Thurstone,Measurement For a New Science

SPRING 2000

Nikolaus Bezruczko, Ph.D.

POPULARMEASUREMENT7

P

R

O

FZLES

ZN

AILEA

UR

EM

99

EN

"He stole fire from the gods, then paid withfactor analysis."

In ten short years, early in his career, Louis L. Thurstone revolution-ized nonphysical scaling by single-handedly adapting the psychophysics de-veloped by Fechner, Wundt, and Miiller to measure mental forces in 20thcentury psychology. In contrast, the long, slow labor offactor analysis over-whelmed himfor more than twenty years, as he tried to develop and defendit. His measurement advances were spectacularly laying the foundations formodem psychometrics, while factor analysis was a dismal burden, consuminghis energy and distracting his attention . Scholars may argue whether factoranalysis wasted his time, but all agree he never returned to absolute scaling .

Thurstone's contributions to social science, however, go deeper thaninventing modern psychometrics . His goal was an entirely new theoreticalpsychology based on instincts, needs, and aspirations "where the dynamic selffinds overt expression" (1923, 356), "We should analyze . . . [human actions]. . . as the expression of cravings that originate in the organism and findparticular modes ofsatisfaction in the stimuli that happen to be available"(1923, 368) . In Thurstone's brave new cosmology, psychology studies theobjective representation ofthese mental forces, his alternative to stimulus-response behaviorism and subconscious psychoanalysis .

His scaling methods conceptualized these mental forces as abstractlinear continua, objectively measured on numerical scales, and their interre-lations expressed as mathematical formulations . Thurstone's sweeping advance, the greatest single achievement rationalizing social experience sincethe Enlightenment, opened the door to a new science ofmind, then stalledwhen he inexplicably succumbed to factor analysis. The ensuing dark cloudobscuredboth his measurement and psychology, costing him the momentumto advance psychology to an objective science . In 1954, at the end ofhiscareer, he expressed surprise over all the attention received by his difficultfactor analytic techniques, while his simple measurement methods neverbecame widely popular (1959, 15) .

His most important works, those which promised a sound, objectivebasis for social research occurred within a short time . By the early 1920s, hehad thought out the important conceptual issues for a new science which hediscussed philosophically in three articles (1919a, 1923, & 1924) . Then in1925 he explained his new scaling method, quickly followed in 1927, 1928,and 1929 with clarification and elaboration . By 1930, it was over. His shift toprimary mental abilities entangled him for years in methods incompatiblewith absolute scaling . The dark cloud drifted over 20th century social sci-ence as factor analysis fatally aroused the naive enthusiasm ofsocial research-ers everywhere .

Volumes could be written about Thurstonian psychometrics : itscentral features, empirical benefits, and implications for advancing socialresearch. Ofcourse this story would start with his decisive rejection ofclassi-

8 POPULAR MEASUREMENT

Nikolaus Bezruczko

Nikolaus Bezruczko is originally from Linz,Austria and immigrated to the U.S . as an infant .He graduated from MESA, University of Chi-cago in 1990 and has published widely in childdevelopment, aesthetics, psychometrics, and edu-cation . He currently lives with wife Ambra'Borgognoni Vimercati and stepdaughter Alice inChicago and Rome, Italy. Favorite recreation issummer hiking in South Tyrolean Alps .

cal psychophysics, as well as raw scores and mental ages . In-consistencies between Weberian and Fechnerian methods,limen determinations, andJND estimation instability are ex-amples of psychophysical concepts Thurstone consideredworthless to social research . To make this methodology mean-ingful, he needed to reconceptualize psychophysics . Insteadofcollecting perceptions of lifted weights and constructing ascale with physical units, he would identify distances betweenmental stimulibased on observer agreement withopinion state-ments using Fechnerian magnitude estimation methods . Thenall he needed was a procedure for transforming ordered pro-portions into scale values and computing their error distribu-tions . He would project mental structures on linear continuaand modeltheir quantitative properties with normal probabil-ity functions . Other improvements were also necessary, suchas shifting from the method of equal appearing intervals topaired comparison, but the decisive step was to conceptualizea response continuum in terms of social objects such as atti-tude, opinion, or preference judgments . His ideas, however,were strange to psychologists and social researchers, andThurstone faced enormous resistance and hostility. He triedto convince skeptics that subjective units were not only sen-sible and necessary, but easily estimated by selecting an arbi-trary item on a continuum and using its error distribution asthe scale unit . "The standard deviation of this dispersion for astandard stimulus could be chosen as a subjective unitofmea-surement." (1952, 307) His responses to objections includedelaborate descriptions ofhis measurement philosophy in pub-lications which fortunately now provide a detailed record ofThurstone's rationale for psychological measurement. Somemain ideas are :

* Mental integrity. Amental integrity independent ofovert behavior underlies the human tendency to engage inparticular actions. Thurstone's defiant reaction to emptyheaded Stimulus-Response psychology, this concept ration-alizes an inferential approach to mental functioning .

* Discriminal process. An automatic perceptualprocess sorts the ambient flow ofexternal stimuli to iden-tify those that may be useful to the organism . Thurstoneasserted they would show an error distribution on the stimu-lus continuum reproducing the subjective qualitative ex-perience .

* Motive forces . A structure ofmotive forces liesdormant in the mental system . Its provocation by itemsreveals mental affinity toward particular stimuli and defines a psychological continuum . "They acquire concep-tual linearity and measurability in the probability with whicheach of them may be expected to associate with any pre-scribed stimulus" (1927b, 51) . "To the extent their prob-abilities ofassociation with stimuli are nearly the same, tothat extent will they tend to be adjacently spaced on theimaginary psychological continuum ." (1927c, 419)

* Arbitrary units . Measuring in general is based onan arbitrary unit ofmeasure whose practical usefulness isits linearity. Thurstone applied Fechner's JND techniqueto estimate unit measure on the subjective continuum.

* Absolute scaling . Social researchers grate atThurstone's insight that scaling must be independent ofthe sample measured and unit ofmeasure . (Many ofthemare still using raw scores/ratings, percentages, and gradeequivalents .) "We have called the method absolute, not inthe sense of measurement from an absolute origin but inthe sense that the scale is independent ofthe unit selectedfor the raw scores and ofthe shape ofthe distribution oftheraw scores" (1927c, 517) . If the associational likelihoodbetween any two points on the continuum "should be af-fected by the opinion of any individual person or group,then it would be impossible to compare the opinion distri-butions oftwogroups on the same base" (1928a, 417) .

* Parameter linearity. "The sum of the subjectiveseparations between the stimulus pairs AB/BC must beequal to the experimentally independent determinationofthe separation AC. Ifthe continuum is unidimensional,then this simple type of check would establish the fact"(1952, 308) . Referring to the additivity axiom in physicalmeasurement, Thurstone presages probabilistic conjointmeasurement for nonphysical observations .

* Item fit. Thurstone was explicit, scale items needboth rational and empirical support . "The scaling methodshould be so designed that it will automatically throw outofthe scale any opinion statements which do not belong inits natural sequence" (1928a, 417) . Thurstone, however,did not support attempts to establish internal consistencycoefficients for this purpose . In general, "correlation pro-cedures constitute an acknowledgment offailure to ratio-nalize the problem and to establish the functions that un-derlie the data" (1929a, 224) . Thurstone was adamant,"correlation coefficients are symbols ofdefeat" (1929a, 240) .

He developed a detailed methodology to apply theseideas . For example, Figure 1 taken from a 1926 article presentspossibly the first cumulative item response curve ever published in a social research journal, now a standard presenta-tion method. Figure 2 shows parallel item trace lines defininglinear structure, the essential empirical evidence for a nu-merical variable . Another Thurstone contribution to socialtheory building is the variable map which positions item byperson dynamics in a quantitative graphic structure. He con-sidered the map an essential foundation for psychologicaltheory and provided many examples . Figure 3 is his S and Rcontinua, Thurstone's theoretical justification for generaliz-ingpsychophysics to nonphysical stimuli (I927a) . In the 1920s,any objective, quantitative representation ofsocial phenom-enawas an extraordinary achievement. Contemporaries suchas Binet, Burt, and Thorndike were pioneering ability and

SPRING 2000 POPULAR MEASUREMENT 9

0d

Scok

Figure 1

Figure 2

1 0 POPULAR MEASUREMENT

achievement tests ; but no one commanded Thurstone'sbreathtaking view on a new science . Over the next 70years,his ideas and methods would take on a life oftheir own ulti-mately to verify Thurstone's heretical assertion, "AttitudesCan Be Measured" (1928b) .

In contemporary social research where hyper-quan-tification and over-parameterization are endemic, Thurstoneis easily dismissed as a historical relic . After all, his wholescaling methodology is based on only two parameters, meanand standard deviation, the scale value and its error distribu-tion. As we all know, the mathematical complications ofcontemporary social research far surpass Thurstonian meth-ods . The surprise, however, is none ofthese complicated meth-ods meetscientific rigor. Each ofthese highly touted methods(multidimensional scaling, cluster analysis, and so on), on closeexamination, suffers from critical defects that destroy its ob-jectivity, generality, and simplicity. All of them obscure theperson in data aggregation . While results are sometimes inter-esting, they are essentially descriptive techniques about spe-cific samples . None offer any scientific advantages overThurstonian measurement .

Newton's expression, "If I have seen farther thanothers, it has been by standing on the shoulders ofgiants" isappropriate here . The evolution of scientific methodologythrough Fechnerto Thurstone and their successors, carriesonan intellectual tradition over 4,500 years old as seen by thebalance scales in Egyptian paintings during the Old Kingdom(Rice, 1990) . We can only speculate how earlier cultureshandled measuring issues . We know humans have an innatetendency to compare objects and abstract their differences .When commensurable with numbers and implemented todescribe patterns ofuniformity in nature, these units enablethe scientific thinking responsible for Western civilization.Separatingperceptual units from the observer and re-express-ing their quantitative properties numerically is the milestonein human history underlying all abstract sciences . Commerceand its evolution into economics, for example, established so-cial science . The failure ofcontemporary social research tocontinue this scientific methodology is responsible for its dis-mal record in the 20th century. Instead ofmodeling universalpatterns, social research remains limited to fragmented, andinconsistent patterns oftestimony, hardly scientific, generallyfailing to meet even minimal standards ofreplication or gen-erality. (Some evidence suggests social research has degener-ated to cult status, that is, dominatedby obtuse methods whichare only accessible to high priests yet without any clear rela-tion to constructing scientific knowledge about human be-havior.) The current absorption ofsocial researchby the physi-cal and biological sciences is a commentary on this failure .

Thurstone provided the architecture for a new sci-ence of mind, as well as the foundations for a nonphysicalmeasuring system : an objective framework in which to conduct scientific thinking. His key ideas are continuity, order,

SPRING 2000

and variability. Continuity is the continuum underlying ob-servations, order is the comparisons among items, and vari-ability is the metric of precision . Georg Rasch, in turn, ad-vanced objectivity by separating the ability and difficulty pa-rameters. This achievement liberates social units from theconfinement to standard deviates of arbitrary populationmeans, and constructs a pure mathematical abstraction, ameasured difference between ability and difficulty on an in-finite continuum. BenWright advanced the framework evenfurther by developing tests of statistical fit to detect depar-tures ofexperience from the abstraction and improve preci-sion andvalidity. Because this information about persons anditems clarifies the dimensionality underlying a scale, it suc-ceeds in eliminating the original motivation to develop factoranalysis . Together they establish a measurement trilogy forthe 20th century.

Biographical information concerning LouisThurstone is documented in several sources (Guilford, 1957 ;Wood, 1962 ; Thurstone, 1952 ; see also Gulliksen, 1968) .Thurstone was born inChicago in 1887 to native Swedes, theThunstr6m family, who changed their name to Thurstone toaccommodate American prejudice against foreigners . As achild, he was interested in music reinforced by his musicianmother. As a teenager, he became interested in trigonometryand in college published an equation for trisecting any angle(1912) . In 1912, he graduated from Cornell University with amechanical engineering degree and immediately went to workfor Thomas Edison in Orange, New Jersey (recruited afterdemonstrating his model of a nonflickering movie projector) .In 1914, he started graduate school inpsychology at the Uni-versity of Chicago . While completing a learning functionthesis, he went to Carnegie Institute of Technology in theDepartment of Applied Psychology. Thurstone returned tothe Chicago Department of Psychology in 1924 where hefounded the Psychometric Society and the journalPsychometrika . (Thurstone spoke on factor analysis to theSigma Xi Society in spring 1948 . After his talk, Ben Wright,then studying physics went to see Thurstone and learnedfrom him his shortcut method for doing factor analysis byhand.) In 1952, Thurstone retired from the University ofChicago and moved his psychometric laboratory to the Uni-versity ofNorth Carolina . (References available on request .)

SPRING 2000

REFERENCES

Guilford, J. J . (1957) Louis LeonThurstone: 1887- 1955 . BiographicalMemoirs, Volume30 . NewYork : Columbia University PressfortheNational Academyof sciences.

Gulliksen,H. (1968) . Louis Leon Thurstone: Experimental andmath-ematical psychologist . American Psychologist, 23, 716-80.

Rice, M. (1990) . Egypt's Making: The OriginsofAncient Egypt 5000,2000BC . London : Routledge .

Wood, D. A. (1962) . Creative Thinker, DedicatedTeacher, Eminent Psy.chologist . Princeton, NewJersey : Educational Testing Service.

Thurstone, L.L. (1912) . A curve whichtrisects anyangle . Scientific Amen,can,73,259-261 .

Thurstone, L. L. (1919a) . Theanticipatory aspect ofconsciousness. jour-nalof Philosophy. Psychology, andScientific Methods, 16,561-568.

Thurstone, L. L. (19196) . Ascoring method for mental tests. PsychologicalBulle 16, 235-240 .

Thurstone, L. L. (1923) . Thestimulus-response fallacy in psychology. P_sLchological Review, 30,354-369.

Thurstone, L. L. (1924) . Contributionsof Freudism to psychology: Influ-ence of Freudism on theoreticalpsychology. PsychologicalReview, 31,175-183 .

Thurstone, L. L. (1925) . Amethod of scalingpsychologicaland educa-tional tests. Journalof Educational Psychology, 16,433-451.

Thurstone, L. L. (1926) . Thescoring ofindividual performance . JournalEducational Psycholoey,17, 446-457 .

Thurstone, L. L. (1927a) . Psychophysical analysis . American JournalofPsychology, 38,368-389 .

Thurstone, L . L. (19276) . Amental unit of measurement . PsychologicalReview, 34,45-423 .

Thurstone, L. L. (1927c) . The unit of measurement in educational scales.ournal ofEducationalPsychology, 18,505-524.

Thurstone,L.L. (1928a) . Themeasurement ofopinion . JournalofAbnor-maland Social Psychology, 22,415-430 .

Thurstone, L.L. (19286) . Attitudes canbe measured. American Journalof

i to

, 33, 529-454Thurstone, L. L. (1929a) . Theory ofattitudemeasurement. Psychological

Revie , 36,222-241 .2=Thurstone, L. L. (19296) . Themental growth curve for the BinetTests.

JournalofEducationalPsychology, 20,569-583 .Thurstone, L. L. (1952) . (Boring, E. G., Langfeld, H.S ., Werner, H., &

Yerkes, R,M., Eds.) A History of Psychology in Autobiography. 295-321 ., Worcester,Massachusetts : Clark University Press .

Thurstone,L.L . (1954) . Themeasurementofvalues . PsvchologicalRe-view,61, 47-58.

Thurstone, L . L. (1959) . The Measurementof Values . Chicago : UniversityofChicagoPress .

Thurstone, L. L. &Chave, E. J . (1929) . Theory ofAttitude Measurement.Chicago : University of Chicago Press .

(See Mark Stone's article "Thurstone's Crime ScaleRe-Visited" onpage 53 . Editor)

Visit the Institute forObjective Measurement Website at,

www.rcrsch.org

POPULAR MEASUREMENT 1 1