20ENERO14 S. S. Stevens - Sobre Teoría de Escalas de Medición

8/13/2019 20ENERO14 S. S. Stevens - Sobre Teora de Escalas de Medicin

1/5

http://www jstor org

On the Theory of Scales of MeasurementAuthor(s): S. S. StevensSource: Science, New Series, Vol. 103, No. 2684, (Jun. 7, 1946), pp. 677-680Published by: American Association for the Advancement of ScienceStable URL: http://www.jstor.org/stable/1671815Accessed: 16/06/2008 17:33

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at

http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR's Terms and Conditions of Use provides, in part, that unless

you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you

may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at

http://www.jstor.org/action/showPublisher?publisherCode=aaas .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed

page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We enable the

scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that

promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].
http://www.jstor.org/stable/1671815?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/action/showPublisher?publisherCode=aaashttp://www.jstor.org/action/showPublisher?publisherCode=aaashttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/1671815?origin=JSTOR-pdf


2/5

S IEN EVol. 103, No. 2684 Friday, June 7, 1946

On the Theory of Scales of MeasurementS. S. Stevens

Director, Psycho-AcousticLaboratory, Harvard University

FOR SEVEN YEARS A COMMITTEE of theBritish Association for the Advancement ofScience debated the problem of measurement.

Appointed in 1932 to represent Section A (Mathe-matical and Physical Sciences) and Section J (Psy-chology), the committee was instructed to considerand report upon the possibility of quantitative esti-mates of sensory events -meaning simply: Is it pos-sible to measure human sensation? Deliberation ledonly to disagreement, mainly about what is meant bythe term measurement. An interim report in 1938found one member complaining that his colleagues

came out by that same door as they went in, and inorder to have another try at agreement, the committeebegged to. be continued for another year.

For its final report (1940) the committee chose acommon bone for its contentions, directing its argu-

ments at a concrete example of a sensory scale. Thiswas the Sone scale of loudness (S. S. Stevens andH. Davis. Hearing. New York: Wiley, 1938), whichpurports to measure the subjective magnitude of anauditory sensation against a scale having the formalproperties of other basic scales, such as those used tomeasure length and weight. Again the 19 members ofthe committee came out by the routes they entered,and their views ranged widely between two extremes.One member submitted that any law purporting toexpress a quantitative relation between sensation in-tensity and stimulus intensity is not merely false butis in fact meaningless unless and until a meaning canbe given to the concept of addition as applied to sen-sation (Final Report, p. 245).

It is plain from this and from other statements bythe committee that the real issue is the meaning ofmeasurement. This, to be sure, is a semantic issue,but one susceptible of orderly discussion. Perhapsagreement can better be achieved if we recognize thatmeasurement exists in a variety of forms and thatscales of measurement fall into certain definite classes.These classes are determined both by the empiricaloperations invoked in the process of measuring and

'

by the formal (mathematical) properties of the scales.Furthermore-and this is of great concern to severalof the sciences-the statistical manipulations that canlegitimately be applied to empirical data depend uponthe type of scale against which the data are ordered.

A CLASSIFICATION FP CALES OF MEASUREMENT

Paraphrasing N. R. Campbell (Final Report, p.340),. we may say that measurement, in the broadestsense, is defined as the assignment of numerals to ob-jects or events according to rules. The fact thatnumerals can be assigned under different rules leadsto different kinds of scales and different kinds ofmeasurement. The problem then becomes that ofmaking explicit (a) the various rules for the assign-ment of numerals, (b) the mathematical properties(or group structure) of the resulting scales, and (c)

the statistical operations applicable to measurementsmade with each type of scale.

Scales are possible in the first place only becausethere is a certain isomorphism between what we cando with the aspects of objects and the properties ofthe numeral series. In dealing with the aspects ofobjects we invoke empirical operations for determin-ing equality (classifying), for rank-ordering, and fordetermining when differences and when ratios betweenthe aspects of objects are equal. The conventionalseries of numnerals yields to analogous operations:We can identify the members of a numeral seriesand

classifythem. We know their order as

givenby convention. We can determine equal differences,as 8-6=4-2, and equal ratios, as 8/4=6/3. Theisomorphism between these properties of the numeralseries and certain empirical operations which we per-form with objects permits the use of the series as amodel to represent aspects of the empirical world.

The type of scale achieved depends upon the char-acter of the basic empirical operations performed.These operations are limited ordinarily by the natureof the thing being scaled and by our choice of pro-cedures, but, once selected, the operations determine

677


3/5

Vol. 103, No. 2684

that there will eventuate one or another of the scaleslisted in Table 1.1

The decision to discard the scale names commonlyencountered in writings on measurement is based onthe ambiguity of such terms as intensive and ex-tensive. Both ordinal and interval scales have at

Thus, the case that stands at the median (mid-point)'of a distribution maintains its position under all trans-formations which preserve order (isotonic group), butan item located at the mean remains at the mean onlyunder transformations as restricted as those of thelinear group. The ratio expressed by the coefficient

TABLE 1

iScale Basic Empirical~ Mathematical Permissible StatisticsOperations Group Structure (invariantive)

NOMINAL Determination of Permutation group Number of casesequality a' =f(a') Mode

f(x) means any one-to-onesubstitution one-tO-e Contingency correlation

ORDINAL Determination of Isotonic group Mediangreater or less $= f($) Percentiles

f(Bx) means any monotonicincreasing function

INTERVAL Determination of General linear group Meanequality of intervals a = a' + b Standard deviationor differences Rank-order correlation

Product-moment correlation

RATIO Determination of - Similar.ity group Coefficient of variationequality of ratios a = aas

times been called intensive, and both interval andratio scales have sometimes been labeled extensive.

It will be noted that the column listing the basic

operationsneeded to create each

typeof scale is cumu-

lative: to an operation listed opposite a particular scalemust be added all those operations preceding it. Thus,an interval scale can be erected only provided we havean operation for determining equality of intervals, fordetermining greater or less, and for determining equal-ity (not greater and not less). To these operationsmust be added a method for ascertaining equality ofratios if a ratio scale is to be achieved.

In the column which records the group structure ofeach scale are listed the mathematical transformationswhich leave the scale-form invariant. Thus, any nu-meral, x, on a scale can be replaced by another numeral,x', where x' is the fanction of x listed in this column.Each mathematical group in the column is containedin the group immediately above it.

The last column presents examples of the type ofstatistical operations appropriate to each scale. Thiscolumn is cumulative in that all statistics listed areadmissible for data scaled against a ratio scale. Thecriterion for the appropriateness of a statistic is in-variance under the transformations in Column 3.

1A classification essentially equivalent to that containedin this table was 1rr.-. ut4.*dbl4efor0e te International Congressfor the Unity of Scitlncet .Sc',teiul.r l141. The writer isindebted to the la tr Pror'. (.;. D. Birkl;,iffl for a stimulatingdiscussion which led to the completion of the table in essen-tially its present form.

of variation remains invariant only under the simi-larity transformation (multiplication by a constant).(The rank-order correlation coefficient is usuallydeemed

appropriateto an ordinal

scale,but

actuallythis statistic assumes equal intervals between succes-sive ranks and therefore calls for an interval scale.)

Let us now consider each scale in turn.

NOMINAL SCALE

The nomrinal scale represents the most vnrestrictedassignment of numerals. The numerals are used only aslabels or type numbers, and words or letters would serveas well. Two types of nominal assignments are some-times distinguished, as illustrated (a) by the 'num-bering' of football players for the identification of theindividuals, and (b) by the 'numbering' of types orclasses, where each member of a class is assigned thesame numeral. Actually, the first is a special case ofthe second, for when we label our football players weare dealing with unit classes of one member each.Since the purpose is just as well served when any twodesignating numerals are interchanged, this scale formremains invariant under the general substitution orpermutation group (sometimes called the symmetricgroup of transformations). The only statistic rele-vant to nominal scales of Type A is the number ofcases, e.g. the number of players assigned numerals.But once classes containing several individuals have

678 SCIENCE


4/5

June 7, 1946

been formed (Type B), we can determine the mostnumerous class (the mode), and under certain con-ditions we can test, by the contingency methods,hypotheses regarding the distribution of cases amongthe classes.

The nominal scale is a primitive form, and quitenaturally there are many who will urge that it is ab-surd to attribute to this process of assigning numeralsthe dignity implied by the term measurement. Cer-tainly there can be no quarrel with this objection, forthe naming of things is an arbitrary business. How-ever we christen it, the use of numerals as names forclasses is an example of the assignment of numeralsaccording to rule. The rule is: Do not assign thesame numeral to different classes or different numeralsto the same class. Beyond that, anything goes withthe nominal scale.

ORDINAL SCALE

The ordinal scale arises from the operation of rank-ordering. Since any 'order-preserving' transformationwill leave the scale form invariant, this scale has thestructure of what may be called the isotonic or order-preserving group. A classic example of an ordinalscale is the scale of hardness of minerals. Other in-stances are found among scales of intelligence, per-sonality traits, grade or quality of leather, etc.

As a matter of fact, most of the scales used widelyand effectively by psychologists are ordinal scales. In

the strictest propriety the ordinary statistics involvingmeans and standard deviations ought not to be usedwith these scales, for these statistics imply a knowl-edge of something more than the .relative rank-orderof data. On the other hand, for this 'illegal' statisti-cizing there can be invoked a kind of pragmatic sanc-tion: In numerous instances it leads to fruitful results.While the outlawing of this procedure would probablyserve no good purpose, it is proper to point out thatmeans and standard deviations computed on an ordinalscale are in error to the extent that the successive in-tervals on the scale are unequal in size. When onlythe rank-order of data is known, we should proceedcautiously with our statistics, and especially with theconclusions we draw from them.

Even in applying those statistics that are normallyappropriate for ordinal scales, we sometimes findrigor compromised. Thus, although it is indicated inTable 1 that percentile measures may be applied torank-ordered data, it should be pointed out that thecustomary procedure of assigning a value to a per-centile by interpolating linearly within a class intervalis, in all strictness, wholly out of bounds. Likewise,it is not strictly proper to determine the mid-point ofa class interval by linear interpolation, because the

linearity of an ordinal scale is precisely the propertywhich is open to question.

INTERVAL CALE

With the interval scale we come to a form that isquantitative in the ordinary sense of the word. Al-

most all the usual statistical measures are applicablehere, unless they are the kinds that imply a knowledgeof a 'true' zero point. The zero point on an intervalscale is a matter of convention or convenience, as isshown by the fact that the scale form remains in-variant when a constant is added.

This point is illustrated by our two scales of tem-perature, Centigrade and Fahrenheit. Equal intervalsof temperature are scaled off by noting equal volumesof expansion; an arbitrary zero is agreed upon foreach scale; and a numerical value on one of the scales

is transformed into a value on the other by means ofan equation of the form x'= ax+ b. Our scales oftime offer a similar example. Dates on one calendarare transformed to those on another by way of thissame equation. On these scales, of course, it is mean-ingless to say that one value is twice or some otherproportion greater than another.

Periods of time, however, can be measured on ratioscales and one period may be correctly defined asdouble another. The same is probably true of tem-perature measured on the so-called Absolute Scale.

Most psychological measurement aspires to create

interval scales, and it sometimes succeeds. The prob-lem usually is to devise operations for equalizing theunits of the scales-a problem not always easy ofsolution but one for which there are several possiblemodes of attack. Only occasionally is there concernfor the location of a 'true' zero point, because thehuman attributes measured by psychologists usuallyexist in a positive degree that is large compared withthe range of its variation. In this respect these at-tributes are analogous to temperature as it is encoun-tered in everyday life. Intelligence, for example, isusefully assessed on ordinal scales which try to ap-proximate interval scales, and it is not necessary todefine what zero intelligence would mean.

RATIO SCALE

Ratio scales are those most commonly encounteredin physics and are possible only when there existoperations for determining all four relations: equal-ity, rank-order, equality of intervals, and equality ofratios. Once such a scale is erected, its numericalvalues can be transformed (as from inches to feet)only by multiplying each value by a constant. An ab-solute zero is always implied, even though the zerovalue on some scales (e.g. Absolute Temperature) may

SCIENCE 679


5/5

Vol. 103, No. 2684

never be produced. All types of statistical measuresare applicable to ratio scales, and only with these

scales may we properly indulge in logarithmic trans-formations such as are involved in the use of decibels.Foremost among the ratio scales is the scale of num-

ber itself-cardinal number-the scale we use whenwe count such things as eggs, pennies, and apples.This scale of the numerosity of aggregates is so basicand so common that it is ordinarily not even men-tioned in discussions of measurement.

It is conventional in physics to distinguish betweentwo types of ratio scales: fundamental and derived.Fundamental scales are represented by length, weight,and electrical resistance, whereas derived scales arerepresented by density, force, and elasticity.

These latter are derived magnitudes in the sensethat they are mathematical functions of certain fun-damental magnitudes. They are actually more numer-ous in physics than are the fundamental magnitudes,which are commonly held to be basic because theysatisfy the criterion of additivity. Weights, lengths,and resistances can be added in the physical sense,but this important empirical fact is generally accordedmore prominence in the theory of measurement than itdeserves. The so-called fundamental scales are. im-portant instances of ratio scales, but they are onlyinstances. As a matter of-fact, it can be demonstrated

that the fundamental scales could be set up even if thephysical operation of addition were ruled out as impos-sible of performance. Given three balances, for ex-ample, each having the proper construction, a set ofstandard weights could be manufactured without itever being necessary to place two weights in the samescale pan at the same time. The procedure is too longto describe in these pages, but its feasibility is men-tioned here simply to suggest that physical addition,even though it is sometimes possible, is not necessarilythe basis of all measurement. Too much measuringgoes on where resort can never be had to the processof laying things end-to-end or of piling them up ina heap.

Ratio scales of psychological magnitudes are rarebut not entirely unknown. The Sone scale discussedby the British committee is an example founded on adeliberate attempt to have human observers judge theloudness ratios of pairs of tones. The judgment ofequal intervals had long been established as a legiti-mate method, and with the work on sensory ratios,started independently in several laboratories, the final

step was taken to assign numerals to sensations ofloudness in such a way that relations among the sensa-

tions are reflected by the ordinary arithmetical rela-tions in the numeral series. As in all measurement,there are limits imposed by error and variability, butwithin these limits the Sone scale ought properly to beclassed as a ratio scale.

To the British committee, then, we may venture tosuggest by way of conclusion that the most liberal anduseful definition of measurement is, as one of its mem-bers advised, the assignment of numerals to things soas to represent facts and conventions about them.The problem as to what is and is not measurementthen reduces to the simple question: What are therules, if any, under which numerals are assigned? Ifwe can point to a consistent set of rules, we are ob-viously concerned with measurement of some sort, andwe can then proceed to the more interesting questionas to the kind of measurement it is. In most casesa formulation of the rules of assignment disclosesdirectly the kind of measurement and hence the kindof scale involved. If there remains any ambiguity,we may seek the final and definitive answer in themathematical group-structure of the scale form: Inwhat ways can we transform its values and still haveit serve all the functions previously fulfilled? We

know that the values of all scales can be multipliedby a constant, which changes the size of the unit. If,in addition, a constant can be added (or a new zeropoint chosen), it is proof positive that we are notconcerned with a ratio scale. Then, if the purposeof the scale is still served when its values are squaredor cubed, it is not even an interval scale. And finally,if any two values may be interchanged at will, theordinal scale is ruled out and the nominal scale is thesole remaining possibility.

This proposed solution to the semantic problem isnot meant to imply that all scales belonging to the

same mathematical groupare

equally preciseor ac-

curate or useful or fundamental. Measurement isnever better than the empirical operations by whichit is carried out, and operations range from bad togood. Any particular scale, sensory or physical, maybe objected to on' the grounds of bias, low precision,restricted generality, and other factors, but the ob-jector should remember that these are relative andpractical matters and that no scale used by mortalsis perfectly free of their taint.

680 SCIENCE

Documents

20ENERO14 S. S. Stevens - Sobre Teoría de Escalas de Medición