|
<br />84
<br />
<br />standing of weather modification, and it is the meteorol-
<br />ogist who shoUld make the judgments and decisions. The
<br />statistician is an expert witness, but the meteorologist
<br />must be the judge-to the 'Elxtent that they can col-
<br />laborate, the judgment is more likely to be a wiser one.
<br />
<br />WilLIAM KRUSKAl *
<br />
<br />Journal o'f the American Statistical Association, March 1979
<br />
<br />REFERENCES
<br />
<br />Gabriell, K. Ruben, and Neumann, J. (1978), "A Note of Explanation
<br />on the 1961~7 Israeli Rainfall Stimulation Experiment" Journal
<br />01 AJrrplied Meteorology, 17, 552-554. '
<br />Tukey, John W. (1977), Ezploratory Data Analy8i8 Reading Mass.:
<br />Addison-Wesley Publishing Co. "
<br />
<br />Comment
<br />
<br />1. INTRODUCTION
<br />
<br />Professor Braham's article raises so many fascinating,
<br />important questions that it requires self-discipline to
<br />avoid spreading oneself over an unseemly number of pages
<br />in discussion. I restrict myself to two topics:
<br />
<br />Physical vs. statistical evidence,
<br />Multiplicity in analysis and criticism,
<br />
<br />and I omit such important other topics as
<br />
<br />Disagreement in' advice from different statisticians
<br />(and analogous disagreements among physicians,
<br />Supreme Court justices, or meteorologists),
<br />T~ what extent should the statistician participate in,
<br />or be a close observer of, an actual experiment?
<br />
<br />2. PHYSICAL V5. STATISTICAL EVIDENCE
<br />
<br />.
<br />
<br />At two or three points in Braham's article, and at a
<br />much sharper level in writings of other meteorologists,
<br />distinctions are made between physical and statistical
<br />experiments (and observations I programs), or between
<br />physical and statistical modes of thought. The comparison
<br />is sometimes to the discredit of statistics.
<br />In one broad sense of the word "statistics," such a
<br />distinction is otiose, for the meteorologist certainly makes
<br />inferences from quantitative da.ta and thus does statistics
<br />,
<br />whether or not the word is used. So presumably some nar-
<br />rower sense of "statistics" is intended. One such (un-
<br />fortunately) limited sense may be that in which "statis-
<br />tics" refers only to randomized experiments designed to
<br />test null hypotheses of no effect. Although such experi-
<br />ments and such hypothesis testing .are of importance,
<br />they are far from the totality of statistics as most stat-
<br />isticians use the term. First, not all analysis-worthy pro-
<br />
<br />· William Kruskal is Ernest DeWitt Burton Distinguished Service
<br />Professor, Dcpartm~nt of Statistics, Uni.v~~ity of Chicago, Chicago;
<br />IL 60637. H~ IS servmg as Dean of the DIVISIOn of the Social Sciences.
<br />Preparation of this comment was aided by National Science Founda-
<br />tion Grant. SOC76-80389. Mr. Kruskal wishes to thank the following
<br />colleagues for helpful comments: Louis J. Battan, Robert F. Boruch,
<br />Robert Kasa, Robert K. Merton, and Sandy Zabell.
<br />
<br />
<br />cedures are randomized experiments j and second, hy-
<br />pothellis testing is by no means the only, or even the
<br />major, mode of statisti(',S,1 analysis.
<br />AllOither sense in which the distinction might be in-
<br />tended is that of divergence between the result of a con-
<br />ventional statistical analysis, perhaps a significance test,
<br />on the one hand, and standard physical knowledge or
<br />possibly the intuition of one or more meteorologists on
<br />the other hand. This distinction has many variations, and
<br />probably most of us have our own stock of relevant
<br />anecdotes. Certainly cumulative scientific theory and the
<br />intuitions of scientists should be given great weight j yet
<br />scientific intuitions often vary widely, and there are
<br />many cases in which accepted doctrine has turned out to
<br />be wrong, 8Ometimp.8 after carrying out controlled ran-
<br />domizled trials. For one discussion in a medical setting,
<br />see Bunker et 0.1. (1978). There are, of course, many
<br />examples in many fields of science of the overthrow of
<br />, scientific consensus j one recent example has been the
<br />amazingly swift change in skepticism toward continental
<br />drift by geologists. Nearly worldwide scoffing has been
<br />replac1ed by general acceptance under the splendid name
<br />of "pll~te tectonics."
<br />We should keep in mind that a major motivation W
<br />significance testin~ in its basic sense is to provide self-
<br />disciplilw. for t~~ enthusiastic investie:a!:pr '!Vho"~y,,be
<br />carried aw~~_ with.!!~!larranted excitement over data..
<br />~t~ns tha.t..!-re.!~~d~;y ~~l~~~~~El tciiQ.n.
<br />~ th,e other hand,. of course, a policy of intellectually
<br />puritanical stri~~~ would,':l?l2B!xJ:~:ii:iiiU:Cli:Iiii'~
<br />from s~, but wou=~ ~ 9f new dis-
<br />~~alO~w.i~L~~t~~_. ;._.1~t ~r~~~~,
<br />statlStlcally Sl~~I;!.t~ This 18 Jle familiar tenslOn be-
<br />-.--"'...,~~~
<br />tween._~~call.c_e1ev.eL.and.poweri- a--tension..that.arises
<br />in one fbrm or another in almost all modes of inferential
<br />statlStics:..-'-~""-' ._--".-' - --,.-----..,,---. '-.^< ,-
<br />
<br />Physical, more generally substantive, knowledge per-
<br />force E!nters any investigation at the most fundamental
<br />level of decisions about variables and measurements. It
<br />enters in decisions about analytic strata and about eligi-
<br />bility of units (as in the case of eligible days for Whitetop).
<br />
|