken.
If not, it is irrelevant. What this illustrates is that the
meteorologist should never put the statistician in the
position of having to evaluate the physical evidence. The
statistician can no doubt help a great deal, but it is the
meteorologist who should have the more profound under-