Backwards Segmentation Using Hierarchical Clustering and Q Factor Analysis

Citation:

Flemming Hansen and T. Bak-Jensen A/S/A.I.M. (1972) ,"Backwards Segmentation Using Hierarchical Clustering and Q Factor Analysis", in SV - Proceedings of the Third Annual Conference of the Association for Consumer Research, eds. M. Venkatesan, Chicago, IL : Association for Consumer Research, Pages: 220-239.

Proceedings of the Third Annual Conference of the Association for Consumer Research, 1972      Pages 220-239

BACKWARDS SEGMENTATION USING HIERARCHICAL CLUSTERING AND Q FACTOR ANALYSIS

Flemming Hansen, Marketing Director

T. Bak-Jensen A/S/A.I.M., Copenhagen

[T. Bak-Jensen A/S/A.I.M.-Marketing is a subsidiary of Booz, Allen and Hamilton Inc.]

THE PROBLEM

The idea behind market segmentation is simple. If you can divide a larger market into smaller segments with different preferences and subsequently adjust your product to the preferences in the different segments, then you reduce the overall distance between what you are offering to the market and what the market requires. By doing so the marketer improves his competitive position.

For market segmentation to be successful, however, two requirements must be met:

1. It must be possible to identify market segments with different behavioral and/or preferential patterns.

2. It must be possible to reach the different segments with differential media, distribution, product, communication message or other strategies.

Although the extent to which the second condition is met may vary from situation to situation, some possibilities will almost always exist when first the segments are identified. Consequently in the following the concern is primarily with the question which has become crucial to most marketers, namely how can different segments be identified?

The question as to the extent to which useful segments can be constructed can be divided into two. First, it must be possible to identify the different segments, secondly--for a differentiated marketing effort to be worthwhile-the segments must respond differently, that is the segments must be characterized by different demand elasticities. The latter, however, can be expected to be the case when segments can be identified which behave differently or which differ in the consumer values governing the behavior.

Therefore it is a major problem what kind of variables to use in an attempt to segment consumers. It is extremely important that the variables chosen really do relate to the behavior of the consumers. In the following pages some of the most commonly used variables are reviewed and following this two studies using different types of variables are reported.

SEGMENTATION VARIABLES

It may be possible to explain consumer behavior by means of personality variables. In the following, the application of personality tests in studies of consumer behavior will be reviewed. The discussion covers studies where larger tests have been used as well as studies where single personality dimensions are explored.

Systematic differences in consumer behavior may be studied based upon socioeconomic and demographic variables also. This will be explored, and in this connection consumer types such as innovators and opinion leaders are discussed.

It is possible that systematic differences in the consumer's self-image may explain differences in consumer behavior. Therefore, studies dealing with the relationship between self-image and product images (image congruence) will be reviewed also.

It may be possible to infer consumer values from observation of behavioral differences. That is, based upon systematic differences in consumer behavior, it may be possible to segment consumers meaningfully. Reports from studies which have applied this kind of "backward segmentation" are reviewed also.

Finally, it may be possible to learn about consumer values simply by asking about the interests, likes and values of the consumer. Also such studies are discussed.

1. The Use of Personality Inventories

In studies of consumer behavior, the most commonly applied personality inventory is "Edwards' (1954) Personal Preference Schedule" (EPPS). This inventory, which builds upon Murray's (1938) study of personality and motivation, is a forced choice, paper and pencil test classifying individuals along 15 personality dimensions. The classical study is reported by Evans (1959) and a replication is reported in Evans (1968). Evans administered selected items from the EPPS to Ford and Chevrolet owners and attempted to predict brand choices based on a discriminant analysis where the personality scores were used as independent variables. With this it was possible to classify 62.9 percent of the owners correctly; a prediction which was only slightly improved when 8 demographic variables were added to the personality variables.

Other authors (Kuehn 1963 & Marcus 1965) have re-analyzed the data and improved the predictions only slightly and other applications of EPPS are reported by Koponen (1960), and by "The Advertising Research Foundation (1964)". These studies deal with grocery products and attempt to predict brand choice, brand loyalty, quantity purchased, and other aspects of consumer choices. Neither of the studies have been very successful. In line with this are the findings reported by Massy, Frank and Lodahl (1968), who tried to explain differences in a number of consumer purchase variables.

Among all the attempts with the EPPS only Claycamp (1965) reports moderately successful results from a study of thrift deposit owners in commercial banks and in saving and loan associations.

Many other personality inventories have been applied. Westfall (1962) was moderately successful with "Thurstone's Temperament Schedule" (Thurstone, 1953) in an automobile study. Similarly, Kernan's (1968) application of the "Gordon Personal Profile" has not produced convincing results, and with the same test Tucker and Painter (1961) found only small correlations between personality types use of studied products (such as headache remedies, vitamins, mouthwash, cigarettes, chewing gum, alcoholic drinks, fashions, deodorants and automobiles). Other studies with only modest success are repeated by Wicks and Nelson (1967), who used the "Guilford-Zimmerman Temperament Survey"; and by Ruch (1966), who analyzed heavy and light users, and loyal and non-loyal users of grocery products, with "McCloskey Personality Inventory". Similarly Myers (1967) found that consumers' attitudes towards private brands are not significantly related to their score on "Cattell's 16 Personality Factor Inventory", and Brim, et al. (1962) were unable to explain differences in choice process behavior with personality scores obtained with "Taylor's Manifest Anxiety Scale ". In line with the results are also those reported by Robertson and Myers (1969) who found no correlations between, on the one hand measures of opinion leadership and innovative buying behavior and on the other hand personality scores obtained with the "California Psychological Inventory". Similarly, Rizzo and Naylor (1964) who applied Allport et al.'s (1960) "Study of Values" report negative results, and so does Scott &957) who applied the 'Minnesota Multiphasic Personality Inventory" in an attempt to predict motion picture preferences.

Altogether, personality inventories have not proven very useful in studies of consumer choice behavior: Wells (1966) summarizes the results nicely: "The findings of these studies have been very consistent. Almost always they have resulted in statistically significant correlations that have been too small to be of much practical value." (p. 187). It would be tempting to conclude that systematical differences in the values which consumers hold do not relate to the choices they make. However, before this view can be adapted, two alternative explanations must be rejected. First, personality tests are highly complex and difficult tests to construct, and those presently available may not be ideal. Secondly, the tests which have been applied have been developed for clinical and other uses, and it is questionable whether they can be expected to explain differences in a completely unrelated area such as consumer choice behavior. To the present author, the last explanation seems the most plausible. As Steiner (1966) suggests "I do not blame psychologists for the failure reported here in attempts to explain behavior by Using certain tests of personality--you cannot take just any tool off the shelf simply because it happens to be there and expect that it will be the best tool for your job."(p. 208).

2. Individual Personality Traits

Some authors, instead of using general personality inventories, have tried to relate consumer behavior to individual personality traits. Some of the more promising attempts have dealt with inner-other directedness, self-esteem and self-confidence, propensity to take risk, achievement motivation, and measures of cognitive style.

Inner-other directedness reflect the individual's tendency to rely upon others (other directedness) in decision making and evaluation of information, as opposed to a tendency to rely upon own judgements and values (inner directedness). The dimension relates to Riesman's (1961) distinction among traditional-, inner- and other-directedness, and it has normally been measured by means of a 36 item social preference scale constructed by W. Kassarjian (1962). With this H. Kassarjian (1965) found inner- and other-directedness to be related to consumers' preferences for persuasive communication and similarly, Arndt (1968) and Donally (1970) found it to relate to innovativeness.

"Self-esteem" or "generalized self-confidence" (Cox and Bauer, 1964) is a personality dimension which reflects the individual's feelings of social adequacy and of confidence in their own ability to cope with problems and aggressions. [The variable should not be confused with 'specific self-confidence' which reflects the individual's confidence in his ability to cope with a specific problem with which he is faced (for a discussion see Cox and Bauer, 1964).] Generalized self-confidence has frequently been measured with some or all of 23 items constructed by Janis and Field in Hovland and Janis (1959). Generally it has been related to individuals' susceptibility to persuasive communication.

However, Schuchman and Perry (1969) question the validity of several of the findings and only slight support is presented in the findings reported by Arndt (1967 and 1968). And, also, Ostlund (1969) was unable to establish any significant relation between generalized self-confidence and innovativeness.

Individual differences in willingness to accept risk are reported by Brim and Hoff (1957) and by Cunningham (1967b). The latter found that subjects who tend to perceive high risk in one product area also tend to perceive high risk in other product areas, and reverse, subjects who perceive little risk in one product area also are more likely to perceive little risk in other product areas.

Several studies confirm the importance of perceived risk as a personality variable. Arndt (1967 and 1968) reports that low risk perceivers are more likely to be innovative, and Cunningham (1964) reports that the extent to which consumers engage in personal communication in connection with grocery products is related to their tendency to perceive risk. Furthermore a number of studies have indicated relationships between loyalty and perceived risk (Cunningham, 1967a; Arndt, 1967; and Brody & Cunningham, 1968), and between innovativeness and a propensity to perceive risk (Arndt, 1968 and Ostlund, 1969).

Achievement motivation has been measured in several different ways. Normally, however, it is inferred from projective measures, most commonly the Thematic Apperception Test (TAT).

It is a basic proposition in the theory of the achievement motive that economic activity is related to achievement. In line with this, Morgan (1966) reports significant differences in income and in spending behavior depending upon achievement motivation. Similarly Boulding (1960) relates differences in consumer behavior to two personality types characterized as "integrated achievers" and "satisfied securers". Moreover, a number of studies have found need achievement to be related to risk taking propensities. For example, Scodel, et al., (1959) conducted a study where they applied a large number of different personality variables, but found only achievement motivation to be significantly related to risk-taking.

A major problem with this research has been that the different need achievement measures do not correlate well, and that no attempts have been made to relate achievement motivation to consumer choices. However, if the measurements problems can be overcome this variable may prove to be of some use in studies of consumer behavior.

Cognitive style reflects the way in which the individual approaches problems: whether they apply more or less wide categories, whether they place major emphasis on problem solving goals or upon social goals, and whether they strive for cognitive clarity or simplicity.

A couple of studies have dealt with these variables. Popielarz (1967) reports that consumers who use wide conceptual categories are more willing to accept new brands, and Phares and Davis (1966) found them to make larger adjustments in expectations following disconfirming experiences. Similarly, Cox (1967b) found that subjects with a high need for cognitive clarity are more susceptible to persuasive influence, and in the same study it also appeared that subjects who could be characterized as clarifiers (those who tend to clarify an issue) as opposed to simplifiers (those who tend to simplify an issue) respond differently to persuasive communication. Finally, Wilding and Bauer (1968) found subjects with predominantly social goals to react significantly different to communication as compared with subjects with predominantly problem solving goals.

Taken together the findings suggest that cognitive style may be an important variable, but so far the interrelations among the different measures is completely unexplored; and not until we have a better understanding of the nature of cognitive styles can more general hypotheses be formulated

In contrast with the traits measured in standardized personality inventories, most of the variables discussed here have emerged in studies of consumer and similar behavior. As shown, there is some evidence in favor of such variables. However, many questions remain to be answered, and the findings which have been reported do not suggest simple relationships between personality variables of this kind and the consumers' choices ant values

3. Socioeconomic and Demographic Variables

The previous discussion suggests that so far personality variables have not proven to be highly useful for the purpose of market segmentation. This would be less important if segmentation could be accomplished with socioeconomic and demographic variables; and of course, to some extent, these variables are useful. For example, it is obvious that homeowners are more likely to purchase outdoors paint and that households with babies are more likely to purchase baby food, etc. That is, socioeconomic and demographic variables can be used to define that segment of the total population which can possibly demand a certain product. However, finer discriminations can rarely be made with these variables. Even though some marked differences may exist among the users of different brands, and retail stores may attract different consumers, it has often been found that the ability of socioeconomic and demographic criterions to discriminate among consumers is relatively limited. As mentioned earlier, Evans (1959) only slightly improved the discrimitive power of his equation which he used to predict ownership of Chevrolet and Ford when he introduced demographic variables, and Frank et al. (1967) found only slight socioeconomic and demographic differences among consumers who purchased more or less expensive grocery items. Similarly, Frank and Boyd (1965) and Myers (1967) found no differences among consumers who prefer private brands as comPared with those who prefer manufacturers' brands, and Kuehn (1966), Frank (1967a), and Massy, et al. (1968) report that loyal consumers cannot be identified by means of demographic and socioeconomic characteristics. Finally, Frank (1967b) reports that heavy versus light buyers of a grocery product do not have different socioeconomic characteristics. Altogether, the available evidence warrants the conclusion by Frank (1968): "For the most part, socioeconomic characteristics are not particularly effective bases for segmentation" (p. 53).

4. Social Class

An individual's social class reflects the way in which he is perceived by others in the society. Warner and Lunt (1941) suggest that a social class consist "of people who are believed to be, and are accordingly ranked by the members of the community, in socially superior and inferior positions" (p. 82).

Several studies have attempted to relate so d al class to consumer behavior. Graham (1956) found different adoption patterns in different social classes for products such as television, canasta, super-markets, and medical services. Similarly, Martineau (1958) reports that many aspects of spending behavior and of store choices are related to social class. However, Brim, et al. (1962) found social class to have only little influence upon decision process variables and with regard to consumer behavior Rotzoll (1967) suggests that finer distinctions among social classes are of doubtful value. It is rarely possible to identify more than two separate classes. Also in the early sixties Martineau (1963) found many social class differences in shopping behavior to be disappearing and Rich and Jain (1968) after reviewing the literature, suggest that with the rapid changes which occur in income, leisure time, education ant the movement to suburbia, social class differences which may have existed earlier, are likely to disappear in the future.

5. Family Life Cycle

Consumers in different stages of the family life cycle are expected to behave differently, and possibly they have differing value structures. Often significant differences in spending behavior, savings, and possession of different durable products have been reported. However, no studies have found important variations in brand choices or in connection with purchases of nondurable products.

As with social class, Rich and Jain (1968) suggest that many differences traditionally associated with the family life-cycle tend to be ruled out by other changing factors in contemporary societies.

6. Innovators and Early Adopters

Much effort has been expended in attempts to identify consumers who are likely to accept new products early. To the extent that such consumers can be defined as a special market segment, they represent a group of consumers of particular importance.

Two questions must be raised in connection with innovators as a special market segment. First, do those who adopt;a particular product early differ from those who adopt it later? Secondly, are those consumers who adopt early in one product area also likely to be innovators for other products?

There is considerable evidence showing that early adopters differ from late adopters. Findings from different areas of research are reviewed by Rogers and Stanfield (1968); and several studies of marketing innovators have found similar differences. (Robertson, 1971). However, most studies report weak relationships and on the whole, it seems that no variables apply uniformly to all products.

This observation suggests that innovativeness in one area does not automatically imply innovativeness in other areas, a conclusion in line with the results from the few studies which have directly explored the amount of overlap between innovativeness in different areas (Warneryd, 1965; Robertson & Myers, 1969; and Arndt, 1968a). They all find practically no overlap among innovativeness in different areas. Therefore, there is little support for innovators as a special market segment.

7. Opinion Leaders

It is a common assumption in communication research that some people act primarily as "opinion leaders" and others primarily as receivers ("followers"). As with innovators the influentials might constitute a special market segment composed of consumers with special values and perceptions.

Findings suggest, however, that opinion leadership does not correlate closely with socioeconomic and demographic variables. Illustrative findings are reported by Myers and Robertson (1969). In an extensive study of 12 products, they found only small correlations with demographic variables and no single variable was significant for all products, and with regard to personality variables even fewer relationships have been found. (King and Summers, 1969; and Myers and Robertson, 1969).

Several researchers have studied the amount of overlap among opinion leaders in different product areas and found only little overlap between unrelated products and only modest overlap among related products.

The concept of opinion leaders rests upon the assumption that some consumers primarily act as sources of personal communication whereas others primarily are receivers. However, few studies have been concerned with the extent to which the opinion leadership studied results in communication between "leaders" and "followers," and recent studies have suggested that people who predominantly act as receivers ("the followers") are rare (Warneryd, 1965; Cerha, 1967; and King & Summers, 1969).

These authors suggest that in most product areas from 60% to 80% of all consumers can be characterized either as both frequent receivers and initiators of communication or as infrequent receivers and initiators of communication. When this is taken together with some of the more consistent characteristics of opinion leaders; their interest in innovations, their better knowledge, and their more frequent exposure to mass communication, it appears that rather than distinguishing among opinion leaders and followers, one should distinguish among consumers engaging in more or less personal communications about the product. Basically those who are interested in a given product also are those who talk about it, and in the process of doing this they provide information for others as well as they acquire additional information for themselves.

8. Purchase Characteristics

Several authors have studied relationships among different aspects of the same consumer's behavior. For example, there is some evidence that brand loyalty is positively related to the market share of the brand (Schuchman, 1968), that "deal-proneness" is related to the number of different brands purchased; to the number of units purchased; and to the brand loyalty (Webster, 1965). Similarly Kollat and Willet (1967) found impouse purchases to be related to number of items purchased to the number of members in the shopping party, and to variables reflecting the structure of the transaction (major or minor purchasing trip); Frank et al. (1964) showed relationships between innovative behavior and purchase characteristics; and Rao (1969b) found brand loyalty and private brand Proneness to be related to store loyalty.

Relationships of this kind may be useful in some attempts to identify special market segments, but it is usually a problem that consumers with special purchase patterns can rarely be identified in other ways than through their purchase behavior. For example, it is of limited use to know that heavy users tend to be more loyal than light users if neither loyal nor heavy users can be identified. But it is a common observation that neither loyal, deal-prone, private brand loyal, heavy users nor innovative consumers, etc. are easily identified. (Frank. 1968).

9. Image Congruence Theories

Consumers' perceptions of purchase alternatives are reflected in the cognitive relationships between on the one hand brands, products, stores, etc. and on the other hand aroused values. Similarly, the consumer's perception of himself can be described in terms of the perceived relationships between the concept of the self and valued concepts. Several authors have suggested that the consumer selects products which are perceived as congruent with his self-image. To the extent that such a relationship can be proven, the self-image would be a valuable set of variables to work with in segmentation studies.

A number of studies have attempted to validate the image congruence hypotheses. Most of these have tested one or both of the following two propositions: (1) there are significant differences in the way in which products are perceived, (2) those products which the consumer owns or prefers have images which deviate less from his self-image than the images of the products which he does not own or does not prefer. Several studies have dealt with automobile brands and supportive evidence is reported by Jacobsen and Kossoff (1963), Birdwell (1964), Grubb and Hupp (1968), and Ito (1967).

To test whether knowledge of the consumers' perception of himself together with information about his images of brands makes it possible to predict his choices, both images should be measured before the choice. This has never been tried. What comes closest is the study reported by Ito (1967). In a nationwide probability sample of car owners, 577 Ford and Chevy owners who were planning to purchase a new Ford or Chevy were identified. Based upon measures of self and product images, it was possible to predict from 51 to 66 percent of the purchase intentions correctly. These percentages are not very high, but of those who intended to switch brands,from 82 to 96 percent were classified correctly.

Another attempt to prove the significance of the self-image rests upon the following reasoning: Since it is not likely that the consumer will change his perception of himself following purchase decisions, significant differences in the self-images of consumers who have chosen different brands can be expected also to have existed before the brand was chosen. Such differences have been identified by Grubb and Supp (1968) in a study where they compared Pontiac owners and Volkswagen owners. They found that Pontiac owners rated themselves significantly higher on dimensions which were positively associated with the Pontiac, whereas the Volkswagen owners rated themselves significantly higher on dimensions which were positively associated with the Volkswagen.

Altogether some positive evidence has been reported supporting the existence of a relationship between self-images and images of brands and products purchase, and it is possible that improved measurement techniques may strengthen this further.

10. Backwards Segmentation

Based upon information about consumers consumption and purchase patterns, and upon product perceptions, it has been tried to classify consumers. Large scale factor-analysis has made this possible and the approach is in line with the conclusions from several of the previous sections: attempts with personality inventories have suggested that consumer types should be identified starting in analysis of the behavior of consumers; attempts to make predictions based upon socio-economic ant demographic studies of different purchase and consumption variables have shown that many of these are correlated and finally the proposal that significant aspects of consumer's conceptual structures may be reflected in their overall life-styles suggests that more systematic utilization of information about different aspects of consumer behavior may lead to meaningful classifications.

Several studies have been reported. Wilson (1966) used factor analysis to identify 20 different variables reflecting aspects of the respondents' product perceptions and similar findings are reported by Pessemier and Tigert (1966). In this research 14 interest and 8 personality factors were identified, many of which closely resemble those identified in the Wilson study. Later the same authors (Bass, Pessemier & Tigert, 1969; and also Welb, 1968) have worked with purchase data alone. Here again product oriented factors emerged closely resembling those reported in other studies.

All of these studies suggest that it is possible to characterize consumers meaningfully based upon information about their perceptions and consumption and purchase patterns. In spite of considerable differences in the samples and in the type of data which have been used, many almost identical factors have been identified.

11. Interests and Values

In several ways the research reviewed in the previous pages has pointed at the importance of interests and values factors governing consumer behavior. It was suggested that innovativeness may be seen in relation to interests and that the frequency with which personal and other information sources are attended to, vary with interest in the issue. Finally, image congruence studies and backward segmentation has pointed at factors of the value interest type.

It is not surprising that interest shows up as an important variable when it is realized that interest in an issue implies that the consumer has a number of important and positive values in relation to the topic. It is more surprising that very little research has been concerned directly with this variable.

Only few studies have been reported.

Cerha (1967) obtained interest scores for 91 products on simple seven point scales and found considerable variations among products, as well as close relationships between interest and exposure to mass and personal communication Moreover, he found relatively small intercorrelations among the different product areas, and based upon factor analysis, he was able to identify seven highly independent interest areas. In another study Pennington and Peterson (1969) used the "Strong Vocational Interest Blank" (Cambell, 1966). Being constructed primarily for purposes of counselling and personal selection, this interest questionnaire would be expected to have the same shortcomings as general personality inventories. Nevertheless, based on the most productive items in the test. and using discriminant analysis, the authors were able to make correct predictions of choices among vacation trips and savings forms in 72 to 80% of the cases. Unfortunately, the products chosen were rather special, but further research with this or similar interest test batteries seems promising. Particularly if tests are developed which are especially relevant to consumer behavior, our ability to predict consumer choices may improve significantly.

SEGMENTATION VARIABLES: SUMMARY

The preceding review suggests that the variables most commonly used in segmentation studies can be grouped along two dimensions. On the one hand, one can distinguish among behavioral and psychological variables. On the other hand it is possible to see the variables as being more or less specific to the ultimate purpose with segmentation studies: to divide consumers into groups behaving differently as consumers. This two way classification is shown in Figure 1.

FIGURE 1

POSSIBLE SEGMENTATION VARIABLES: AN OVERVIEW

With regard to the more psychologically oriented variables, the precedent review suggests that the more product specific variables are the most promising.

Among the behavioral variables, the more specific ones have in a number of contexts proven useful. However, practical applications of these variables often raise problems. It is normally difficult to identify marketing strategies aimed at market segments defined in terms of the behavioral variables to be influenced by the marketing strategies chosen. Normally, the researcher tries to define the segments identified in this way in terms of some attitude or interest variables which can be deducted from the behavioral variables grouped together.

For this reason in a number of segmentation studies we have chosen to work with different kinds of product related interest, attitude and perceptual variables. Below some of our experiences are reviewed.

Backwards Segmentation With Interest, Attitude And Perceptual Variables

Backwards segmentation was described above as segmentation based upon those behavioral interest and attitude variables in which one is ultimately interested. In the simple case with only one variable involved, backwards segmentation is trivial. For example, dividing consumers into heavy users and light users of a product based upon a single measure of the amount used of the product, is a simple form of backwards segmentation. However, what is most commonly discussed as backwards segmentation is the case where a number of different variables form the basis for the segmentation, and here the interrelationships among the variables become critical. As an operational definition of backwards segmentation the following is proposed: Backwards segmentation is the process in which segments are established departing from those variables which are dependent in relation to the project.

A Segmentation of TV-Viewers

In this study the major emphasis is on a comparison between the use of two different techniques with the same set of dates.

In one of a regularly mode TV audience surveys, carried out for the Danish Broadcasting Corporation, special interest was paid to the news programs. Criticism had been raised about the TV-news coverage. Particularly the reporters were said to be difficult to understand, using an academic language and being politically biased. To examine the extent to which this criticism existed among the viewers also, a number of questions about TV's news coverage were included in the spring survey of 1971. The questions were concerned with how viewers regard the coverage of local news, foreign news, objectivity, actuality, the use of filler items, understandability, use of foreign words, unclear speaking, etc. With these items no real support for the criticism was found, neither in the total population nor in any geographical, socioeconomic or other segments. For this reason it was decided to use a backwards segmentation technique in an attempt to test if segments existed in which the criticism could be identified.

To identify the dimensions along which the 1500 respondents evaluate the newscasts, a normal (R) factor analysis was carried out. In this way three major dimensions were localized. Briefly described they are:

I Quality of information

II Understandability

III Type of information

The main results of this analysis are presented in Table 1.

TABLE 1

Subsequently a Q-factor analysis was carried out in an attempt to identify segments having special viewers along the three major dimensions. Briefly the nature of the Q-factor analysis can be understood as follows. Whereas the R analysis departs in a matrix of correlations like the one shown in Figure 2a containing correlations between responses computed over respondents, the Q analysis works with a matrix like the one shown in Figure 2b. Here the correlations are between respondents and computed so that two respondents with the same responses will have a correlation of 1.00 and two respondents with very different responses will have a low correlation.

FIGURE 2

ORIGINAL AND REDUCED CORRELATION MATRICES IN R AND Q FACTOR ANALYSIS

TABLE 2

PROFILE FOR SAMPLE AND SEGMENTS. A HIGH SCORE MEANS THAT RESPONDENTS TRIED TO OPEN WITH THE STATEMENTS (6 POINT LIKERT SCALES ARE USED)

The computational procedure, however, is the same in Q and R analysis.

In this way 9 segments were identified (Table 2). Of those, however, the 5 largest contained 86% of the respondents and the results are shown in Table 2. Among the segments,the first (including a little less than 50% of all respondents) can be described as satisfied viewers. They are more positive than the average on all dimensions. The remaining four groups are characterized by their negative evaluation along one or more of the dimensions. Group two is claiming that rather than giving a broad coverage of local and international news,biased news are broadcasted. The group constitutes 15% of all respondents. Group 3 is also dissatisfied with the news coverage, but rather than seeing the material broadcasted as biased,they perceive it as "filler items"This segment constitutes 13% of the respondents. The 2nd last segment does not analyze the news coverage as such, but it is claimed that it is biased and less relevant. It is noteworthy that this group to a significant extent says that only few foreign words are used, suggesting that it is' a more educated group. Finally the last group (4) says that many foreign words are used, that it is not easy to understand and that it is biased. Seemingly this segment constitutes a group which has difficulties in understanding what is being broadcasted.

It should be noted also, that the segments differing significantly in their view on Radio and TV-news are quite similar in their socio-economic characteristics. The only significant difference lies in the first segment which tends to consist of more males in the Copenhagen area.

In an attempt to test the validity of the grouping mode with the Q analysis a sample of 150 respondents was reanalyzed using a different clustering procedure. In this case comparisons among respondents was based upon rank-order correlations between respondents,and the matrix of similarities constructed in this way was then used as input for The Johnson's hierarchical clustering procedure. In this way 5 clusters plus 16 unclassifiable respondents were identified. Particularly the 5 clusters agree very closely with the 5 major segments identified in the Q analysis. There were only few and insignificant differences in the average scores and the overlap between groups runs as high as 60-90%. That is, of the respondents assigned to one segment by the hierarchical clustering procedure from 60 to 90% were assigned to the corresponding group in the Q analysis.

A ComParison of Segments Established Based Upon Different Types of Data

To understand the segmentation problem encountered in this study it is necessary to say a few words about some of the results from another part of the study.

The second study to be reported deals with educational toys. In a given market,demand for the product is highly dependent upon two things. First, the number of consumers adopting the product and secondly the frequency with which those who adopt continues to purchase it. The product sales differ significantly between England and Germany. To examine the extent to which this was ascribable to differences in the market structure a comparative study was designed. In this study measurements were obtained on 7 point likert scale. The questionnaire included 35 general attitude statements relating to child rearing, purchases of toys, etc. Moreover, 43 specific image variables revealing the perception of the product were included. With these sets of variables using R-factor analysis it was found that the structure in the general attitudes as well as in the specific images were quite similar in the two countries.

TABLE 3

PERCENTAGE OF RESPONDENTS FROM SEGMENTS ESTABLISHED BASED UPON GENERAL ATTITUDES FALLING IN RESPONDENTS ESTABLISHED BASED UPON SPECIFIC IMAGE VARIABLES.

Moreover, in the segments subsequently identified separately in England and Germany the same conceptual structures applied.

Actually the factors were so alike that it was a natural way of carrying out the comparisons between segments to run a joint factor analysis on the combined data from the two countries, resulting in one common image structure (14 factors) and one common general attitude structure (12 factors). Further comparisons could be made with the use of these common factors.

With regard to the general attitudes compared with the specific image it was found that significant differences were found between the two countries in 6 and 8 factors. However, when the variables were related to the amount of the toy in the household (by means of stepwise regression analysis) and to propensities to buy more of the toy, only the specific image variables had any importance. Opposite, when discriminant analysis was used to analyze ownership versus no ownership,the most important variables turned out to be general attitude ones. Seemingly, are the two sets of variables playing a different rate and the question relating to further segmentation was, what set of variables to use.

With the results from the previous study in mind (and those from a similar one on automobiles) it was decided to use only Q-factor analysis, and this was carried out on both sets of data separately. Here only results from one of the countries are shown. With the image data 5 segments were identified and with the general attitude data 6 segments resulted. How the segments compare can be seen from Table 3, where it appears that even though some agreement exists (the zero-hypotheses that the two groups of segments are unrelated is rejected with p > 0.999 _ x2 test) there is no close correspondence between the segments. Closest come the two segments, number two having approximately 50% of the respondents in common. In further analysis both sets of the segments were studied in relation to amount of the toy at home and propensity to purchase. With regard to the first variable only small differences were found among the segments. This applies to both types of segments. With regard to propensities to buy, however, there were large and significant differences between the segments derived from the image data. This is shown in Table 4 where the four most important segments are shown.

The first segment can be described as overall possible, the second as overall negative and the third and the fourth as partially negative. Finally, it should be noted that there are no significant differences between the segment on socio-economic criterion.

The implications for the company of these findings shall not be discussed here, but some conclusions concerning the technique of backwards segmentation should be provided.

TABLE 4

FOUR MAJOR SEGMENTS AND THEIR PERCEPTION OF THE PRODUCT

CONCLUSIONS

Two basic questions facing the researcher when dealing with backwards segmentation relate to the choice of the data handling technique and the choice of variables to work with. The findings reported here suggest that the latter question is far more crucial than the first. Whereas choice of technique to a large extent may be a question of the type of data available (metric/nonmetric, etc.) the choice of variables to work with turns out to be highly critical. For that reason it may be advisable to put more time into the latter question.

Another conclusion established in the studies reported here is that backwards segmentation is useful in the sense that it makes it possible to identify segments, which differ in relation to the product studied. Moreover in the two studies reported here (and a couple of others not described) it seems that the segments which emerge normally will consist of one totally positive and a number of segments being negative for different reasons.

Finally, a somewhat negative--but not unexpected--conclusion can be mentioned. Seemingly the segments thus identified do not differ largely in terms of more traditional variables such as income, age, sex. etc.

REFERENCES

Advertising Research Foundation. Are There Consumer Types? New York: Advertising Research Foundation, 1964.

Aliport, G. W., Vernong, P. E., Lindzey, G. Study of Values. Boston, Mass.: Houghton-Mifflin Company, 1960.

Arndt, J. A Study of Work of Mouth Advertising. Markedskommunikation. 1967, 4, 94-117.

Arndt, J. New Product Diffusion: The Interplay of Innovativeness, Opinion Leadership, Learning, Perceived Risk and Product Characteristics. Markedskommunikation, 1968a, 5, pp. 1-9.

Arndt, J. Profiling Consumer Innovators. In J. Arndt (Ed.), Insights Into Consumer Behavior. Boston: Allyn and Bacon, 1968b.

Birdwell, E. A. Influence of Image Congruence on Consumer Choice. L. G. Smith (ed.): Reflections on Progress in Marketing. Chicago: American Marketing Association, 1964.

Boulding, E. Orientation Toward Achievement or Security in Relation to Consumer Behavior. Human Relations, 1960, 13, 365-82.

Brim, O. G.; Glass, D. G.; Lavin, D. E., & Goodman, N. Personality and Decision Processes. Stanford, Calif.: Stanford University Press, 1962.

Brim, O. G.; Hoff, D. B. Individual and Situational Differences in Desire for Certainty. Journal of Abnormal and Social Psychology, 1957, 54, 225-29.

Brody, R. P.; Cunningham, S. M. Personality Variables and the Consumer Decision Process. Journal of Marketing Research, 1968, 5, 50-57.

Cambell, D. P. Manual for the Strong Vocation Interest Blank. Palo Alto, Calif.: Stanford University Press, 1966.

Cerha, J. Selective Mass Communication. Stockholm: P. A. Norstedt and Soner, 1967.

Claycamp, J. H. Characteristics of Thrift Deposit Owners. Journal of Marketing Research, 1965, 2, 163-70.

Cox, D. F. The Influence of Cognitive Needs and Styles on Information Handling in Making Product Evaluations. In D. F. Cox (Ed.) Risk Taking and Information Handling in Consumer Behavior. Boston: Grad. School of Business Administration, Harvard University, 1967, pp. 370-73.

Cox, D. F.; Bauer, R. A. Self Confidence and Persuadability in Women. Public Opinion Quarterly, 1964, 28, 453-66.

Cunningham, S. M. Perceived Risk as a Factor in Product-Oriented Word-of-Mouth. Behavior: A First Step. In L. G. Smith (Ed.), Reflections on Progress in Marketing. Chicago: American Marketing Association, 1964.

Cunningham, S. M. Perceived Risk and Brand Loyalty. In D. F. Cox (Ed.), Risk Taking and Information Handling in Consumer Behavior. Boston: Graduate School of Business Administration, Harvard University Press, 1967a.

Cunningham, S. M. The Major Dimensions of Perceived Risk. In D. F. Cox (Ed.), Risk Taking and Information Handling in Consumer Behavior. Boston: Graduate School of Business Administration, Harvard University, 1967b.

Donnelly, Jr., J. H. Social Character and Acceptance of New Products. Journal of Marketing Research, 1970, 7, 111-13.

Edwards, A. L. Personal Preference Schedule: Manual. New York: The Psychological Corporation, 1954.

Evans, F. B. Psychological and Objective Factors in the Prediction of Brand Choice. Ford versus Chevrolet. Journal of Business, 1959, 32, 340-69.

Evans, F. B. Ford versus Chevrolet: Park Forest Revisited. Journal of Business, 1968, 41, 445-59.

Frank, R. E. Is Brand Loyalty a Useful Basis for Market Segmentation? Journal of Advertising Research, 1967a, 7, 27-33.

Frank, R. E. Correlates of Buying Behavior for Grocery Products. Journal of Marketing, 1967b, 31, 48-53.

Frank, R. E. Market Segmentation Research: Findings and Implications. In F. M. Bass, C. W. King and E. A. Pessemier (Eds.), Applications of the Sciences in Marketing Management. New York: John Wiley and Sons, Inc., 1968.

Frank, R. E. & Boyd, H. W. Are Private Brand Prone Grocery Customers Really Different? Journal of Advertising Research, 1965, 5, 27-35.

Frank, R. E.; Green, R. E.; Sieber, H. F. Household Correlates of Purchase Price for Grocery Products. Journal of Marketing Research, 1967, 4, 54-58.

Frank, R. E., Massy, W. F., & Morrison, D. G. The Determinants of Innovative Behavior with Respect to a Branded, Frequently Purchased Food Product. In G. Smith (Ed.), Reflections on Progress in Marketing. Chicago: American Marketing Association, 1964, 312-23.

Graham, S. Class and Conservation in the Adoption of Innovation. Human Relations, 1956, 9, 91-100.

Grubb, E. L., & Hupp, G. Perception of Self, Generalized Stereotypes and Brand Selection. Journal of Marketing Research, 1968, 5, 58-63.

Hovland, C. I., & Janis, I. L., (Eds.), Personality and Persuadability. New Haven: Yale University Press, 1959.

Ito, Pikuma. Differential Attitude of New Car Buyers. Journal of Advertising Research, 1967, 38-42.

Jacobsen, E., & Kossoff, J. Self Percept and Consumer Attitudes Toward Small Cars. Journal of Applied Psychology, 1963, 47, 242-45.

Kassarjian, H. H. Social Character and Differential Preference for Mass Communication. Journal of Marketing Research, 1965, 2, 146-53.

Kassarjian, W. M. A Study of Riesman's Theory of Social Character. Sociometry, 1962, 25, 213-30.

Kernan, J. B. Choice Criteria, Decision Behavior, and Personality. Journal of Marketing Research, 1968, 5, 155-64.

King, C. W., & Summers, J. O. Technology, Innovation and Consumer Decision Making. In R. Moyer (Ed.), Changing Marketing Systems. Chicago: American Marketing Association, 1967, 63-68.

Kollat, D. T., & Willett, R. T. Customer Impulse Purchase Behavior. Journal of Marketing Research. 1967, 4, 21-31.

Koponen, A. Personality Characteristics of Purchasers. Journal of American Research, 1960, 1, 6-12.

Kuehn, A. A. Demonstration of a Relationship Between Psychological Factors and Brand Choice e Journal of Business, 1963, 36, 237-41.

Kuehn, A. A. Mathematical Models of Consumer Behavior. In J. W. Newman (Ed.), On Knowing the Consumer. New York: John Wiley and Sons, Inc. 1966.

Marcus, A. S. Obtaining Group Measures from Personality Test Scores: Auto Brand Choice Predicted From the Edwards Personal Preference Schedule. Psychological Reports, 1965, 17, 523-31.

Martineau, P. D. Social Class and Spending Behavior. Journal of Marketing, 1958, 121-30.

Martineau, P. D. Customers Shopping Center Habits Change Retailing. Editor and Publisher, October 26, 1963, p. 11.

Massy, W. F., Frank, R. E., & Lodahl, T. Purchasing Behavior and Personal Attributes. Philadelphia: University of Pennsylvania Press, 1968.

Morgan, J. N. The Achievement Motive and Economic Behavior. In J. W. Atkinson and N. T. Feather (Eds.), A Theory of Achievement Motivation. New York: John Wiley and Sons, Inc. 1966.

Murray, H. A. Explorations in Personality. New York: Oxford University Press, 1938.

Myers, J. G. Determinants of Private Brand Attitudes. Journal of Marketing Research, 1967, 4, 73-81.

Ostlund, L. E. The Role of Product Perceptions in Innovative Behavior. Paper presented at Fall Conference of American Marketing Association, Cincinnati, Ohio, 1969.

Pennington, A. L., & Peterson, R. A. Interest Patterns and Product Preferences: An Exploratory Analysis. Journal of Marketing Research, 1969, 6, 284-90.

Phares, E. J., & Davis, W. L. Breadth of Categorization and the Generalization of Expectancies. Journal of Personality and Social Psychology, 1966, 4, 461-63.

Popielarz, B. T. An Exploration of Perceived Risk and Willingness to Try a New Product. Journal of Marketing Research, 1967, 4, 368-73.

Rao, T. R. Are Some Consumers More Prone to Purchase Private Brands? Journal of Marketing Research, 1969, 6, 447-50.

Rich, S. U., & Jain, S. L. Social Class and Life Cycle Predictions of Shopping Behavior. Journal of Marketing Research, 1968, 5, 41-49.

Riesman, D. The Lonely Crowd. New Haven: Yale University Press, 1964.

Rizzo, J. R., & Naylor, J. C. The Factorial Structure of Selected Consumer Choice Parameters and Their Relationship to Personal Values. Journal of Applied Psychology, 1964, 48, 241-48.

Robertson, T. R., & Myers, J. G. Personality Correlates of Opinion Leadership and Innovative Buying Behavior. Journal of Marketing Research, 1969, 6, 164-68.

Rogers, E. M., & Stanfiels, J. D. Adoption and Diffusion of New Products: Emerging Generalizations and Hypotheses. In F. M. Bass, E. A. Pessemier and C. W. King (Eds.), Applications of the Sciences in Marketing Management. New York: John Wiley and Sons, Inc., 1968.

Rotzoll, K. B. The Effect of Social Stratification on Market Behavior. Journal of Advertising Research, 1967, 7, 22-27.

Ruch, D. M. Limitations of Current Approaches to Understanding Brand Buying Behavior. In J. W. Newman (Ed.), On Knowing the Consumer. New York: John Wiley and Sons, Inc., 1966.

Schuchman, A. Are There Laws of Consumer Behavior? Journal of Advertising Research, 1968, 8, 19-28.

Schuchman, A., & Perry, M. Self Confidence and Persuadability in Marketing. Journal of Marketing Research, 1969, 6, 146-55.

Scodel, A., Ratoosh, P., & Minas, J. S. Some Personality Correlates of Decision Making Under Conditions of Risk. Behavioral Science, 1959, 4, 19-28.

Scott, E. M. Personality and Movie Preferences. Psychological Reports, 1957, 3, 17-18.

Steiner, G. A. Consumer Behavior: Where Do We Stand? A Psychologist's Appraisal. In J. W. Newman (Ed.), On Knowing the Consumer. New York: John Wiley and Sons, Inc. 1966.

Thurstone, L. L. Examiner Manual for the Thurstone Temperament Schedule. Chicago: Science Research Associates, 1953.

Tucker, W. T., & Painter, J. J. Personality and Product Use. Journal of Applied Psychology, 1961, 45, 325-29.

Warner, W. L., & Lunt, P. S. The Social Life of a Modern Community. New Haven: Yale University Press, Yankee City Series, Vol. 1, 1941.

Warneryd, B. Innovation, Indflydelse och Information (with an English summary). Stockholm, Sweden: Almquist och Wicksell, 1965.

Webster, F. E., Jr. The Deal-Prone Consumer. Journal of Marketing Research, 1965, 2, 186-89.

Wells, W. D. Children as Consumers. In J. W. Newman (Ed.), On Knowing the Consumer. New York: John Wiley and Sons, Inc., 1966.

Wells, W. D. Backward Segmentation. In J. Arndt (Ed.), Insights Into Consumer Behavior. Boston: Allyn and Bacon, 1968.

Westfall, R. L. Psychological Factors in Predicting Product Choice. Journal of Marketing, 1962, 26, 34-40.

Wicks, J. H., & Nelson, C. C. Preliminary Investigation of Some Psychological Determinants of Consumption Propensity. Southern Economic Journal, January, 1967, 383-87.

Wilding, T., & Bauer, R. A. Consumer Goals and Reaction to a Communication Source. Journal of Marketing Research, 1968, 5, 73-77.

Wilson, L. L. Homemaker Living Patterns and Marketplace Behavior--a Psychometric Approach. In J. S. Wright and J. L. Goldstucker (Eds.), New Ideas for Successful Marketing. Chicago: American Marketing Association, 1966.

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