|
<br />88
<br />
<br />..
<br />
<br />tistical error associated with natural differences between
<br />the seeded and non-seeded eJ(perimental units. Type I
<br />statistical errors in weather modification experiments
<br />have been suggested by others, but usually for the reverse
<br />reason (i.e., a "bad" draw instead of a "good" draw as in
<br />. the present discussion; cf. Gelhaus, Dennis, and Schock
<br />1974).. '
<br />Objective partitioning criteria of various types have
<br />been devised by Colorado State University meteorologists
<br />to index large-scale influences on precipitation (based on
<br />available upper air information) in an effort to weed out
<br />the events responsible for these region-wide natural differ-
<br />enceS in precipitation between the seeded and non-seeded
<br />experimental units. Attempts to use existing control
<br />stations (i.e., stations having elevations similar to gage
<br />lQcations of the Climax experiments) have been hampered
<br />by the small number of such stations, and by the fact
<br />that even these could have been contaminated by seeding
<br />in the Climax experiments or by other seeding operations
<br />for certain wind directions. The reanalysis is continuing.
<br />However, with additional partitioning based on more
<br />complicated meteorological criteria, the sample sizes are
<br />so small that the Climax I and II experimental data.
<br />. must be pooled. This (i) prevents comparisons between
<br />the replicated Climax I and II experimental data; and
<br />(ii) is likely to prohibit any strong conclusions concerning
<br />seeding effects, particularly since the degree of upper air
<br />detail required to adequately qUantify natural precipita-
<br />tion pattern variations between control and target loca-
<br />tions does not exist. Additional replications of the Climax
<br />experiments. with adequate instrumentation and the
<br />(' necessary models to quantify natural processes are needed
<br />I to help clarify many questions. (A comprehensive <lis-
<br />, cussion of the problems associated with the Climax I and
<br />f II experiments (jointly authored by G.W. Brier; L.O.
<br />Grant, G.T. Meltesen, G.J. Mulvey, J.O. Rhea, and
<br />. myself) is forthcoming.)
<br />'--
<br />
<br />FREDERICK MOSTELLER*
<br />
<br />Journal of: the American Statistical As~ociation, Mal'th1979
<br />
<br />The above discussion shows that even though the
<br />wintertime mountain precipitation process may be simple
<br />comps.red to summertime precipitation such as rainfall
<br />from thunderstorms, it is still extremely. eomplex.
<br />Furthl~r, it is controlled by so many scales of influences
<br />that highly heterogeneous samples may be difficult to
<br />. avoid, even in careflilly planned randomized experiments
<br />which use partitioning on many but not all of the in-
<br />fluences. It thus b~hooves both meteorologists and stat-
<br />isticians to design projects which consider all of the
<br />knowIll influence factors, and to make every effort to
<br />quanti.fy their degree of control over the precipitation
<br />process in order to diminish the uncertainties due to
<br />heterogeneous samples. (There is no inexpensive way to
<br />do this.) This complexity also emphasizes the need for
<br />statisticians and physical scientists to be involved at all
<br />stages of the project, including the operations (as pointed
<br />out by Professor Braham), to gain a better understanding
<br />of the natural variability of these many influences.
<br />
<br />REFERENCES
<br />
<br />Elliott, Robert D., Shaffer, Russell W., Court, Arnold, and Hanna-
<br />ford, .Jack F. (1978), "Randomized Cloud Seeding in the SanJuan
<br />Mountains, Colorado," Jcyuf'fWol. of Applied Meteorology, 17,
<br />1298-1318. . .
<br />Gelhaus, J.W., Dennis, A.S., and Schock, M.R. (1974), "Possibility
<br />of a 1l'ype I Statistical Error in Analysis of a Randomized Cloud
<br />Seeding Project in South Dakota," JO'U1'1I41. oj Applied Meteorology,
<br />13,383-386. .
<br />Grant, ll.ewis 0., and Mielke, Paul W., Jr. (1967), "A Randomized
<br />Cloud! Seeding Experiment at Climax, Colorado, 1960-1965,"
<br />Proce.1dingB of the Fifth Berkeley Sympolium on Mathematical Sta.-
<br />tistics and Probability, V, 115-131.
<br />Mielke, Paul W., 'Jr. (1978), "On Criticisms Concerning the Israeli
<br />ExpeIiment," Jcyuf'fWol. of Applied Meteorology, 17,555-556.
<br />-, Grant, Lewis 0., and Chappell, Charles F. (1971), "An In-
<br />dependent Replication of the Climax Wintertime Orographic
<br />Cloud Seeding Experiment," JoUf'fWol. of Applied Meteorology, 10,
<br />1198-1212; Corrigendum, 15, SOL
<br />Neymalll, J. (1977), "Experimentation with Weather Control and
<br />Statistical Problems Generated by It," in Applications of Statistics,
<br />ed. Ps,ruchuri R. Krisbnaiah, Amsterdam: North-Holland Publish-
<br />ing Co., 1-25.
<br />
<br />Comment
<br />
<br />. I discuss Dr. Braham's instructive article by describing
<br />some additional material and raising some questions.
<br />Specifically, I discuss collaboration between atmospheric
<br />scientists and statisticians, the need for cost-benefit
<br />
<br />· Frederick Mosteller is Roger I. Lee Professor of Biostatistics,
<br />Department of Biostatistics, School of Public Health, Harvard Uni-
<br />versity, Boston, MA 02115. The author wishes to thank Miriam
<br />Gasko-Green, David Hoaglin, Nan Laird, and Michael Stoto for
<br />clarifying suggestions.
<br />
<br />l
<br />'>"~c',,,,,-,,"J~A,j,,,,-,: ,.j ,
<br />
<br />.' ...... --' ," .. .
<br />~. ii:;:; ,. :'_~'::~~j,~ :;,,~:-~~i:,-ci_:,::,:,_ >'. ,-'.<i,-~,-~:"';-~~. ~~~~~~..__
<br />
<br />analysE~, and the problems of multiple analyses and of
<br />reanalyses and their relation to the interpretation of non-
<br />randomized studies.
<br />Dr. lBraham encourages statisticians to be involved in
<br />the tot:!).l eff<?rt of investigations of weather modification,
<br />not just the design and analysis portions. Let me under-
<br />line thiis. In 1971,' the Panel on Weather and Climate
<br />Modifi(lation of the National Academy of Sciences' Com-
<br />
|