Industrial Research 



287 



accidental or significant. An example is found in the 

 field of geochemical prospecting. The surface soil 

 overlying regions in which there is oil contains a higher 

 proportion of hydrocarbons and waxes than occur in 

 other locations. Chemical analj'sis of surface soil 

 therefore affords a means of prospecting for oil. Mr. 

 Eugene McDcrmott writes: 



In the geochemical method, it was found necessary to deter- 

 mine between samples showing significantly high analysis values, 

 and those w'hich were normal values. These normal sample 

 values, of course, had considerable variation between themselves, 

 due to analysis and in larger part sampling errors. After exam- 

 ining these data for a long period of time, it was decided to 

 approach the problem statistically. This disclosed at once that 

 areas surveyed could be divided into positive (having significant 

 values, and hence favorable from the standpoint of petroleum 

 possibilities), negative (no significant values and unfavorable for 

 petroleum) and marginal (indeterminate). The latter case is 

 always the most difficult one in surveying, and while we are now 

 able to recognize it, further work is needed to fully interpret it. 

 This kind of mathematics is being applied at the present moment, 

 and bids fair to solve the problem. 



(6) In planning the kind of experiments from which 

 such data arise. Wliether variations are or are not 

 significant depends in no small degree upon the fashion 

 in which the data were taken. Consideration of the 

 experiment in advance from a statistical point of view 

 often results in economy of procedure, or even points 

 the difTerence between a trustworthy and a meaningless 

 result. 



The following example is quoted from an address by 

 Dr. R. H. Pickard, Director of the British Cotton 

 Research Association: 



To illustrate the advantage of good experimental design I may 

 refer to some experiments carried out at the Shirley Institute 

 to find the effect of various treatments on a quality of cloth. 

 This quality varies considerably at different parts of the same 

 piece of cloth, and in order to measure the effect of the treatments 

 the tests are repeated systematically so that the variations are 

 "averaged out." Some of the natural variation, however, is 

 systematic, and by adopting a "Latin Square" arrangement of 

 treatments on the cloth (such as is much used in agricultural 

 yield trials), these systematic variations are eliminated from the 

 comparison, and in the instance quoted the result was to reduce 

 by one-half the number of tests necessary for a given significance 

 as compared with a random arrangement.' 



To the extent to which biology becomes an important 

 element in industrial research — and it would appear to 

 be on the point of doing so in such fields as food manu- 

 facturing — it can be expected that the type of statistical 

 work listed under (a) and (6) will rapidly increase. 



(c) In laying out an inspection routine. Manufac- 

 turing inspection frequently yields data which are best 

 interpreted statistically, either because only spot-checks 

 are taken, or because the method of inspection gives 

 measurements which are themselves subject to acci- 



dental fluctuation. In such cases statistical theory is 

 of great advantage in setting up an efl'ectivc and eco- 

 nomical inspection program. It is being so used in 

 certain industries, notably in electrical manufacturing 

 and textiles, but the potential field of usefulness is far 

 from covered. 



The following example is quoted from an address by 

 Mr. Warner Eustis, staff officer on research of the 

 Kendall Company: 



Surgical sutures are twisted strands of sheep intestine, which 

 has been slit lengthwise * * * After a stated number of days 

 a sewing with such material, implanted in the body during a 

 surgical operation, will be digested and disappear as the healing 

 processes progressively take up the load originally held by the 

 suture * * * Here is a product which it is impossible to 

 test in any way without destroying the product, especially as 

 each suture is sealed in an individual, sterilized tube. Our 

 final product tests must all be conducted by breaking open a 

 sterile tube and testing the product therein. The quality 

 appraisal of such a product naturally rests upon probability, 

 rather than upon an actual testing of each item. Due to the 

 nature of such a product, in which a single failure may destroy 

 human life, the need for accurate quality appraisal is super- 

 lative.' 



(d) In the control of manufacturing processes. In- 

 spection is not merely a means of discarding bad prod- 

 uct; it is also a means of detecting trouble in the factory. 

 This is obvious in the extreme cases when the product 

 is unusually bad. By the use of suitable routines set 

 up in accordance with statistical theory, the day-to-day 

 results of inspection can be used to detect incipient 

 degradation in the process of manufacture which might 

 otherwise escape notice. This procedure is used 

 extensively by the Western Electric Company in assur- 

 ing uniform quality in many items of manufacture, and 

 to a lesser extent in other industries. Of it, Mr. J. M. 

 Juran, manufacturing engineer of the Western Electric 

 Company, says: 



Too frequently we have seen an inspection group grow lax 

 in vigilance until a complaint from the customer wakes them up. 

 They promptly swing the pendulum a full stroke in the opposite 

 direction, and the factory groans in its effort to meet the now 

 unreasonable demands. A sound and steady control, like a 

 sound currency in commercial relations, gives factory foremen a 

 feeling of confidence and gives the consumer a feeling that 

 control is being exercised before the product reaches him.'" 



(e) In writing rational specifications. Obviously, if 

 such a procedure helps the manufacturer to assure uni- 

 form quality, it is also of value to the purchaser of his 

 products. Hence the subject of statistics enters into 

 the writing of the buyer's specifications. It has been 

 so used to a limited extent in the Bell System in con- 

 nection with telephone apparatus, and by the United 



' Pickard, R. H. The application of statistical methods to production and research 

 in industry. Journal of the Royal Statistical Society, Supplement, 1, No. 2, 9-10 (1934). 



• Eustis, Warner. Wliy the Kcnddll Company is interested In statisticul methods. 

 Industrial Stutistics C<mference, Proceedings, 143-144 (held at Ma.ssachusetts Institute 

 ot Technology, Cambridge, Mass., September 8-9, 1938). 



"> Juran, J. M. Inspectors' errors in quality control. Mechanical Engineering, 

 57, 643-644 (October 1935). 



