Editorial. Sanity in Statistics

---

4.3

1

6-

L_

a

A

‘

-.a”,

* “

.+ax

@

ANALYTICAL C H E M I S T R Y Walter J. Murphy, Editor - -_

Sanity in Statistics 1E

cxtreiiie results are tl i c oiie? n i o F t neai,ly correct, the siinpk c:tlcnlat~ionof t’he mean of several chemical determinations may prove grossly misleading. Often it is far better, statisticians say, to report, all the raTv data-and not just a single calculated result-principally as a means of exposing the caliber of the experjmental findings. I n addition, statisticians have a cautioning word about, the glib use of formulas for the rejection of results that fall out of line. Even when truly random phenoiiiena are concerned, observations must not be blindly rejected. The mere fact that one set of dat’a’maydiffer markedly from the rest is not, they eniphasize, sufficient evidence that such unexpected (lata may properly be ignored. William C. Schlecht of the U. S. Geological Survey’s Chemical Laboratory has said: “Statistical methods are no substil u t e for understanding. We should hy all means use them \\.hen we are ready for them. B u t before we can study a given :inalytical method statistically, we must have adequate experience on which to base our study. T h a t requires both a large number of experiments under carefully controlled contlitions and a sound theoretical untlerst’anding of the process. These must be sweated out for each kind of determination :ind material. When we do use statistical methods, we niust do so responsibly and with a reasonalde appreciation of their meaning, plus the extreme skepticisin traditionally expected of scientist,s.” Among chemists the1.e is developing these days a gro\ving :t\wreness that conipeteiit statistical analyses niay oftentimes i,equire the skill of a professional statistician. The run-ofthe-mill chemist who has dabbled briefly in statistics and perhaps is fortified with an elementary knowledge of its jargon, is apt’to be hopelessly beyond his depth in dealing with st,atiPt i c d problems. For thoroughgoing statistical work, the cheniist must turn to the specialist.. Increasingly, chemists :ire doing just that. Statist’icians. ill turn, are proving their worth in chemical research not only by their critical appraisal of data but by their conipet’ent formulation of experimental p1,ocedures. It is a sign of the times that, in molie and more hlioratories today, the chemically trained statistician 113s bei’oine an essential meqber of the research team. l h c k in 1949, Beverly L. Clarke of Xerck & Co., Inc., concvlritlecl an address before a n analytical symposium with the \\-oi,cts: ,‘Iurge you not, to delay beginning the process of iliaking your organization statistics-conscious. For the longer you delay, the further you will lag behind your bolder rivals who are making more and more extensive use of this youngest and fast-growing accessor!. technique to chemical analysis.” The enthusiastic i n h e s t in statistical methods expressed at last month’s Kashington symposium has made it C’1:ii.ke’sinjunction two years ago more than obvious that 111.. did not fall upon deaf ears.

LAST month‘s anal) tical s j iiiposium IS any indication, chemists these days are developing a lively intereqt in statistical methods. Although none of the speakers at the Waqhington meeting singled out statistics foi full-(11eis 1 eview, several gave the topic considerably more than a passing nod. The speakers were not the only ones. A great many of those attending the three-day symposium debated the merits and shortcomings of statistics-earnestly, critically, and a t length. Over the years, statistical methods have had both theii energetic partisans and their noisy detractors. T o some scientists, statistics is the sun, the moon, and the stars. T o others, equally as vocal yet not nearly so numerous, statistics is just so much academic hogwash. Needless to say, the great majority of chemists-those who u p until now have managed to remain on the sidelines-are beginning to wonder rather seriously what all the controveisy is about. Today, many of those most skeptical of statistical methods come from the ranks of the disillusioned. Their position i5 easily understood. J u d a few years ago, statistics \$as heialded as a panacea, a short cut to knowledge, a final answer t o the ills of the laboratory. The hope was all too prevalent that somehow-by some magical means-statistics would be able to breathe unheard-of accuracy into half-baked results. Obviously, no self-respecting statistician would have enciorseu the extravagant clainis that nere characteristic of the day. IThen, inevitably, chemists found that statistics could not live up to its advance billing, many turned their backs. X a n y still have their backs turned. Others, however, have come to realize that the sins of the past have been the sins of misuse. Statistics has a place in chemistry-a veiy impoitant placebut its limitations must be recognized and its methods must be intelligently applied. More and more, t h i ~is being understood. I n fact, one of the most notable features of last month’s symposium was its unmistakable evidence that a t long last there is evolving a balanced and enlightened app o a c h to statistics on the pait of analytical chemists and their laboratory brethren. I n the past, chemists have been all too prone to plug any and all analytical data into standard statistical forniulas in the expectation that, with a few turns of the crank. the correct result n ould necessarily emerge. Many standard teutbooks and handbooks, by reprinting simple statistical forniulas without even a superficial explanation of their undeilying :issumptions, have lent tacit support to just such a belief. Experience shows, however, that ready-made formulas applied without a n intelligent appreciation of their meaning are orse than useless. Basically, statistical formulas are designed for the analysis of random errors. In many cases, the significant errors in chemical analysis are not random. Because frequently the 937