Clothing As Communication: an Empirical Investigation

Rebecca H. Holman, Pennsylvania State University
ABSTRACT - This paper reports an empirical study of how clothing expresses something about the user. It continues a tradition of research on product meaning but explicitly adopts a communication paradigm. The "language" of clothing use in one social system, one situation, and one role was partially decoded and its meaning explained in terms of attributions made to a user.
[ to cite ]:
Rebecca H. Holman (1980) ,"Clothing As Communication: an Empirical Investigation", in NA - Advances in Consumer Research Volume 07, eds. Jerry C. Olson, Ann Abor, MI : Association for Consumer Research, Pages: 372-377.

Advances in Consumer Research Volume 7, 1980     Pages 372-377

CLOTHING AS COMMUNICATION: AN EMPIRICAL INVESTIGATION

Rebecca H. Holman, Pennsylvania State University

[This paper is drawn, in part, from the author's doctoral dissertation. The support given this research by the Eugene and Dora Bonham Memorial Fund, The University of Texas at Austin is acknowledged. Robert A. Peterson and Mark I. Alpert are thanked for their assistance with the original work. The helpful comments from two anonymous reviewers are also appreciated.]

ABSTRACT -

This paper reports an empirical study of how clothing expresses something about the user. It continues a tradition of research on product meaning but explicitly adopts a communication paradigm. The "language" of clothing use in one social system, one situation, and one role was partially decoded and its meaning explained in terms of attributions made to a user.

INTRODUCTION

The literature of consumer behavior provides numerous comments upon the phenomenon of products serving as a means by which consumers express something about themselves. An early paper by Levy stated that "People buy things not only for what they can do, but for what they mean" (1959, p. 118). Robertson elaborated upon Levy by asserting:

Products vary in the degree to which social-symbolic meaning is important. Cars and clothing are both products which are high in visual display and recognized in our society as "saying something" about a person (1970, p. 3).

As Belk (1978) pointed out, and attempted to correct, systematic attempts to study products as means by which individuals communicate with one another have been few in number. The work presented here is in intellectual sympathy with Belk, although differing somewhat in methodology. Unlike Belk, this research focuses upon one product system, that of clothing, and explicitly studies a single product usage situation for one social system. The intent of this work is similar to Belk's, namely to initiate a stream of research examining the impact of consumption of visible products upon those observing the consumption, but differs from Belk in that a communication paradigm specifically underlies the research. The explicit assumptions of this research are that clothing use is communicative; [While many empirical studies have demonstrated the in-formation-value of clothing (especially in a first impression situation), relatively few have addressed the issue of how much information is conveyed about an individual relative to other cues (facial expression, body type, verbal behavior, etc.). One noteworthy exception is Buckley and Roach (1974).] that being communicative, its use is systematic within a social system (much like verbal language); and that if the use is systematic, it can be decoded ["Decoding" means that the recurrent features of the system are analyzed and described in a logical manner. For more information on what is involved in communication system decoding see Scheflen (1966).] and described much as verbal communication has been described by structural linguists (see Hill 1958, for example). The research presented here illustrates this particular approach to product use as an expression of a variety of attributes about the user.

LITERATURE REVIEW

Clothing was selected as the focus of study here for two reasons. One was that many authors, in writing either about products as indicators of "something" about their users or about products as forms of nonverbal [A very large body of research on nonverbal communication has sprung up during the past twenty-five years. Relatively little of this, however, deals with products as communication modalities. See Scheflen (1974, p. 42) for a brief synopsis of the various ways in which humans communicate, and Knapp (1978) for a recent overview of nonverbal communication research.] communication, have listed clothing as one category likely to function in this manner. (See Holman 1976, p. 49-51, and Holman forthcoming, for discussions of the authors that have taken one or the other of these approaches.) The other reason that clothing was selected for study is because of a long and very rich literature dealing with clothing's ability to express something about its user. [Some works have explicitly dealt with this topic in great detail. See Horn (1968) and Laver (1969) for two excellent examples.]

One of the earliest authors to assert that clothing consumption was more than for mere protection of the body was Veblen (1953, originally published in 1899) who developed a theory of fashion centering on the chattel status of those being required to wear the outrageous clothing of the times (e.g., women and servants). Veblen's work has been pursued in recent years and represents one of the major approaches to clothing use, specifically that it symbolizes some of the core values of society. Thus, Cassell (1974), who deals with clothing and the women's liberation movement, is typical of this type of approach; empirical studies documenting the effects of clothing upon behavior toward people dressed differently from one another can be found in Lefkowitz, Blake, and Mouton (1955); Bickman (1971); Suedfeld, Bochner, and Mates (1971); Darley and Cooper (1972); and Buckley and Roach (1974).

Another major focus of study on clothing treated it as an extension of the psychological state of its user. The first work of this kind may have been Dearborn (1918), but the approach has been popular ever since publication of his monograph, showing up most recently in Rosenfeld and Plax (1977) who constructed an instrument to determine personality differences in those using clothing differently from one another.

A third major theme of research on clothing conceptualizes clothing as indicative of the wearer's group membership or position within a group. Crawley (1931) addressed this in his paper on primitive cultures as did Gill (1931) for modern society. Bogatyrev's work (1971 published originally in 1937) and that of Barthes (1967) remain the most comprehensive works of this type, although Lasswell and Parshall (1961); Coursey (1973); and Jonaitis (1978) provide very recent examples.

In the above-cited works (and others which have not been referenced due to space limitations) there has been only one attempt to specifically decode the communication system arising out of clothing use. This one was the work of Barthes (1967) whose research was limited by the fact that he dealt not with actual consumption or use of clothing, but with description of clothing illustrations taken from fashion magazines. The advantages of being able to specify the "language" code for clothing use lie in more precise understanding of clothing use, and hence a better ability to satisfy consumers' needs. The research described below was a first attempt at decoding clothing usage within one context, and thus makes a contribution as being the first attempt to describe actual usage of clothing as a language.

METHODOLOGY

Because of the assumption that clothing use is systematic only within one social system and for one usage situation, it was deemed essential to hold these factors constant for an initial study of the communicational code for clothing. Additionally, the role played by the individuals studied in the chosen situation was held constant, a procedure in line with Belk's conclusions (1978, p. 46). Due to the exploratory nature of this research, convenience was the major criterion used in deciding whom to study. Thus, women students at The University of Texas at Austin while on the way to (or from) class formed the social system, the situation, and the role studied here.

Two data collection and analysis phases were used for the study. The first was for the purpose of obtaining a sample of the clothing behavior, so that regularities of the usage could be detected. The second phase of the research was to determine what the discrete clothing messages, identified in the first phase of the research, meant to members of the system. These two phases of research are necessary for understanding the communicational code for any type of behavior. More detail about the rationale for this appears in Scheflen (1966; 1974).

Phase I Procedures

Clothing, as a relatively non-rapid-fading communication code (i.e., it does not change in form as rapidly as do eye movements for example), can be completely and unambiguously recorded using color photography. Photographic slides were used here because of their economy, portability, and ease of enlargement to show detail.

Almost one thousand slides of women students on the university campus were obtained during one week in the Fall of 1973. A woman [Women's clothing was arbitrarily selected for study; men's clothing could also have been chosen, but as the systems appear to be separate, one or the other had to be the focus of this preliminary investigation, or the complexity would have been out of bounds.] was operationally defined as a student by the presence of books, notebooks, and/or a hook bag, satchel, or briefcase. Only full head to toe, front or three-quarters front photographs were used. Half of the slides were randomly eliminated from this analysis, so that subsequent analysis could be used to validate, the results obtained here. While photographs were taken on all days of the week and in a large number of locations during all times at which students were going to or coming from class, no formal randomization of sampling times and locations was used. It was felt more important to obtain a sample that would encompass the range of variation present in the population, and to record the clothing as it was actually worn on campus, than to insist upon a true random sample. All photographs were taken unobtrusively to ensure as little interference with the behavior as possible. Thus it was felt that the sample obtained here was representative, if not strictly probabilistic. After examining the slides to delete any that contained unusable photographs, an enumeration of individuals yielded a final sample of 392 different clothing ensembles.

Three individuals were used to code the ensembles in terms of a system that had been developed especially for unambiguous, exhaustive, and objective description of clothing (see Holman 1976; Holman forthcoming). The system used produced 171 numeric codes which recorded the form of the clothing (omitting all details of the cloth used to construct the clothing). For ease of coding, the body was divided into twenty-one areas (e.g., head, upper thigh, lower arms, etc.). Coders recorded eight pieces of information which described the clothing present at each of the body areas. An additional three codes noted whether the clothing was bilaterally symmetrical (a simplifying assumption made to reduce the total number of codes); whether the legs were distinguishable by the clothing (i.e., the presence of pants); and whether the arms were distinguishable by the clothing (i.e., the presence of sleeves, not a cape). The eight aspects (e.g., presence of cloth, presence of an opening or closing) for each of the twenty-one areas, plus the three additional codes yielded the 171 notations made for each ensemble. [Numeric codes used here were strictly nominal so that a score of 2 on any variable for any area of the body was as dissimilar from a 3 as it was from a 9; furthermore, no ordinal relationships were implied by the numbers. The only ordinary mathematical significance was with the code 0, which always indicated absence of a characteristic.]

Raters were thoroughly trained on the system and reported no difficulty in learning the system. Formal inter- and intra-rater reliability measures were obtained, and ranged from .8418 to .9276. [Inter- and intra-rater reliability coefficients were defined as m/n where m = number of matches, and n = number of codes used. Inter-rater reliability coefficients were based upon twenty ensembles rated by each pair of raters; intra-rater reliability coefficients were based upon ten ensembles coded twice by one rater, once at the beginning of the coding task and once at the end of it.] Thus, it was concluded that the system was used consistently across all raters, and from one coding to a recoding at a later point in time (the intra-rater reliability coefficients).

A matrix of similarity between each pair of ensembles was produced from the data generated by the raters. A m where m is the simple matching coefficient defined as m/n, number of "matches" (i.e., identical codes on one variable), and n is the number of variables being compared (171 in this case), was the data entered in the similarity matrix. Such a similarity matrix represents the proportion of codes for which each pair of ensembles is identical. Since the purpose of this phase of the data collection was to identify the clothing "messages" used by the members of the social system being studied while in the situation and role being enacted, such a matrix can indicate the similarities and dissimilarities in clothing usage (as recorded by the notation system) present in the data gathered.

Hierarchical cluster analysis using Ward's method (Anderberg 1973, p. 162) formed an initial partitioning of the data. Then the data from the configuration so-obtained were regrouped using a nonhierarchical algorithm referred to as Forgy's method (see Forgy 1965; Anderberg 1973, p. 161). These procedures produced groups of ensembles that were maximally similar to one another and maximally dissimilar from all other ensembles or groups of ensembles. The final configuration of groups of clothing ensembles should be a good representation of the types of clothing ensembles present in the obtained sample and thus should be the clothing messages present in the system, situation, and role from which they were sampled (if the sample is representative, and the data coding manipulations valid).

However, knowing the clothing messages does not provide insight into the significance or meaning of those messages unless further information is obtained. The purpose of the second phase of the research was to gain some insight into just what was indicated to members of the social system when an individual was observed wearing selected clothing in the situation and role for which the first phase data were representative.

Phase II Procedures

Phase II was designed to answer two research questions.

1.  Is the meaning attributed to person X's wearing of an ensemble different from the meaning attributed to X's wearing of any other ensemble (where ensembles are defined and limited by Phase I results)?

2.  What is the content of each message about person X generated by her use of each ensemble?

The meaning of person X's use of clothing was determined from attributions made about X by others. Specifically, subjects (students recruited from classes at The University of Texas at Austin, and hence members of the social system) were presented one of six photographs of a woman and asked to make attributions about her.

The woman in the photographs was dressed in ensembles representative of six clusters (those that were easily interpretable), obtained in Phase I. (A modal description of each clothing cluster provided the description of the clothing to be used, which was then matched with the model's clothing.) Photographs were taken as the model walked past the same spot on campus and were matched for similarity in posture and facial expression. The model carried books and a notebook, and while no other students were visible in the background, buildings of the university clearly identified the situation to subjects; the books identified the model's role within the situation.

Booklets containing one photograph were randomly distributed to subjects who were asked to provide biographical information about themselves, and were asked to describe the model in terms of twenty-seven attributes using a six-point scale ranging from "very definitely yes" to "very definitely no." The attributes selected for use in the study were obtained from past research on clothing, from instrument pretesting, and from a focus group interview with five male students.

The following instructions were given subjects:

We are interested in impressions that people form when first meeting another person. Study the photograph below. Assume that you have just seen this girl (the photograph was taken recently on campus). Then answer the questions on the opposite page, stating what you think this girl is like. Please circle the number of the answer that best expresses your thoughts about her for each question. Work quickly and do not go back to change answers. Remember that we are interested in your first impressions.

A great deal of research exists on the importance of first impressions in human interaction, and upon the role of clothing in first-impression formation. See Kleinke (1975) for a review of previous research on this topic.

RESULTS

Phase I Results

As mentioned, a total of 392 ensembles were clustered. The hierarchical cluster analysis determined that ten groups might be the ideal number of clusters, as a merger of two groups to form a total of nine groups produced an increase of 30.6% in total within group error. This was the largest percentage change in error except at the step from six groups to five, at which point there was an increase in within-group error of 37.5%. Examining these two solutions (the ten-group and the six-group) revealed that the ten-group configuration was more easily interpretable, and hence was the one chosen.

The nonhierarchical procedures yielded convergence in groups after three iterations. Because cluster analysis is a method requiring a great deal of researcher interpretation, the use of a further criterion of "ease of interpretability'' was deemed appropriate at this stage of research and yielded the six groups referred to previously.

These six clusters represented 142 data points, leaving 250 data points in groups not readily interpretable. The balance of the communicational structure remains unanalyzed at this point in time, but can serve as the basis for further research, should it be desired. The cluster analysis identified six distinctive clothing messages which served as input into Phase II. This was all that was intended from Phase I, so that it may be judged to have been successful.

Further analysis of cluster membership was necessary before stimulus items to use in Phase II could be constructed. A modal description of cluster members was obtained for each cluster. The modal description was a valid measure of "central tendency" of the cluster because the codes used for describing the clothing were strictly nominal, prohibiting any other mathematical manipulation. As stated, these modal descriptions were used to select the clothing to be worn by the model in the photography stage of Phase II. An examination of the slides from which the data were originally derived facilitated the interpretation of the modal descriptions.

Phase II Results

A two group discriminant analysis tested the hypothesis of no differences in attributions between males and females and yielded an F = 1.710 (p = .021). Univariate F's revealed that perceptions of the age of the model was the only variable discriminating between males and females. For ease in interpreting the results, the age variable was deleted and male and female subjects were treated identically for the balance of the analysis.

A six-group discriminant analysis produced an overall F of 1.628 (p = .0001) indicating significant differences in attributions across groups of subjects exposed to different clothing ensembles. The null hypothesis of no differences in perceptions due to clothing was rejected.

One way analysis of variance for each variable resulted in significant differences for eight variables (p < .05). A summary of the analysis of variance is presented in Table 1.

Significant differences among groups seeing each clothing ensemble for the eight variables showing significant differences are presented in Table 2. An ex post facto examination of the photographs revealed the presence of an eyebrow furrow for some of the photographs and not for others, and consequently differences between the two groups were examined. Table 2 also presents the significant differences resulting from the comparison of the group with eyebrow furrow and the group without eyebrow furrow. (Two group discriminant analysis for this latter comparison yielded F = 1.709 (p = .025), rejecting the hypothesis of no difference due to the eyebrow furrow.)

DISCUSSION OF RESULTS OF PHASE II ANALYSIS

It is a bit discouraging to note that only two variables, fashionable and sexy looking were free of any possible confound due to the presence of the eyebrow furrow. Of interest, however, is that while both stimulus-objects B and C contained an eyebrow furrow, they differed in terms of perceived femininity (B less feminine than C). This suggests that one basis for the perception of femininity lies in differences between B and C. Such an observation, however, cannot be generalized to other stimulus objects and the balance of the discussion of results focuses upon fashionability and sexy looking.

TABLE 1

SUMMARY OF ANALYSES OF VARIANCE

Stimulus-object A, C, D, E, and F were perceived to be significantly different from B in fashionability (B less fashionable). (Table 3 describes the clothing used in the study and how the clothing pieces were combined to form the different ensembles used in photography.) Because the shirt (blouse) and shoes in B were identical (in form) to those located in other ensembles these perceptions must have been due to a light-dark color dimension not properly controlled in the study, or to differences in pants, belt, or interactions. The pants in B were more straight-legged than were popular at the time of the study, and the belt had a wide buckle, a type worn frequently also by males. It is not possible to determine which of these formed the "unfashionable" cue, although further experimentation could do so.

Stimulus-objects A was perceived as being significantly more sexy looking than stimulus-objects B, E, and F. Exposure of the body has been informally linked to perceived sexiness. These data support that hypothesis. Ensemble A contained short cut-off jeans; ensembles B, E, and F contained long loose jeans. Ensembles C and D, while not significantly different from B, E, and F, were perceived as more sexy; C's dress was short exposing the legs; D's blouse was scooped low and outlined the breasts. The blouse in E also had a low scooped neck, but the breasts were obscured by the looseness of the blouse. Ensemble F, which contained an overblouse that obscured the entire torso, was perceived as being the least sexy looking except for B (the perception of which may have been affected by perceptions of the eyebrow furrow).

Presence of the eyebrow furrow seemed to indicate non-sorority membership, more enjoyment of beer drinking, less use of personal hygiene products, more sympathy with women's liberation, more athletic, less feminine, less good sense of humor, and a less conservative lifestyle. These variables appear to comprise a stereotyped pro-feminist, almost masculine, image when considered together. Thus, a tentative conclusion is that the presence of the eyebrow furrow may have connoted feminism, in this context, when ignoring the effect, if any, of clothing worn. The research design used here did not allow for detection of any interactions of the kinesic factors with the clothing factors. Due to the overlap in variables showing significance for clothing and for facial expression, it is reasonable to posit interactions. Thus, any interpretation of the effect of either clothing or eyebrow furrow must be tempered by consideration of the effect of the other.

CONCLUSIONS

This study, while limited in its design (analysis of only one social system, one situation, and one role) has been successful in achieving its purpose. It successfully identified six different clothing messages and investigated part of the meaning of those messages. The interpretation of the meaning was hampered due to a facial expression confound, but some interpretations were possible none-the-less.

The study's primary strength lies in its demonstration of an approach to products as communication not in evidence in the past. A communication paradigm formed the theoretical base, and a research procedure was adopted which operationalized that paradigm. This research opens the way for further work on clothing which is clearly needed, but also for research into other product categories as they too are used as communication.

An interesting side note also results from what could be perceived as a flaw of the study, improper control of facial expressions. While the design used here did not permit detection of interactions between clothing use and facial expressions, such an interaction was strongly suggested. If in fact such interactions are significant, the whole issue of how people communicate by using products (such as clothing) is greatly complicated. What should be studied is not product use in isolation from other forms of communication, but as it is integrated with both verbal and nonverbal forms.

TABLE 2

ATTRIBUTIONS MADE TO EACH CLOTHING STIMULUS OBJECT (SIGNIFICANT VARIABLES ONLY a < .05)

Of even more significance to consumer behavior researchers is the overall importance of products (like clothing) as factors in human communication. Research exists that suggests that appearance is extremely important in some situations (Walster, Aronson, and Abrahams 1966, for example), and that clothing is an extremely important cue in first-impression formation (Hamid, 1969). There is one study that even finds carryover effects due to initial impressions based upon clothing cues (Coursey 1973). What remains to be sorted out, in addition to the interactions of product-based and body-based modes of communicating, is the extent to which cues transmitted in one mode (e.g., facial expressions) can be supplanted by cues transmitted by products. Naturally, it follows that investigation is needed into the extent to which consumers purchase products for the purpose of communicating with them and if these serve as redundant messages or as additional ones. These and other topics result in the conclusion that the entire area remains one promising challenge to consumer behavior researchers in the future.

TABLE 3

DESCRIPTION OF CLOTHING AND IDENTIFICATION OF ENSEMBLE PIECES

REFERENCES

Anderberg, Michael (1978), Cluster Analysis for Applications, New York: Academic Press.

Barthes, Roland (1967), SystTme de la Mode, Paris: Editions du Seuil.

Belk, Russell W. (1978), "Assessing the Effects of Visible Consumption on Impression Formation" in Advances in Consumer Research, Vol. V, ed. H. Keith Hunt, Ann Arbor: Association for Consumer Research.

Bickman, Leonard (1971), "The Effect of Social Status on the Honesty of Others," The Journal of Social Psychology, 85, 87-92.

Bogatyrev, Petr (1971), The Functions of Folk Costume in Moravian Slovakia, (translated by Richard G. Crum), The Hague: Mouton.

Buckley, Hilda Mayer and Roach, Mary Ellen (1974), "Clothing as a Nonverbal Communicator of Social and Political Attitudes," Home Economics Research Journal, 3, 94-102.

Cassell, Joan (1974), "Externalities of Change: Deference and Demeanor in Contemporary Feminism," Human Organization, 1, 85-94.

Coursey, Robert D. (1973), "Clothes Doth Make the Man, in the Eye of the Beholder," Perceptual and Motor Skills, 36, 1259-1264.

Crawley, Alfred Ernest (1931), Dress, Drinks and Drums; Further Studies of Savages and Sex, London: Metheun & Co., Ltd.

Darley, John M. and Cooper, Joel (1972), "The 'Clean for Gene' Phenomenon: The Effect of Students' Appearance on Political Campaigning," Journal of Applied Social Psychology, 2, 24-33.

Dearborn, George Van Ness (1918), The Psychology of Clothing, Princeton, NJ and Lancaster, PA : Psychological Review Co.

Forgy, E. W. (1965), Cluster Analysis of Multivariate Data: Efficiency Versus Interpretability of Classifications, Riverside, CA: Biometric Society Meetings.

Gill, Eric (1931), Clothes: An Essay Upon the Nature and Significance of the Natural and Artificial Integuments Worn by Men and Women, London: J. Cape.

Hamid, Paul N. (1969), "Changes in Person Perception as a Function of Dress," Perceptual and Motor Skills, 29, 191-194.

Hill, Archibald A. (1958), Introduction to Linguistic Structures, New York: Harcourt Brace.

Holman, Rebecca H. (1976), "Communicational Properties of Women's Clothing: Isolation of Discriminable Clothing Ensembles and Identification of Attributions Made to One Person Wearing Each Ensemble," unpublished dissertation, Department of Marketing, The University of Texas--Austin.

Holman, Rebecca H. (forthcoming), "A Transcription and Analysis System for the Study of Women's Clothing Behavior," Semiotica.

Horn, Marilyn J. (1968), The Second Skin, Boston: Houghton Mifflin Co.

Jonaitis, Aldona (1978), "Reconciliation of Complementary Opposites: The Yakut Shaman Costume," Anthropology, 2, 61-66.

Kleinke, Chris L. (1975), First Impressions: The Psychology of Encountering Others, Englewood Cliffs, NJ: Prentice-Hall.

Knapp, Mark L. (1978), Nonverbal Communication in Human Interaction, New York: Holt, Rinehart, and Winston.

Lasswell, Thomas E. and Parshall, Peter F. (1961), "The Perception of Social Class from Photographs," Sociology and Social Research, 45, 407-414.

Laver, James (1969), Modesty in Dress, Boston: Houghton Mifflin Co.

Lefkowitz, Monroe, Blake, Robert R., and Mouton, Jane Srygley (1955), "Status Factors in Pedestrian Violation of Traffic Signals," Journal of Abnormal and Social Psychology, 51, 704-706.

Levy, Sidney J. (1959), "Symbols for Sale," Harvard Business Review, 37, 117-124.

Robertson, Thomas S. (1970), Consumer Behavior, Glenview, IL: Scott-Foresman and Co.

Rosenfeld, Lawrence B. and Plax, Timothy G. (1977), "Clothing as Communication," Journal of Communication, 27, 24-31.

Scheflen, Albert E. (1966), "Natural History Method in Psychotherapy: Communicational Research," in Methods of Research in Psychotherapy, eds. L. A. Gottschalk and A. A. Auerbach, New York: Appleton-Century-Crofts.

Scheflen, Albert E. (1974), How Behavior Means, Garden City, NY: Anchor Press.

Suedfeld, Peter, Bochner, Stephen, and Matas, Carol (1971), "Petitioner's Attire and Petition Signing by Peace Demonstrators: A Field Experiment," Journal of Applied Social Psychology, 1, 278-283.

Veblen, Thorstein (1953), The Theory of the Leisure Class, New York: The New American Library.

Walster, Elaine, Aronson, Vera, and Abrahams, Darcy (1966), "Importance of Physical Attractiveness in Dating Behavior," Journal of Personality and Social Psychology, 4, 508-516.

Winer, B. J. (1962), Statistical Principles in Experimental Design, New York: McGraw-Hill.

----------------------------------------