Professional Documents
Culture Documents
A historical analysis of the period from 1925 to 1950 was conducted to inves-
tigate the incorporation of analysis of variance (ANOVA) techniques in psycho-
logical research. In addition to attempts to identify the earliest uses of ANOVA
in psychology, the gradual incorporation of the technique was examined by
counting its appearance in 6,457 articles appearing in major American psycho-
logical journals from 1935 through 1952. Expository articles and texts directed
at psychologists were also identified, and a questionnaire survey of graduate
psychology programs was conducted to establish how psychologists were intro-
duced to ANOVA. Finally, the phylogeny of major contributors to American psy-
chological statistics was established for the period. The data suggest a segmen-
tation into three historical periods: (a) an initial, expository phase lasting until
the onset of World War II, (b) a wartime interregnum during which use of
ANOVA declined, and (c) a postwar resurgence, characterized by the institu-
tionalization of ANOVA training. Although ANOVA certainly had a major impact
on experimental psychology, the data do not permit the conclusion that the
incorporation of ANOVA in psychology in itself constituted a revolutionary
"paradigm shift."
In 1957 Cronbach, in a presidential ad- Cronbach asserted that "Fisher made the
dress to the American Psychological Associa- experimentalist an expert puppeteer, able to
tion, described two disciplines of scientific keep untangled the strands to half-a-dozen
psychology: the correlational and the experi- independent variables (p. 675)."
mental. He characterized the experimental Although it may be superfluous to reassert
school by its reliance on analysis of variance that the ANOVA has been the workhorse sta-
(ANOVA) techniques for research design, in tistic of experimental psychology, it is not
contrast to the correlational techniques fa- superfluous to consider how this technique
vored by the other school. Referring to the carne to dominate experimental design in psy-
English statistician who developed ANOVA, chology. Not until the late 1930s did ANOVA
begin to appear in psychological journals, and
yet by 1952 it was fully established as the
The authors would like to acknowledge contribu- most frequently used technique in experi-
tions in the form of personal communications from
Harold Bechtoldt, Isadore Blumen, Alphonse Cha-
mental research. It took less than 15 years
panis, C. J. Cranny, Lee Cronbach, Allen L. Ed- for psychology to incorporate ANOVA.
wards, Nathaniel Gage, Bert Green, J. P. Guilford, The purpose of the present study was to
Harold Gulliksen, Lloyd Humphreys, E. F. Lind- develop a historical account of psychology's
quist, Quinn McNemar, Don Ragusa, T. A. Ryan,
Patricia C. Smith, and Ray Tucker.
incorporation of ANOVA from Fisher's initial
A brief version of this article was presented at presentation of the technique (1925b)
the annual convention of the American Psychologi- through the early 1950s. It became clear in
cal Association in Toronto, Ontario, Canada, August the course of this research that experimental
1978. design in psychology changed radically in the
Requests for reprints should be sent to Anthony
J. Rucci, who is now at the American Hospital
period in question: Single-variable designs
Supply Corporation, 1 American Plaza, Evanston, were being used less frequently from 1940
Illinois 60201. on, whereas factorial designs were used more
166
ANALYSIS OF VARIANCE 167
divided the total variance of the results of an today, had its beginning with Sir Ronald A. Fisher
experiment to test whether different varieties and began in the second quarter of the twentieth
century. (Federer & Balaam, 1973, p. 40)
of potatoes responded differently to potash
fertilizers. The subcomponents of variance for Early Uses of ANOVA in Psychology
differences between manures and differences
between varieties were tested for significance Our first approach was to trace the devel-
against a variance due to variations between opment of variance techniques in psychology
plots. These data were reanalyzed in the by identifying early published articles in be-
1925(b) text because, according to Yates havioral research that used ANOVA or proposed
(1964), Fisher realized the need to partition it as a design and analysis alternative. An
the error term due to the arrangement of extensive review was made of all located
plots in the original design. articles that used ANOVA prior to 1940. In ad-
The second 1923 reference to ANOVA was an dition, a review was made of all located arti-
article by Student (Gossett, 1923), on testing cles through 1950 that recommended the use
varieties of cereals. Student quoted a letter of variance techniques in psychology or dem-
that he received from Fisher regarding the onstrated their application to psychological
breakdown of variances for trials and varie- research problems.
ties. Fisher presented a table in this letter Table 1 lists 17 published studies in psy-
that is nearly identical to the form in which chological research that used ANOVA prior to
ANOVA results are reported today. 1940. These studies were identified using
The 1925(b) text, Statistical Methods for personal communications from Cronbach
Research Workers, was the first didactic pre- (Note 1) and a systematic survey of approxi-
sentation of ANOVA, It was not a compre- mately 50 pre-1940 journals. Although some
hensive presentation, however, and in fact pre-1940 studies may not have been discov-
only two chapters out of eight were directly ered, Table 1 should contain the majority of
related to ANOVA. The remainder of the text published psychological research articles to
was devoted to a discussion of the scope and use ANOVA prior to 1940.
characteristics of statistics, the Poisson dis- A number of similar articles appear in
tribution, chi-square, and the t statistic. Be- Table 1. Three used covariance analysis
sides the presentation of ANOVA, the fourth (Dressel, 1939; Gaskill & Cox, 1937; Snede-
edition of this text (Fisher, 1932) included cor, 1935), 2 investigated memorization of
Fisher's first presentation of the analysis of piano music (Rubin-Rabson, 1937, 1939), 2
covariance (ANCOVA). studied differences in alpha rhythms between
Soon after the initial presentation of ANOVA, normal and abnormal individuals (Kreezer,
Fisher became acutely aware of the relation- 1939; Rubin, 1938), 3 applied ANOVA to
ship between design and analysis (Fisher, evaluating validity of test items and test
1926). The 1935 text, The Design oj Experi- reliability (Baker, 1939; Jackson, 1939; Lev,
ments, was the culmination of the interfacing 1938), and 4 were concerned with the quanti-
of design and analysis. This was the first tative study of trance personalities (Caring-
comprehensive presentation of factorial design ton, 1934, 1936, 1937; Thouless, 1937). Of
and analysis. The first 113 pages of this text the 17 articles listed, 16 involved research
are an elegant presentation of the concepts primarily relating to or using human subjects.
of null hypothesis testing, randomization, and Only 1 study (Crutchfield, 1939) used ANOVA
inductive inference. in research with animals. Twelve of the arti-
Fisher's contributions to statistics were cles were in psychological publications, 4
numerous, and although some colleagues appeared in educational journals, and 1 ap-
claimed that his work was not new in a revo- peared in a statistical journal.
lutionary sense, he clearly exerted a profound The earliest behavioral research to use
influence on contemporary experimental de- ANOVA was reported by Reitz (1934) in the
sign and analysis: Journal of Experimental Education. Reitz
presented the formulas for computing a z
It might be stated that experiment design as known value, noted the assumption of homogeneity
ANALYSIS OF VARIANCE 169
Table 1
Psychological Articles Using Analysis of Variance Prior to 1940
of variance, and checked his computed z was a direct application of covariance analy-
value by outlining the relationship between sis to a behavioral research problem. This
the correlation ratio (rj-) and z. Given that article represents one of the earliest uses of
only Fisher's editions of the 1925(b) text applied variance techniques. Smith, the head
were available to Reitz, this was a remarkably of the mathematics department at Iowa State,
competent use of the technique. wanted to determine if different grading prac-
Four of the earliest articles to use ANOVA tices by teachers were responsible for dif-
were in research on trance personalities ferences between mathematics classes or if the
(Carington, 1934, 1936, 1937; Thouless, different grades reflected different levels of
1937) through the London-based Society for achievement. Snedecor applied covariance
Psychical Research. After these four very analysis to the problem and tested for class
early articles, however, use of variance tech- differences on a common exam.
niques completely disappeared from the jour- A wide range of statistical sophistication
nal. was demonstrated by these early articles. The
Although Snedecor's (1935) covariance Snedecor (193S) and Gaskill and Cox (1937)
analysis of grades appeared in a statistical articles were complex statistical applications,
journal, we have included it here since it particularly in view of the newness of the
170 ANTHONY J. RUCCI AND RYAN D. TWENEY
techniques. Even more striking, however, was of Peters and VanVoorhis (1940) in a chap-
the extent to which the authors of these ter entitled "The Technique of Controlled
early articles acknowledged the new technique Experimentation":
being used. Reitz (1934), Gaskill and Cox
( 1 9 3 7 ) , and Thouless (1937), for example, ll seems scarcely necessary for us to say here that
care must be exercised to keep all other conditions
made a point of telling the reader that the constant in the two situations except the one experi-
statistical technique being used was both new mental factor; this is the law of the single variable,
and particularly appropriate for research so fundamental to all scientific experimentation.
problems in psychology. On the other hand, (p. 44V)
Rubin-Rabson (1939) did not acknowledge
The pervasiveness of the single-variable law
the newness of variance analysis, and Rubin
in behavioral research was felt well into the
(1938) cited no statistical text or article rele-
late 1930s and early 1940s.
vant to the statistical treatment used!
Baxter (1941) noted that Fisher's methods
Although the pre-1940 articles were pio-
had only recently begun to appear in psycho-
neering applications of ANOVA, some of these
logical journals, but these methods frequently
authors displayed uncertainty regarding the
appeared in journals of agriculture and biol-
technique. Kreezer (1939) provides the best
ogy. Baxter's article outlined the analogy
example of this uneasiness. After pointing out
between the agricultural terminology associ-
that F was equal to t~, Kreezer analyzed his
ated with Fisher's method and psychological
data using both t and F, saying, "Both meth-
terminology. The fundamental analogy pro-
ods have, nevertheless, been used in this study
posed by Baxter was
so that they may provide a check on each
other'' (p. 5 2 2 ) . Treatment: Soil: :
A systematic survey of psychological jour- Experimental variable: Organism.
nals through 19SO turned up 66 articles that
were identified as expository, written with Further, he stated that "treatment corre-
the stated intention of presenting a new sta- sponds to the traditional term independent
tistical technique to be applied to psychologi- variable" (p. 270). An article by Garrett and
cal research. These included articles that en- Zubin (1943) gave extensive treatment to po-
dorsed ANOVA for psychological research and tential applications of ANOVA in psychological
also those that demonstrated its application research. Grant (1944) published a rejoinder
to psychological research problems. to this article, pointing out errors of omission
Three articles, Crutchfield (1938), Baxter in Garrett and Zubin's discussion of some
(1941), and Garrett and Zubin (1943), were early psychological articles using ANOVA.
major efforts toward "psychologizing" the Many other articles that need not be re-
ANOVA technique. In 1938, Crutchfield il- viewed here endorsed ANOVA for psychology
lustrated the application of factorial design to (Bloomers & Lindquist, 1942; Burt, 1938;
a study of the perseverative performance of Crutchfield & Tolman, 1940; Dunlap, 1938,
rats in a string-pulling apparatus. Only the 1940, 1941; Garrett, 1943; Grant, 1944,
layout of the research design, in accordance 19SO; Hotelling, 1935; Royce, 19SO; Thou-
with Fisher's technique, was reported, not the less, 1939).
results of the study. Crutchfield was un- A major portion of the 66 expository arti-
equivocal in his endorsement of factorial cles were applications of variance analysis.
designs: The titles of a number of these articles began
with the words "An application of" ANOVA or
Whenever, in experimental, comparative, or social factorial design to a particular research issue.
psychology, a systematic investigation of the pri- Jackson's (1940) research bulletin was fre-
mary effects and the interacting effects of a number quently cited by later researchers. Other arti-
of experimentally controllable factors is being con-
ducted, the principles of efficient factorial design cles on applications ranged from studies of
can be invoked with inestimable benefit, (p. 341) educational development (Johnson & Tsao,
1945) to galvanic skin response data (Hag-
Contrast this quote by Crutchfield with that gard, 1949), from liberalism scores (Schrader,
ANALYSIS OF VARIANCE 171
1940) to reaction times (Baxter, 1942), and fact, as late as 1944, Grant used the future
from lightness and saturation (Helson, 1942) tense to refer to variance techniques in psy-
to practice effects (Owens, 1942). chology: "Analysis of variance and kindred
The Latin square design received much techniques developed by Fisher and his co-
early attention from psychologists. No fewer workers will very likely become valuable
than seven articles were found that were statistical tools in psychological research" (p.
devoted wholly or partially to the Latin 158).
square in research design (Baxter, 1941;
Bugelski, 1949; Edwards, 19SOb; Garrett & ANOVA Growth Curve
Zubin, 1943; Grant, 1944; Thomson, 1941).
An issue that began to receive attention in We carried out a frequency count of articles
the late 1940s was that of post hoc tests. The appearing in major psychological journals
common practice was to compute t tests be- from 1935 through 1952 to determine the
tween individual comparisons following a sig- size of the impact of ANOVA. Six major Amer-
nificant overall F value. Webb and Lemmon ican psychological journals were selected for
(19SO) and Walker (1947), however, com- analysis: American Journal oj Psychology,
mented on this practice. Webb and Lemmon Journal oj Applied Psychology, Journal oj
maintained that it was too rigidly applied, Educational Psychology, Journal of Experi-
and Walker saw post hoc techniques as a yet mental Psychology, Journal oj General Psy-
unresolved problem in research. chology, and Journal oj Psychology.
An examination of publication dates showed Every article from 1935 to 1952 in each
that 45 out of the 66 expository articles journal was reviewed to determine what type
(687c) appeared between 1938 and 194S. of statistical analysis was employed. A total
This was a critical period with respect to the of 6,457 articles was included in the count.
development of ANOVA techniques in psychol- Editorial articles, notes and discussions, and
ogy. Nearly all of these articles were explicit articles reporting new apparatus were not
introductions to variance techniques. This included.
suggests that at the time of the publication Figure 1 shows the proportion of articles
of these articles, the ANOVA was still not that used ANOVA, t tests, critical ratio tests,
widely used by psychological researchers. In and correlational analysis for each year
CRITICAL RATIO • • • • •
CORRELATIONAL
1935 36 38 40 42 44 46 48 50 52
YEAR OF JOURNALS
Figure 1. Proportion of articles using selected statistical tests as a function of year.
172 AXTHOXY J. RUCCI AND RYAX D. TWENEY
from 1935 through 1952. These categories are proportion of articles that used correlation in
the most reflective of the transition from the 1952 is nearly equal to that in 1935. This
conventional method (critical ratio) to the indicates that the use of correlational meth-
Fisher techniques. Use of both t and ANOVA ods remained relatively constant, even after
increased gradually prior to World War II, the Fisher methods were incorporated. With
declined during the war, and increased im- the aid of hindsight, it becomes clear that
mediately thereafter. Use of the critical ratio variance techniques enabled researchers to
technique declined in exact proportion to the fill the void that existed in experimental
increase in ANOVA and t, which confirms the methodology. To use Cronbach's terminology,
idea that Fisherian procedures were supplant- the correlational school probably would have
ing the conventional technique for examining been sustained whether Fisher's methods ap-
group differences. Note the marked similarity peared or not.
in the profiles for ANOVA and t. Every in- Some aspects of the frequency distribution
crease or decrease in the use of ANOVA is par- for individual journals deserve brief com-
alleled by a concomitant increase or decrease ment: The American Journal of Psychology
in /-test usage. By f949 use of the / test sur- and the Journal oj Applied Psychology
passed that of the critical ratio, and in 1950 showed increases in the use of ANOVA and t
ANOVA usage surpassed the critical ratio. Sur- up to 1942. From 1942 to 1945, however,
prisingly, use of the t test did not precede there was a decrease in the use of these
that of ANOVA, although it had been developed statistics. From 1945 on, use of ANOVA and t
much earlier and was very similar to the increased, with a concomitant decrease in the
already established critical ratio technique. use of the critical ratio. The percentage of
Figure 1 shows a decline in the use of articles that used both ANOVA and the t test
Fisher's techniques during the war years, surpassed that of articles that used the criti-
1942-194S. Note also the general increase in cal ratio only after 1947. The Journal oj Ex-
the use of the critical ratio during the same perimental Psychology and the Journal oj
period. Those research psychologists young Psychology showed the same pattern, though
enough to be inducted into military service growth of ANOVA was faster in the former and
were also those most likely to have had slower in the latter. The Journal oj General
graduate training in variance techniques Psychology was late to include ANOVA and did
(Chapanis, Note 2). Therefore, those re- not use it at all in 1948, although it increased
searchers most likely to use ANOVA were taken rapidly thereafter.
out of the publishing ranks during the war The results for the Journal oj Educational
years, which could account for the decline. In
Psychology were anomalous compared with
addition, those who continued to publish dur-
ing the war may have been those too old to the results of the other five. It evidenced the
be inducted and who lacked variance training. largest initial increase in ANOVA through 1942.
Their published research, then, would rely This may be attributed to a text by Lind-
more heavily on the critical ratio technique, quist (1940) on ANOVA for educational re-
and this could account for the increase in its searchers, which was widely cited. Although
use from 1943 to 1945. This explanation of a decrease in the use of variance techniques
the war year effect is consistent with Kuhn's from 1942 to 1946 was also evident in this
( 1 9 7 0 ) account of paradigm shifts. Kuhn has journal, the depression lasted much longer
suggested that new ideas are adopted differ- than in the other journals. It was not until
entially by young and old scientists. 1950 that the use of ANOVA surpassed its
The use of correlational analysis (also prewar use. Thus the paradox—Although ed-
shown in Figure 1) from 1935 to 1952 sug- ucational researchers initially adopted the
gests a striking conclusion: The Fisher meth- technique more quick!)' than others, the im-
ods did not supplant the use of correlational mediate postwar upsurge never materialized.
analysis. The curve of the use of correlational It was not until 1952 that usage of the criti-
analysis runs parallel to the abscissa. The cal ratio was surpassed by either ANOVA or /.
ANALYSIS OF VARIANCE 173
Table 2
.1 rticles Containing Statistical Treatments Related to Analysis of Variance.
Table 2 (continued)
researchers from Columbia University, Iowa articles using ANOVA. The two Snedecor
State College, the University of Iowa, and the (1934, 1937) texts, Fisher's (1935) text,
University of Minnesota. This is discussed in Goulden (1939), Rider (1939), Peters and
more detail later in this article. VanVoorhis (1940), and Walker (1943) ac-
A small subset of the textbooks listed in count for most of the citations through the
Table 3 was heavily cited by psychological early 1940s. In general, however, the texts by
Table 3
Statistical Textbooks Containing Analysis of Variance Coverage
Author Affiliation
Snedecor (1937) and Lindquist (1940) had the ordering by doctorate with each of the
the largest impact on psychological research- three category orderings. The rank-order cor-
ers well into the late 1940s. The Lindquist relation (Spearman's rho corrected for ties)
book represents the first comprehensive text between doctoral year and year of first vari-
on ANOVA that was intended for behavioral ance training was .53; between doctoral year
researchers. Snedecor was a statistician at the and year of first psychology course in ANOVA,
Iowa State College agricultural station, and p = .63; between doctoral year and first year
his text was oriented to agricultural research. of program requirement, p = AS (all ps <
However, a number of the early psychology .01). The older the doctoral program, (a) the
programs offering variance training used this earlier variance training was started, (b) the
text. earlier the department offered a variance
The texts by Kelley (1947), McNemar course, and (c) the earlier variance training
(1949), Johnson (1949), Cochran and Cox became a graduate degree requirement.
(19SO), Edwards (T9SOa), and Guilford In the 5 years after World War II, there
( 1 9 S O ) proved to be important to psycho- was a dramatic increase in the number of
logical researchers through the 1950s. psychology departments in which graduate
Once again the effect of the war years on students were receiving training in ANOVA
the use of ANOVA is seen in Table 3. Only techniques. This result coincides with a similar
three texts appeared between 1940 and 1946. increase in journal and textbook use of and
From immediately after the war, however, attention to ANOVA immediately following the
through 1950, the production of texts on war. Again, the return of young psychologists
ANOVA averaged nearly three per year, not to academic positions following the war may
including later editions of earlier texts. have brought an increased emphasis on vari-
ance techniques.
ANOVA Training in Psychology
Phylogeny of Variance Training
Our fifth and final approach was to deter-
mine when ANOVA was incorporated into the Based on the results of the preceding
training of psychologists. It was expected analyses, the line of training was traced for
that graduate training in ANOVA would have those researchers identified as prominent con-
lagged behind incorporation into the research tributors to the development of variance tech-
literature. niques in psychology. (See Ben-David & Col-
A questionnaire survey was conducted of lins, 1966, and Boring & Boring, 1948, for
88 psychology departments that offered grad- examples of the method used.) Figure 2 was
uate degrees in 1940. The questionnaire was constructed using 1940 editions of the Ameri-
intended to determine when ANOVA training can Psychological Association membership
was introduced into the graduate program at directory, textbook prefaces and acknowledge-
each department. Of the 88 surveys mailed, ments, and personal communications (Cron-
41 ( 4 7 / J - ) were returned. Fourteen of the bach, Note 1; Chapanis, Note 2; Humph-
returned surveys were unusable due to incom- reys, Note 3; Bechtoldt, Note 4; Cronbach,
plete information. Therefore, Table 4 lists Note 5; Edwards, Note 6; Gage, Note 7;
the results of the 27 usable questionnaires. Green, Note 8; Gulliksen, Note 9; Lind-
The universities are listed in the order in quist, Note 10; McNemar, Note 11; Ryan,
which the}' began to offer the doctorate in Note 12; Smith, Note 13) from a number of
psychology degree. In all three categories, individuals.
those departments offering the doctorate in The phylogeny of the development of Fish-
psychology prior to 1940 incorporated ANOVA er's techniques in psychology and in Ameri-
training earlier than other departments. The can statistics as well was centered around
27 departments were rank ordered by the three statistical researchers: G. W. Snedecor
year in which they began to offer the doc- at Iowa State College, Harold Hotelling at
torate and by the year of the three categories. Columbia University, and Palmer Johnson at
Rank-order correlations were computed for the University of Minnesota. These three
ANALYSIS OF VARIANCE 177
Table 4
Summary of Questionnaire Results for Incorporation of Variance Training
stand at the beginning of the entire line of 1931 he took a position as professor of Eco-
training in Fisherian statistics in the United nomics at Columbia University. Under Ho-
States. All three gained exposure to variance telling's influence Columbia University be-
techniques directly from Fisher himself. came one of the centers of statistical research
Hotelling visited Fisher's laboratory at the in the United States. Hotelling (1931) him-
Rothamsted Experiment Station in 1929. In self is responsible for T-, the multivariate
178 ANTHONY J. RUCCI AND RYAN D. TWENEY
^,'Cox
0**
**''^, ..--Cochran • -Lindquist
[U. Iowa)
/ I Iowa St.) • Gaskill
,, Bartlett
S''\ Walker
[Wisconsin |
-Hotel lings:-!- MeNemar :*''
„' Edwards
I Columbia )"%>„ I Stanford I "**»
v\ *% N
3errett
s
" * Humphreys ''.
I Northwestern I
IU. Washington)
S
\v ""•Chapanis
Zubin (Johns Hopkins!
^Johnson-
Tsao
I Minnesota I^VV*«.S """ 1
""."^^ ~~" Broiek
\^> v
N "Baxter
x
Alexander
educational researchers was the first bona fide began an uninterrupted increase immediately
statistics text devoted to the application of following the war. Psychologists became ac-
variance techniques in behavioral research. tively involved in statistical issues related to
This text was widely cited by researchers in the technique in the middle-to-late 1940s.
psychology who used ANOVA prior to 19SO. Textbooks in statistics written explicitly for
Of psychologists directly involved in the psychological researchers did not appear until
statistical development of ANOVA, the line of the late 1940s. Finally, graduate training in
training clearly traces back to Quinn McNe- ANOVA was the last in chronological order to
mar at Stanford University. McNemar, how- develop, the median year being 1951.
ever, received his exposure to statistical tech- Thus, introductory articles were followed
niques from Hotelling, who was an associate by a gradual increase in the application of
professor at Stanford University until 1931. the technique in the research literature. When
Through McNemar, at Stanford University, it became clear that the technique was es-
L. G. Humphreys was trained before going to tablishing itself as a primary method of de-
Northwestern University, where he was sta- sign and analysis, psychologists began refin-
tistical adviser for the research of Allen Ed- ing the technique to meet the unique
wards. Edwards became a prominent figure in requirements of behavioral research. Text-
the development of variance techniques in books written by psychologists for psycholo-
psychology, publishing numerous articles, an gists then appeared. When it was clear that
elementary textbook (Edwards, 1946), and in competent psychologists must receive ex-
1950 a comprehensive textbook of ANOVA. posure to variance techniques, graduate pro-
David Grant graduated from Stanford Uni- grams began offering and requiring courses in
versity in 1941 before going to the University ANOVA.
of Wisconsin. Grant was perhaps the most The history of psychology's incorporation
active of all psychologists both in introduc- of variance techniques can be logically di-
ing ANOVA to psychology and in applying it vided into three periods: (a) the expository
in his own research. From Stanford Univer- period from 1925 to 1942, (b) the war period
sity, therefore, came directly and indirectly from 1942 to 1945, and (c) the postwar
four of the most prominent names in the institutionalization period from 1946 to 1957.
development of variance techniques in psy- The expository period is characterized by
chology: McNemar, L. G. Humphreys, Ed- early attempts to use ANOVA, but even more
wards, and Grant. so by articles exhorting the use of variance
L. G. Humphreys may even have played an analysis by psychologists. The majority of
important role in introducing variance train- these articles appeared between 1936 and
ing at Yale University. While on a postdoc- 1945, with some of the most important (e.g.,
toral appointment at Yale in 1938, Humph- Baxter, 1941; Crutchfield, 1938; Crutchfield
reys reported on Fisher's techniques in a & Tolman, 1940; Garrett & Zubin, 1943)
seminar offered by Donald Marquis. In 1942, appearing between 1938 and early 1943.
at Yale, Chapanis received variance training The war years effect was pronounced. The
from Carl Hovland. Chapanis went on to gradual upsurge in the use of ANOVA between
introduce the ANOVA into applied experimental 1934 and 1942 was wiped out during World
research (Chapanis, 1951; Chapanis, Garner, War II. It was 1947 before use of the tech-
& Morgan, 1949). nique regained the high point it had reached
just before the war. The war, therefore, was
Discussion instrumental in how quickly the technique
was adopted and may have delayed its in-
Our results show that ANOVA was incorpo- corporation by some 5 years. This conclusion
rated into psychology in logical and orderly must be qualified, however. To establish such
steps. First came the expository articles from an effect conclusively would require extensive
1938 to 1945. The use of variance techniques review of the research conducted by psycholo-
in journals began to increase in the early gists for the military during the war. Many
1940s, was deterred during World War II, and psychologists may have received their initial
180 ANTHONY J. RUCCI AND RYAN D. TWENEY
exposure to variance analysis and experience service as psychologists. This period shows a
with the technique while in military service clear drop in the publication of articles using
(Chapanis, Note 2). If so, then the war may ANOVA from 1942 to 1946, whereas, at the
actually have accelerated use of the technique same time, ANOVA techniques were increas-
in the late 1940s. ingly used by military research psychologists
The postwar institutionalization period was (e.g., Chapanis & Schachter, Note 14;
dramatic. The use of ANOVA in the journals Schachter & Chapanis, Note 15). Following
began an uninterrupted increase immediately the war, these psychologists entered the aca-
following the war. By 19S2 nearly all psy- demic world and began to publish in regular
chology graduate programs were offering journals, using ANOVA. Further, there was
course work in ANOVA, clearly signaling its clearly resistance from older researchers.
institutionalization as an accepted analytic Peters (1943, 1944; Peters & VanVoorhis,
technique familiar to virtually all research 1940) was the most vociferous critic of Fish-
psychologists. er's techniques. He pointed out that Fisher
The proposed trichotomy of the history of presented only extensions of traditional sta-
psychology's adoption of variance techniques tistics, nothing fundamentally new. He re-
qualifies the usual claim (Hearnshaw, 1964; ferred to the ANOVA as "magic" on two occa-
Stanley, 1966) that the incorporation of vari- sions in print and warned,
ance techniques into psychology was a post-
World War II phenomenon. By 19S2 psychol- if educationalists and psychologists, out of some
ogy had completed a shift with respect to sort of inferiority complex, grab indiscriminately at
them [variance techniques] and employ them where
experimental design and analysis, a process they are unsuitable, education and psychology will
that began well before the war, in the 1930s. suffer another slump in prestige such as they have
Can this change be considered an instance often hitherto suffered in consequence of the pursuit
of a paradigm shift in Kuhn's (1970) sense? of fads. (Peters, 1943, p. 549)
The criteria that would have to be met to
substantiate such a claim are (a) that a major The last criterion for a paradigm—success-
change was produced in psychologists' way of ful incorporation of anomalies—does not seem
perceiving and/or approaching basic concep- to fit this case. Although examples can be
tual issues, (b) that the new techniques were found of pre-ANOVA research that could have
differentially adopted by younger psycholo- benefitted from use of ANOVA, it is not the
gists against the resistance of older workers, case that felt anomalies were present that
(c) that anomalies that could not be ac- needed to be attended to. ANOVA is, of course,
counted for with older techniques were in- a technique, not a theory, and one could ar-
corporated in the newer. gue that no methodological technique could
The first criterion can be supported by ever meet this criterion, by definition. If,
reference to Cronbach's (1957) characteri- however, one considers the concepts of
zation of the experimental discipline of psy- crossed classifications and interaction as being
chology. The approach to basic experimental anomalies prior to ANOVA, then, in that sense,
issues was radically different following the ANOVA did incorporate anomalies. Even so, we
adoption of Fisherian techniques. It is unde- feel that it is more reasonable to regard
niable that factorial design and analysis fa- ANOVA not as something fundamentally new,
cilitated the demise of the single-variable law in a revolutionary sense, but as an extension
in psychology. By 1957 it was standard ex- and explication of a set of design strategies.
perimental procedure to manipulate multiple Although ANOVA gave the experimenter a
levels of multiple independent variables. powerful set of analytic tools, the nature of
ANOVA was, therefore, the ideal analytic sta- psychological experimentation changed during
tistic for such designs. the period in question as psychologists moved
That the second condition held is suggested from single-variable studies to multiple-varia-
by the effect of the war on incorporation of ble studies. Thus, the history of ANOVA is
ANOVA. By and large, it was younger, less likely to be most interpretable as a component
established researchers who entered wartime of a broader, as yet unattempted, history of
ANALYSIS OF VARIANCE 181
the use of experimental research methods in Air Technical Service Command, Engineering
Division, Aero Medical Laboratory, October
psychology. 1945.
Thus, we feel that our study represents 15. Schachter, S., & Chapanis, A. Distortion in
only the beginning of a necessary but much glass and its effect on depth perception (Memo-
larger effort. We cannot legitimately assess randum Rep. TSEAL-695-48B). Wright Field,
the place of ANOVA in the history of psychol- Dayton, Ohio: Army Air Forces Air Technical
ogy without the context provided by knowl- Service Command, Engineering Division, Aero
Medical Laboratory, April 1945.
edge of the explicit ways in which experimen-
tation changed during the period in question.
Furthermore, neither statistical techniques nor References
design techniques can be assessed indepen- Alexander, H. W. A general test for trend. Psycho-
dently of the concomitant changes in the logical Bulletin, 1946, 43, 533-557.
empirical issues considered by psychologists Alexander, H. W. The estimation of reliability when
or independently of the changes in the theo- several trials are available. Psychometrika, 1947,
12, 79-99.
retical structures underlying psychological Ansbacher, H. L., & Mather, K. Group differences in
research. Our study is thus very much a size estimation. Psychometrika, 1945, 10, 37-56.
beginning effort only. By characterizing the Baker, K. H. Item validity by the analysis of vari-
growth and development of a single impor- ance. Psychological Record, 1939, 3, 242-248.
Bartlett, M. S. The use of transformations. Bio-
tant dimension of psychological method, we metrics, 1947, 3, 39-52.
hope it will contribute to the emergence of a Baxter, B. Problems in the planning of psychologi-
broader, more comprehensive history of scien- cal experiments. American Journal oj Psychology,
tific method in psychology. 1941, 54, 270-280.
Baxter, B. A study of reaction time using factorial
design. Journal oj Experimental Psychology, 1942,
Reference Notes 31, 430-437.
Ben-David, J., & Collins, R. Social factors in the
1. Cronbach, L. J. Personal communication, July origin of a new science: The case of psychology.
16, 1977. American Sociological Review, 1966, 31, 451-465.
Bloomers, P., & Lindquist, E. F. Experimental and
2. Chapanis, A. Personal communication, July 19,
statistical studies: Application of newer statistical
1977. techniques. Review of Educational Research, 1942,
3. Humphreys, L. G. Personal communication, July 12, 501-520.
18, 1977. Boring, M. D., & Boring, E. G. Masters and pupils
4. Bechtoldt, H. P. Personal communication, Au- among the American psychologists. American
gust 10, 1977. Journal oj Psychology, 1948, 61, 527-534.
5. Cronbach, L. J. Personal communication, July Brozek, J., & Alexander, H. A note on estimation of
25, 1977. the components of variation in a two-way table.
6. Edwards, A. L. Personal communication, Decem- American Journal oj Psychology, 1947, 60, 629-
ber 6, 1977. 637.
Brozek, J., & Alexander, H. The formula t~ = F.
7. Gage, N. L. Personal communication, July 21,
American Journal of Psychology, 1950, 63, 262-
1977. 269.
8. Green, B. F. Personal communication, August 10, Bugelski, B. R. A note on Grant's discussion of the
1977. Latin square principle in the design of experi-
9. Gulliksen, H. Personal communication, July 15, ments. Psychological Bulletin, 1949, 46, 49-50.
1977. Hurt, C. Recent developments of statistical methods
10. Lindquist, E. F. Personal communication, De- in psychology: 1. Occupational Psychology Land-
cember 4, 1977. marks, 1938, 12, 169-177.
11. McNcmar, Q. Personal communication, Decem- Carington, W. The quantitative study of trance
ber 20, 1977. personalities: 1. Proceedings oj the Society for
Psychical Research, 1934, 42, 173-240.
12. Ryan, T. A. Personal communication, July 27, Carington, W. The quantitative study of trance
1977. personalities: 2. Proceedings oj the Society for
13. Smith, P. C. Personal communication, May 10, Psychical Research, 1936, 43, 319-361.
1977. Carington, W. The quantitative study of trance
14. Chapanis, A., & Schachter, S. Depth perception personalities: 3. Proceedings of the Society for
through a P-80 canopy and through distorted Psychical Research, 1937, 44, 189-222.
glass (Memorandum Rep. TSEAL-69S-48N). Chapanis, A. Theory and methods for analyzing
Wright Field, Dayton, Ohio: Army Air Forces errors in man-machine systems. Annals of the
182 ANTHONY J. RUCCI AND RYAN D. TWENEY
.\eui York Academy of Sciences, 1951, 51, 1179- of squares for interactions in the analysis of vari-
1203. ance. Psychometrika, 1950, 14, 17-24.
Chapanis, A., Garner, VV. R., & Morgan, C. T. Eisenhart, C. The assumptions underlying the analy-
Applied experimental psychology. New York: sis of variance. Biometrics, 1947, 3, 1-21.
Wilej', 1949. Eisenhart, C., Hastay, M. W., & Wallis, W. A.
Clark, H. H. The language-as-fixed-effect fallacy: Selected techniques of statistical analysis. New
A critique of language statistics in psychological York: McGraw-Hill, 1947.
research. Journal of Verbal Learning and Verbal Engelhart, M. D. The analysis of variance and
Behavior, 1973, 12, 335-359. covariance techniques in relation to the conven-
Clark, H. H., Cohen, J., Smith, J. E. K., & Keppel, tional formulas for the standard error of dif-
G. Discussion of Wike and Church's comments. ference. Psychometrika, 1941, 6, 221-234.
Journal oj Verbal Learning and Verbal Behavior, Federer, W. T., & Balaam, L. N. Bibliography on
1976, IS, 257-266. experiment and treatment design—pre 1968. New
Cochran, W. G. Some consequences when the as- York: Hafner, 1973.
sumptions for the analysis of variance are not Festinger, L. An exact test of significance for means
satisfied. Biometrics, 1947, 3, 22-28. of samples drawn from populations with expo-
Cochran, VV. G., & Cox, G. M. Experimental designs, nential frequency distribution. Psychometrika,
New York: Wiley, 1950. 1943, S, 153-160. (a)
Coombs, C. H. The role of correlation in analysis of Festinger, L. A statistical test for means of samples
variance. Psychometrika, 1948, 13, 233-243. from skew populations. Psychometrika, 1943, 8,
Cronbach, L. The two disciplines of scientific psy- 205-210. (b)
chology. American Psychologist, 1957, 12, 671- Fisher, R. A. On the distribution of the standard
684. deviation of small samples: Appendix 1. to papers
Croxten, F. E., & Crowden, D. J. Applied general by "Student" and R. A. Fisher. Biometrika, 1915,
statistics. New York: Prentice-Hall, 1939. W, 522-529.
Crutchfield, R. S. Efficient factorial design and Fisher, R. A. Applications of "Student's" distribu-
analysis of variance illustrated in psychological tion. Metron, 1925, 5, 90-104. (a)
experimentation. Journal oj Psychology, 1938, 5, Fisher, R. A. Statistical methods for research work-
339-346. ers. London: Oliver & Boyd, 1925. (b)
Crutchfield, R. S. The determiners of energy expendi- Fisher, R. A. The arrangement of field experiments.
ture in string pulling by the rat. Journal oj Psy- Journal of the Ministry of Agriculture, 1926, 33,
chology, 1939, 7, 163-178. 503-513.
Crutchfield, R. S., & Tolman, E. C. Multiple-varia- Fisher, R. A. Statistical methods for research work-
ble design for experiments involving interaction of ers (4th ed.). London: Oliver & Boyd, 1932.
behavior. Psychological Review, 1940, 47, 38-42. Fisher, R. A. The design of experiments. London:
Davenport, C. B., & Ekas, M. P. Statistical methods Oliver & Boyd, 1935.
in biology, medicine, and psychology. New York: Fisher, R. A. Statistical methods for research work-
Wiley, 1936. ers (14th ed.). London: Oliver & Boyd, 1970.
Dressel, P. The effect of high school grades on col- Fisher, R. A., & MacKenzie, W. A. Studies in crop
lege grades. Journal oj Educational Psychology, variation: 2. The manurial response of different
1939, JO, 611-616. potato varieties. Journal of Agricultural Science,
Dunlap, J. W. Recent advances in statistical theory 1923, 13, 311-320.
and applications. American Journal of Psychology, Fisher, R. A., & Yates, F. Statistical tables for bio-
1938, 51, 558-S71. logical, agricultural, and medical research. New
Dunlap, J. W. Applications of analysis of variance York: Hafner, 1938.
to educational problems. Journal oj Educational Garrett, H. E. AN OVA in psychological research.
Research, 1940, 33, 434-442. Journal of Educational Research, 1943, 36, 631-
Dunlap, J. W. Recent advances in statistical theory 632.
and applications. American Journal oj Psychology, Garrett, H. E. Statistics in psychology and educa-
1941, 54, 583-601. tion. New York: Longmans, Green, 1947.
Edwards, A. L. Statistical analysis for students in Garrett, H. E., & Zubin, J. The analysis of vari-
psychology and education. New York: Rinehart, ance in psychological research. Psychological Bul-
1946. letin, 1943, 40, 233-267.
Edwards, A. L. Experimental design in psychologi- Gaskill, H. V., & Cox, G. M. 1. Respiration: Use
cal research. New York: Rinehart, 1950. (a) of analysis of variance and covariance in psycho-
Edwards, A. L. Homogeneity of variance and the logical data. Journal of General Psychology, 1937,
Latin square design. Psychological Bulletin, 1950, 16, 21-38,
47, 118-129. (b) Gilliland, A. R., & Humphreys, D. W. Age, sex,
Edwards, A. L. On the use of interactions as "error method and interval as variables in time estima-
terms" in the analysis of variance. Educational tion. Journal oj Genetic Psychology, 1943, 63,
and Psychological Measurement, 1950, 10, 214- 123-130.
223. (c) Godard, R. H., & Lindquist, E. F. An empirical
Edwards, A. L., & Horst, P. The calculation of sums study of the effect of heterogeneous within-groups
ANALYSIS OF VARIANCE 183
variance upon certain /''-tests of significance in Johnson, P. 0., & Neyman, J. Tests of certain linear
analysis of variance. Psychometrika, 1940, S, 263- hypotheses and their application to some educa-
274. tional problems. Statistical Research Memoirs,
Gossett, W. S. (pseudonym, Student). The probable 1936, ;, 57-93.
error of a mean. Blomelriha, 1908, 6, 1-25. Johnson, P. O,, & Tsao, F. Factorial design in the
Gossett, W. S. (pseudonym, Student). On testing determination of differential limen values. Psycho-
varieties of cereals. Biometrika, 1923, 25, 271-293. metrika, 1944, 9, 107-146.
Gottsdanker, R. M. An experimental study of fixa- Johnson, P. 0., & Tsao, F. Factorial design and
tion of response by college students in a multiple covariance in the study of individual educational
choice situation. Journal oj Experimental Psy- development. Psychometrika, 1945, 10, 133-162.
chology, 1939, 25, 431-444. Kelley, T. L. A variance-ratio test of the uniqueness
Goulden, C. H. Methods of statistical analysis. of principal-axis components as they exist at any
Minneapolis, Minn.: Burgess, 1939. stage of the Kelley iterative process for their
Grant, D. A. On "The analysis of variance in psy- distribution. Psychometrika, 1944, 9, 199-200.
chological research." Psychological Bulletin, 1944, Kelley, T. L. Fundamentals of statistics. Cambridge,
41, 158-166. Mass.: Harvard University Press, 1947.
Grant, D. A. The Latin square principle in the Kogan, L. S. Analysis of variance—Repeated mea-
design and analysis of psychological experiments. surements. Psychological Bulletin, 1948, 45, 131-
Psychological Bulletin, 1948, 45, 427-442. 143.
Grant, D. A. Statistical theory and research design. Kogan, L. S. Variance designs in psychological re-
In C. P. Stone (Ed.), Annual review of psychol- search. Psychological Bulletin, 1953, 50, 1-40.
ogy (Vol. 1). Stanford, Calif.: Annual Reviews, Krcezcr, G. Intelligence level and occipital alpha
19SO. rhythm in the Mongolian type of mental defi-
Guilford, J. P. Fundamental statistics in psychology ciency. American Journal oj Psychology, 1939, 52,
and education. New York: McGraw-Hill, 1950. 503-532.
Haggard, E. A. On the application of analysis of Kuhn, T. S. The structure oj scientific revolutions.
variance to GSR data: 1. The selection of an Chicago: University of Chicago Press, 1970.
appropriate measure. Journal of Experimental Lev, J. Evaluation of test items by the method of
Psychology, 1949, 39, 378-392. analysis of variance. Journal of Educational Psy-
Hearnshavv, L. S. A short history of British psy- chology, 1938, 29, 623-630.
chology. New York: Barnes & Noble, 1964. Lewis, D. Quantitative methods in psychology. Iowa
Helson, H. Multiple-variable analysis of factors City, Iowa: Bookshop, 1948.
affecting lightness and saturation. American Jour- Lindquist, E. F. Statistical analysis in educational
nal of Psychology, 1942, 55, 46-57. research. Boston: Houghton-Mifflin, 1940.
Hotelling, H. The generalization of "Student's" Lindquist, E. F. Goodness of fit and trend curves
ratio. Annals oj Mathematical Statistics, 1931, 2, and significance of trend differences. Psycho-
360-378. metrika, 1947, 12, 65-78.
Hotelling, H. Review of Snedecor, G. W.: Calcula- Mahalanobis, P. C. Professor Ronald Aylmer Fisher.
tion and interpretation of the analysis of variance Sankhya, 1938, 4, 265-272. •
and covariance. Journal oj the American Statisti- Mann, H. B. Analysis and design oj experiments.
cal Association, 1935, 30, 118. New York: Dover, 1949.
Hotelling, H. Dr. Peters' criticism of Fisher's statis- McNcmar, Q. Psychological statistics. New York;
tics. Journal oj Educational Research, 1943, 36, Wiley, 1949.
707-711. Mood, A. Introduction to the theory oj statistics.
Hotelling, H. The impact of R. A. Fisher on statis- New York: McGraw-Hill, 1950.
tics. Journal of the American Statistical Associa- Mueller, C. G. Numerical transformations in the
tion, 1951, 46, 35-46. analysis of experimental data. Psychological Bul-
Hoyt, C. Test reliability by analysis of variance. letin, 1949, 46, 198-223.
Psycttometrika, 1941, 6, 153-160. Owens, VV. A. A new technique in studying the
Humphreys, L. G. The strength of a Thorndikian effects of practice upon individual differences.
response as a function of the number of practice Journal oj Experimental Psychology, 1942, 30,
trials. Journal oj Comparative Psychology, 1943, 180-183.
35, 101-110. Peters, C. C. Misuses of the Fisher statistics. Jour-
Jackson, R. W. B. Reliability of mental tests. Brit- nal of Educational Research, 1943, 36, 546-549.
ish Journal oj Psychology, 1939, 29, 267-287. Peters, C. C. Interaction in analysis of variance in-
Jackson, R. W. B. Application of the analysis of terpreted as intercorrelation. Psychological Bulle-
variance and covariance methods to educational tin, 1944, 41, 287-299.
problems. Department of Educational Research, Peters, C. C., & VanVoorhis, W. R. Statistical pro-
University of Toronto Bulletin, 1940, 11, 67-74. cedures and their mathematical bases. New York:
Johnson, P. 0. Use of Fisherian statistics. Journal McGraw-Hill, 1940.
oj Educational Research, 1943, 36, 627-630. Reitz, W. Statistical techniques for the study of
Johnson, P. 0. Statistical methods in research. New institutional differences. Journal oj Experimental
York: Prentice-Hall, 1949. Education, 1934, 3, 11-24.
184 ANTHONY J. RUCCI AND RYAN D. TWENEY