Involvement, Familiarity, Cognitive Differentiation, and Advertising Recall: a Test of Convergent and Discriminant Validity

ABSTRACT - Involvement, familiarity, and cognitive differentiation are three measures of individual difference which have been hypothesized to be related to consumers' ability to recall advertising messages. Here, these three relationships are examined, and an attempt is made to establish a purified measurement procedure for operationalizing each of these constructs. With this last purpose in mind, tests of convergent and discriminant validity are reported; and a group of indicators is tentatively proposed for measuring each construct in our hypothesized model.

Citation:

George M. Zinkhan and Aydin Muderrisoglu (1985) ,"Involvement, Familiarity, Cognitive Differentiation, and Advertising Recall: a Test of Convergent and Discriminant Validity", in NA - Advances in Consumer Research Volume 12, eds. Elizabeth C. Hirschman and Moris B. Holbrook, Provo, UT : Association for Consumer Research, Pages: 356-361.

Advances in Consumer Research Volume 12, 1985      Pages 356-361

INVOLVEMENT, FAMILIARITY, COGNITIVE DIFFERENTIATION, AND ADVERTISING RECALL: A TEST OF CONVERGENT AND DISCRIMINANT VALIDITY

George M. Zinkhan, University of Houston

Aydin Muderrisoglu, University of Houston

ABSTRACT -

Involvement, familiarity, and cognitive differentiation are three measures of individual difference which have been hypothesized to be related to consumers' ability to recall advertising messages. Here, these three relationships are examined, and an attempt is made to establish a purified measurement procedure for operationalizing each of these constructs. With this last purpose in mind, tests of convergent and discriminant validity are reported; and a group of indicators is tentatively proposed for measuring each construct in our hypothesized model.

INTRODUCTION

Semantic confusion about the concept of "involvement" and involvement-related concepts has been going on in the consumer behavior literature for some time now. A lot has been written since Krugman's (1965) introduction of the low-involvement concept. One would expect that both empiricists and academicians interested in involvement would have a clear conceptualization of this construct. One would further expect an understanding of the relationships between "involvement" and other concepts used to explain it. While a great deal of interest exists, there is not a methodological consensus to study involvement. A number of different definitions and conceptualizations as well as distinct methodological approaches have been used. But these have led to further semantic confusion instead of contributing to a clear understanding of what involvement is, what it is not, what its causes and effects are. There are basically two reasons for this semantic confusion. First, the construct of involvement is extremely difficult to define and subsequently operationalize. This is especially so in light of the second reason which is the lack of a nomological network of relationships that help explain the construct of involvement. In 1955, Cronbach and Meehl stated that "to make clear what something is means to set forth the laws in which it occurs: a nomological network." To understand the relationship between involvement and other hypothetical constructs (i.e., to set forth laws in which involvement occurs), we need clearer conceptualizations of involvement and the constructs related to it. Thus, the purpose of this paper is to make an attempt to start building a nomological network of involvement. To better understand the nature of involvement and to begin developing the theoretical structure in which it is embedded, we generated testable propositions from four interrelated constructs. These constructs are involvement, familiarity, cognitive differentiation and their relationship to recall. Involvement and related concerts have most frequently been used in a communication/advertising context. Thus with the model in Figure 1, we intend to show how similar -- and at times, synonymously used -- concepts of involvement. familiarity and cognitive differentiation influence recall in a persuasive communication setting.

INVOLVEMENT

It is plausible to conceptualize that each one of the concepts has a positive direct effect on recall of advertising messages. That is, the higher a person's involvement and familiarity with a product and the higher the ability to cognitively differentiate between features of that product, the higher the recall of the contents/characteristics of an advertisement for that product.

Among the many definitions of involvement is one that views it as a psychological/internal state of commitment (Mitchell 1979, 1981) that is activated by a certain stimulus in a given situation (Cohen 1983). If activation of this internal state is potentially high, caused possibly by a greater degree of attention to that particular stimulus, then subsequent memory performance and recall should also be high.

This perspective is also analogous to Greenwald and Leavitt's (1984) description of involvement at four levels: preattention, focal attention, comprehension and elaboration. As antecedents to these four levels of involvement, a distinguishing characteristic observed by the person may be made salient in his/her mind. Stimuli will not be very salient at the preattention and focal attention level. From a processing point of view, stimuli may be processed at a shallow level, leading to a lower degree of commitment to the stimulus object in the cognitive domain. On the other hand, comprehension and elaboration require not only more cognitive capacity (Greenwald and Leavitt 1984), but will result in "deeper" processing leading to a higher activation potential for the stimulus and thus more commitment and involvement. High involvement due to comprehension and elaboration should also lead to better recall of message characteristics.

FAMILIARITY

Familiarity with the product and/or brand is an important factor in the study of consumer behavior. Familiarity is a variable that describes the nature of the cognitive structure a person has towards a product. As such it should be seen as an intervening variable between the hypothetical construct of involvement and, say, an importance scale at the measurement level. Familiarity has been operationalized in the past as frequency of use (Raju and Reilly 1979), knowledge about the product class (Lastovicka 1979) and previous experience (Russo and Johnson 1980). But Marks and Olson (1981) disagree with these definitions and suggest that familiarity is the "cognitive representation of the (past) experiences that are stored in memory... These representations can be considered to be organized in a memory as a product-related cognitive structure or schema." This information processing approach to product familiarity is appropriate in light of the above conceptualization of the involvement construct. When an individual is exposed to an external stimulus, this information is going to be encoded and represented either isomorphically or in a form different to its external existence. By encoding we mean an individual will create mental associations between the features of the stimuli and cues or pieces of knowledge that already exist in his/her memory. Reseated exposure-encoding-representations will lead the individual to form more elaborate and possibly more complex memory structures about the stimulus object. Thus it is these well- or less-developed memory structures that define the amount of familiarity an individual has with an object. If an individual receives a persuasive message about a product/brand for which the individual has a well-developed memory structure, then that individual will be able to activate more concepts from memory to use in interpreting the attended stimuli. This may also mean that the individual may present a higher activation potential (a la Cohen 1981) to process the external information. This high level of activation may explain the level of involvement of a person more familiar with that product and/or brand. Because this individual has more knowledge and concepts in memory to use in judging the external stimuli, the possibility of the features of stimuli being associated with already existing cues is larger. This will lead the individual to have better memory performances and subsequent recall. On the other hand, an individual who is not very familiar with a product will have less-developed memory structures. This will cause the individual to link the features of the external stimuli with a fewer number of existing cues in memory, ending with not so elaborate an encoding and thus with poorer recall of the product and/ or message features.

FIGURE 1

ESTIMATED MODEL

This view of familiarity is based on some basic, widely-in-use principles of the information processing paradigm, and is also useful in explaining the involvement construct. The probability that a person is more involved with a stimulus object is greater when that individual is more familiar with that object and thus has a better developed memory structure about it. As such, familiarity should be seen as an intervening variable for involvement. If involvement is a psychological and internal state whose activation is triggered by a particular stimulus, then the more an individual knows (i.e., more familiar s/he is) the higher will be his/her involvement.

COGNITIVE DIFFERENTIATION

One important feature of a memory structure is that it enables an individual to perceive differences in the features of a stimulus object and to make fine distinctions between that object and others. This is cognitive differentiation, and researchers suggest it is a key characteristic of-cognitive complexity or simplicity. A person who can perceive these fine distinctions in the stimulus object will do so because of the extensive knowledge the person has about that stimulus and the detailed/intricate way these knowledge cues are interconnected with one another. The more well-developed a memory structure is (the more familiar a person is with a stimulus) the higher the probability of being able to make fine distinctions between that stimulus object and others." (1965) describes high involvement as having more personal connections and bridging experiences. The more there are of these, the higher the ability of the individual to make finer distinctions. These connections and bridging experiences should be considered as cognitive traces, i.e., pieces of knowledge stored in memory. The more involved a person is, the more elaborate these knowledge cues will be. These cues may then be used to make cognitive differentiations. This state of more interconnectedness between knowledge cues in a memory will lead people to have better memory performances and recall. Even though we suggest that involvement, familiarity and cognitive differentiation will have similar effects on recall, they are not identical constructs. Familiarity may be an antecedent and cognitive differentiation can both be an antecedent and consequence of involvement.

PURPOSE OF THIS STUDY

As is apparent from the above discussion, familiarity, involvement, and cognitive differentiation are expected to be related to advertising recall. In addition, it can be seen that these three predictors are very similar from a theoretical perspective. They also are very similar from a measurement perspective; and these two facts (conceptual and operational similarity) have led to some confusion in the consumer behavior literature. For example, the following item: "I am familiar with this product category," has sometimes been proposed as a measure of involvement. In this manner, lack of conceptual clarity among these three constructs may have led to measurement confusion, and vice versa. The purPose of this paper is twofold:

1) To clear up some of this confusion of measurement by attempting to isolate a group of indicators that will measure involvement, as distinct from familiarity and cognitive differentiation. of course the same goal of measurement purity exists with respect to familiarity and cognitive differentiation

2. To investigate the relationships of involvement, familiarity, and cognitive differentiation tn ad recall.

To accomplish the first objective, tests of convergent and discriminant validity are reported. To accomplish the second goal, a causal model with multiple measures is estimated.

THE STUDY

Stimulus Objects

Constructs such as cognitive differentiation seem most relevant for describing consumers' perceptions of complex products (products with many salient attributes) as opposed to simpler products. For this reason, relatively complex products were chosen for study -- automobiles and stereo systems. Both products are comprised of a potentially wide range of attributes, and both may involve a relatively complex decision process on the part of the target audience. For each product category, a one-minute radio commercial was professionally produced for a fictitious, new brand in that category. Fictitious brands were used to minimize the impact of previous promotional efforts. These radio spots were embedded in regular programming material, consisting of an international news broadcast, in order to simulate a natural listening environment.

Subjects

The total sample size is 90. One-half of the subjects were exposed to the stereo advertisement. The remaining 45 subjects were exposed to the automobile ad. All of the subjects held full-time jobs, while attending graduate school part-time, and all were prescreened to establish that they were in the target audience for the advertised product and fully expected to make a purchase in that product category sometime within the next year.

Measurement

Subjects participated in three sessions. During the first session measures of cognitive differentiation, familiarity, and involvement were administered. These measures were randomly rotated in order to minimize any ordering effect. The first group of subjects completed these measures for the domains of world events and stereo systems. The second group was tested in the domains of world events and automobiles. During the second session, subjects were exposed to a radio broadcast consisting of a news program (international news and events), along with the target ad. In the third session. day-after recall measures were administered

Measures

One of the main purposes of this study is to isolate one group of indicators to measure involvement, such that involvement alone is measured as opposed to familiarity or cognitive differentiation. The same goal exists with respect to familiarity and cognitive differentiation -- to isolate purified indicators. With this goal of discriminant validity in mind, groups of measures were compiled for each construct. An attempt was made to put together a list of all measures which had appeared in the literature and which made sense from a theoretical perspective. After a series of pretests, the measurement instruments were refined and reduced in number. Table 1 presents a description of those which were finally selected for inclusion in this study. Note that Some measures employ a Likert-type scale. while others require that subjects complete an objective task. Complete descriptions and definitions of the measurement procedures are provided in Table 1.

ANALYSIS PROCEDURE

Our purpose, then, is to investigate the validity of the proposed measurement procedures while, at the same time, testing the strength of the causal relations between the three predictors and ad recall. When faced with this type of analysis problem there are two main choices: LISREL or Partial Least Squares (Fornell and Bookstein 1982; Fornell and Zinkhan 1982).

Recently, some problems associated with LISREL modeling have begun to surface (see, for example, Fornell and Larcker 1981). In particular, it seems as if data gathered in the course of consumer behavior research may not often satisfy some of the assumptions required by maximum likelihood estimation under LISREL. such as multinormality, interval scaling, or relatively large sample size. In addition there is the rather disturbing problem of improper solutions (e.g., negative error variance). Because of these problems there has been increasing interest in alternative modeling procedures such as Partial Least Squares (PLS). Unfortunately, some of these PLS applications have been a bit difficult to follow. For example, Jagpal (1981, 1982) has applied PLS estimation to the problem of evaluating hierarchy-of-effects models in advertising. In this application, advertising expenditure was measured and sales was measured, but no intervening constructs (such as awareness or preference) were explicitly measured. This leads to a rather odd model representation wherein advertising expenditure is represented as an indicator (rather than a cause) of awareness. Likewise, sales is represented as a measure, rather than as an outcome, of preference. Here an attempt is made to show a clearer application of PLS modeling in order to demonstrate its potential usefulness for consumer behavior research.

An advantage offered by so-called causal modeling techniques is the ability to separate empirical measures from underlying theoretical constructs and, in this way, assess construct validity. In this context, construct validity refers to the extent to which an observed measure reflects the underlying theoretical construct that the investigator has intended to measure (Andrews 1984; Cronbach and Meehl 1955). One step associated with testing construct validity involves an examination of the empirical relationship between measures and their underlying constructs (Zeller and Carmines 1980). One way to accomplish this is through consideration of convergent and discriminant validity where convergent validity refers to the degree to which multiple measures of the same underlying construct are in agreement, and discriminant validity refers to the degree to which two hypothetical constructs can be shown to be different (Campbell and Fiske 1959). In this sense, construct validity is a bit different from the notion of reliability and is different from other types of validity, such as ecological, content, or predictive validity (Zeller and Carmines 1980).

One common, though not universal, practice is to study a measure's validity by examining the correlation between a measure and its underlying construct. In the context of PLS modeling, this is represented by a loadings coefficient. By squaring this coefficient it is possible to get an idea of the proportion of valid variance in an indicator (Andrews 1984). Thus, if an indicator correlates .7 with its underlying construct, then that indicator contains 49 percent valid variance. Also note that the measure shares more variance in common with error than with its theoretical construct. Hence, a rule of thumb has developed that an indicator should correlate above .71 with its underlying construct and should share more variance in common with the construct than with error. If this condition is satisfied for an indicator, this is taken as evidence of convergent validity (Fornell et al. 1979).

Andrews (1984), in a study of six different surveys involving a total of 7706 respondents, apportioned total variance into three components: 1) valid variance; 2) correlated error variance; and 3) random error variance. That is, following classic measurement theory, a respondent's recorded answer to a particular survey item is assumed to reflect three types of influences: 1) the way the respondent really feels about the concept of interest (e.g., how involved the subject is with automobiles); 2) the way the respondent reacts to the data collection method (e.g., systematic error caused by the measurement instrument itself); and 3) everything else that might affect a recorded response (such as misunderstanding, fatigue, lapses of memory, etc.). Andrews (1984) found that, on average, a typical survey measure consisted of 66 percent valid variance, 3 percent method variance, and 28 percent error variance. These findings are encouraging in that an average survey item can be expected to share more in common with its theoretical construct than with error, and these findings can serve, along with others, as a benchmark against which subsequent measurement procedures can be compared. Also of note is the fact that the major component of survey error appears to be random rather than systematic.

In accordance with Andrews' results, we set out to determine the relative amount of residual variance as compared to the amount of valid variance in the constructs of interest: involvement, familiarity, and cognitive differentiation. It is, of course, possible to study systematic (or method) variance through an application of causal modeling (see Bagozzi 1978; Fornell, et al. 1982); but, since the methods used here are not radically different from one another and since Andrews (1984) has found that systematic variance is a relatively small portion of total error variance, only residual variance is considered here in relation to valid variance.

RESULTS

The relationships of the indicators to their constructs, along with the causal relationships among the constructs themselves, are summarized by the model shown in Figure 1. Following the procedure developed by Fornell, et al. (1982), this hypothesized model is examined with respect to convergent and discriminant validity.

Convergent Validity

As mentioned above, one way to assess convergent validity is to examine the loadings coefficients. As shown in Figure 1, all indicators save one share more variance in common with their underlying construct than with random error. The breakdown occurs for the Krugman task as an indicator of Involvement, where error variance (.51) exceeds shared variance with the latent variable (.49). Note, however, that this indicator just barely fails the test.

Another method for examining convergent validity has been developed by Fornell et al. (1982) where the variance shared by a construct is estimated by calculating the average of the squared loadings coefficient (Pvc) for a construct (see Fornell and Zinkhan 1984). Thus, Pvc provides a measure of the average amount of valid variance that an indicator shares with a latent variable.

A condition for satisfying convergence is that the value of PVC for a construct be greater than 0.5, i.e., the true variance should at least be greater than the error variance (Fornell and Larcker 1981). As summarized in Figure 1, all four constructs, on average, share more variance in common with their indicators than with error.

For example, all values for PVC are greater than .5, with the lowest Pvc value-being observed for Involvement (Pvc = .608) and highest value being observed for Familiarity (Pvc = .757)

In summary, the estimated model shown in Figure 1 seems to be acceptable in terms of convergent validity. The indicators load highly on the constructs which they are designed to measure.

Discriminant Validity

Fornell et al. (1982) demonstrates how PLS can be used to assess discriminant validity, the degree to which a construct differs from other constructs. If the squared correlation between any two constructs is lower than PVC for a construct, then there is evidence of discriminant validity. That is, discriminant validity is indicated if the variance shared between any two different constructs is less than variance shared between a construct and its measures (Fornell and Zinkhan 1984).

All four constructs share more variance in common with their indicators than with other constructs in the model. This can be clearly seen in Table 2 which summarizes the PVC values for each latent variable and displays the correlations among the latent variables themselves. In the PLS estimation procedure, the constructs are allowed to correlate among themselves; there is no assumption of indePendent factors.

As can be seen in Table 2, the largest squared correlation coefficient between constructs is .096 (r = .31 between Familiarity and Recall). Therefore, the smallest Pvc value is more than six times greater than the largest squared correlation between constructs. In short, the relationships among the constructs themselves are never greater than the relationships between a construct and its indicators. Discriminant validity is achieved for all four constructs.

Examination of Construct Relationships

Using PLS, it is possible to test the significance of a path coefficient through a jackknifing procedure. As shown in Figure 1, all three hypothesized paths between Recall and its predictors are significant at the .05 level; and all three paths are positive, as expected. Together Involvement, Familiarity, and Cognitive Differentiation explain 158 of the variance in Recall scores.

Thus, the hypothesized model is moderately successful in explaining the phenomenon of advertising recall. It is also revealing to examine the relationships among the three predictors, as shown in Table 2. Involvement, Familiarity, and Cognitive Differentiation correlate positively with one another, with the smallest relationship being observed between Involvement and Cognitive Differentiation (r = .17) and the largest relationship observed between Familiarity and Cognitive Differentiation (r = .30). In general, the relationships among the predictors are about equal in strength to the relationships observed between the predictors and ad recall.

LIMITATIONS

Before discussing the implications of these findings, some limitations should be re-emphasized. First, only one medium (radio) and two product categories were investigated. Second, subjects experienced forced exposure to one advertising message, and, in this sense, natural listening conditions were not fully simulated. Third, graduate students were used as subjects. However, somewhat ameliorating this problem is the fact that all graduate students held full-time jobs and all qualified as members of the target audience for the advertised products. Finally, our design did not allow us to investigate possible causal ordering among the predictors themselves. This final point is further discussed in the following section.

TABLE 1

DESCRIPTIONG OF INDICATORS

TABLE 2

CORRELATION BETWEEN CONSTRUCTS

DISCUSSION

The main contribution of this paper is in terms of measurement procedures. Specifically, five indicators have been proposed for Involvement, three indicators for Familiarity, and two for Cognitive Differentiation (see Table 1). With one minor exception for one of the Involvement indicators (the Krugman task), these proposed indicators seem adequate to the task of operationalizing the three predictor constructs. In particular, rather stringent tests of convergent and discriminant validity are passed within the context of PLS modeling. Also, the three indicators of ad recall appear to be acceptable as these too pass the validity tests.

Within the proposed model, however, rather weak relationships are reported between the three predictors and ad recall. Together, the predictors explain only 15% of the variance in recall scores. Although the hypothesized relationships are statistically significant, they are rather weak in nature. However, it must be pointed out that this level of explained variance is rather typical for consumer behavior research when measures of individual differences are used.

It is also interesting, within the context of the PLS model, to examine the relationships among the three predictors themselves. As can be expected, the predictors do correlate significantly among themselves; but again, none of the relationships is particularly large. For example none of the predictor-to-predictor is larger than any of the predictor-to-criterion relationships. As expected, Involvement, Familiarity, and Cognitive Differentiation are interrelated. Unfortunately, we haven't made much empirical progress toward disentangling these interrelationships; only correlation coefficients are examined.

We have, however, made some progress in isolating purified measures of each construct which can pass certain validity checks. In addition, within the context of the proposed model, we are able to examine the relationships of the three predictors to ad recall and, given these relationships, are able to examine the resulting correlations among the predictors themselves. What remains for future researchers is the task of elaborating upon the specific circumstances or situations necessary for establishing some causal ordering among the predictors. Little progress towards this end has been made to date; but, perhaps, an investigation employing some of the methodological advances suggested here may prove more fruitful.

REFERENCES

Andrews, Frank M. (1984), "Construct Validity and Error Components of Survey Measures: A Structural Modeling Approach," Public Opinion Quarterly, 48 (No. 2), 409-442.

Bagozzi, Richard P. (1978), "The Construct Validity of the Affective, Behavioral, and Cognitive Components of Attitude by Analysis of Covariance structures," Multivariate Behavioral Research, 13, 9-31.

Bieri, James, Alvin L. Atkins, Scot Briar, Robin Lobeck Leaman , Henry Miller, and Tony Tripodi (1966), Clinical and Social Judgement The Discrimination of Behavioral Information, New York: Wiley.

Buchanan, Dodds I. (1964), "How Interest in the Product Affects Recall: Print Ads vs. Commercials," Journal of Advertising Research, 4 (No. 1), 9-14.

Campbell, Donald T. and Donald W. Fiske (1959), "Convergent and Discriminant Validity by the Multitrait-Multimethod Matrix," Psychological Bulletin, 56 (March), 81-105 .

Cohen, J. B. (1983), "Involvement and You: 1000 Great Ideas," in R. P. Bagozzi and A. M. Tybout (Eds.), Advances in Consumer Research, 10, Ann Arbor: Association for Consumer Research, 325-328

Cronbach, L. J. and P. E. Meehl (1955), "Construct Validity in Psychological Tests," Psychological Bulletin, 52, 281-302.

Fornell, Claes and Fred Bookstein (1982), "Two Structural Equation Models: LISREL and PLS Applied to Consumer Exit-Voice Theory," Journal of Marketing Research, 19 (November), 440-452.

Fornell, Claes and David F. Larcker (1981), "Evaluating Structural Equation Models with Unobservable Variables and Measurement Error," Journal of Marketing Research, 18 (February), 39-50.

Fornell, Claes, Gerry J. Tellis, and George M. Zinkhan (1982), "Validity Assessment: A Structural Equations Approach Using Partial Least Squares," in An Assessment of Marketing Thought and Practice, B. J. Walker, et al. (eds. ), Chicago: American Marketing Association, 405-409.

Fornell, Claes and George M. Zinkhan (1982), "Classification Schemes for a Second Generation of Multivariate Analysis," in (R. Bush and S. Hunt, eds.) Marketing Theory: Philosophy of Science Perspectives, Chicago, American Marketing Association, 295-299.

Fornell, Claes and George M. Zinkhan (1984), "A Test of the Learning Hierarchy in High- and Low-Involvement Situations," working paper, University of Houston.

Greenwald, Anthony G. and Clark Leavitt (1984), "Audience Involvement in Advertising: Four Levels," Journal of Consumer Research, 11 (June), 581-592.

Jagpal, H. S. (1981), "Measuring Joint Advertising Effects in Multiproduct Firms," Journal of Advertising Research 21 (1), 65-69.

Jagpal, H. S. (1982), "Multicollinearity in Structural Equation Models with Unobservable Variables," Journal of Marketing Research, 19, (November), 431-439.

Krugman, Herbert E. (1965), "The Measurement of Advertising Involvement," Public Opinion Quarterly, 29 (Fall), 349-356.

Lastovicka, John L. (1979), "Questioning the Concept of Involvement Defined Product Classes," in Advances in consumer Research. 6. William Wilkie (Eds.), Ann Arbor: ACR, 174-179.

Mitchell, A. A. (1981), "The Dimensions of Advertising Involvement," in Kent B. Moore (Ed.) Advances in Consumer Research. 8. Ann Arbor: ACR, 25-30.

Mitchell, A. A. (1979), "Involvement: A Potentially Important Mediator of Consumer Behavior," in William L. Wilkie (Ed.), Advances in Consumer Research, 6, Arbor: ACR. 191-195.

Raju, P. S. and Reilly, Micheal D. (1979), "Product Familiarity and Information Processing strategies: An Exploratory Investigation," Journal of Business Research 8, 187-212.

Russo, J. Edward and Johnson, Eric J. (1980), "What Do Consumers Know About Familiar Products?" in Advances in Consumer Research, 7, Jerry C. Olson (Ed.), Ann Arbor: ACR, 417-423.

Scott, William A. (1962), "Cognitive Complexity and Cognitive Flexibility," Sociometry, 25, 405-414.

Zeller, R. A. and E. C. Carmines (1980), Measurement in the Social Sciences. New York: Cambridge University Press.

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