Integrating Cognitive Structure and Cognitive Response Approaches to Monitoring Communications Effects

Richard J. Lutz, University of California, Los Angeles (student), University of California, Los Angeles
John L. Swasy,
ABSTRACT - Communications research generated by cognitive structure models and cognitive response models is briefly reviewed and critiqued. Then, a joint cognitive structure/cognitive response model is proposed which capitalizes on the strengths of both the two previous approaches. Discussion of the new model focuses on its conceptual and operational implications, as well as its potential role in the development of policy-oriented communications models.
[ to cite ]:
Richard J. Lutz and John L. Swasy (1977) ,"Integrating Cognitive Structure and Cognitive Response Approaches to Monitoring Communications Effects", in NA - Advances in Consumer Research Volume 04, eds. William D. Perreault, Jr., Atlanta, GA : Association for Consumer Research, Pages: 363-371.

Advances in Consumer Research Volume 4, 1977   Pages 363-371


Richard J. Lutz, University of California, Los Angeles

John L. Swasy (student), University of California, Los Angeles

[This research was supported, in part, by an intramural grant from the Research Committee of UCLA.]


Communications research generated by cognitive structure models and cognitive response models is briefly reviewed and critiqued. Then, a joint cognitive structure/cognitive response model is proposed which capitalizes on the strengths of both the two previous approaches. Discussion of the new model focuses on its conceptual and operational implications, as well as its potential role in the development of policy-oriented communications models.


The general problem of monitoring the effects of persuasive communications has been of importance to advertisers, politicians and religious leaders for many years. Any communicator or policy maker who is interested in influencing public opinion or behavior must have feedback to allow an assessment of how well the persuasion attempt has worked. Thus, an advertiser may be interested in sales dollars, the politician in votes, and the religious leader in converted souls or dollars of contributions.

As important as these ultimate performance indicators may be, however, they are at best only descriptive of how well a campaign has accomplished its goal. What is even more important for the modern communicator to know is why the campaign has succeeded or failed. Only by moving to the level of explanation of communication effects can feedback be obtained which is capable of providing insights into how communications can be modified to achieve better results.

Much of the past empirical research on communications has been characterized by a simplistic, black-box model of the communications process. The dominant research strategy has been one in which attitudes (or some other dependent variable) following a persuasive message are compared with those prior to the message, or to those held by some control group. Literally thousands of studies, then, have focused on the manipulation of a plethora of independent variables, falling into the general categories of source, channel, message, and receiver. The major purpose of these studies has been to identify consistent patterns of response to the various independent variables, with the eventual goal of stating generalizable "principles" of communication. In large part, these efforts have been unsuccessful (Fishbein and Ajzen, 1972), and their failure in many cases can be attributed to the overly simplistic treatment of the communication process.

This simplicity is ironic in light of the fact that one of the earliest notions in modern communications research, as represented by the Yale research program, was that of certain cognitive and/or affective mediators of communications effects (Hovland, Janis and Kelley, 1953). This concept has endured over the past two decades and has been advocated by both advertising researchers (Lavidge and Steiner, 1961) and psychologists (e.g., McGuire, 1968). However, much of the research generated by these models either has failed to include operationalizations of the postulated mediators and simply inferred their existence from the observed data, or has measured only a subset of the mediators and used these data to infer the levels of other mediators. Thus, despite the seemingly powerful conceptualizations of the communications process available to researchers, very little empirical work has moved beyond the black-box model.

There are, however, at least two streams of research which have not suffered from the above problems. These two approaches to the communications process have focused on mediating cognitive variables as underlying explanations of communications effects. For convenience, throughout the remainder of this paper, these two general approaches will be referred to as the cognitive structure model and the cognitive response model. The purpose of this paper is to briefly describe and critique these two models and then integrate them to provide a more powerful model for monitoring communications effects.


The basic kernel of the cognitive structure model can be traced back to Cartwright's (1949) article on mass persuasion:

To influence behavior, a chain of processes must be initiated within the person. These processes are complex and interrelated, but in broad terms they may be characterized as (i) creating a particular cognitive structure, and (ii) creating a particular motivational structure, and (iii) creating a particular behavioral (action) structure. In other words, behavior is determined by the beliefs, opinions, and "facts" a person possesses [and] by the needs, goals, and values he has... (Cartwright, 1949, p. 255)

Cartwright's view of the persuasion process is consistent with a general expectancy x value (ExV) orientation, wherein an individual's behavior is seen as basically goal-directed and therefore capable of being modified by messages focusing either on the goals themselves or on the perceived paths to these goals (means-ends analysis).

While ExV models, or variants thereof, have recently been quite popular in consumer research as models of attitude structure (Wilkie and Pessemier, 1973), little research utilizing these models has been directed at attitude change. [Exceptions include articles by Axelrod (1963) and Lutz (1975a).] This has not been the case in psychology; here, several studies have been conducted under the general format shown in Figure 1 (e.g., Carlson, 1956; Peak, 1960).

Basically, this approach to monitoring communications effects goes a step beyond the black-box model by actually measuring certain cognitive variables that are assumed to mediate attitude change. This step is an important one, for it not only allows a more molecular assessment of what has occurred following a persuasion attempt, hut in addition provides a vehicle for diagnosing the basis of an attitude prior to initiating an influence attempt. Thus, the message construction process is made more scientific, in that it is focused on the presumed cognitive determinants of the dependent variable in question.



As shown in Figure 1, cognitive structure, represented by the ExV formulation, [There are many other possible dimensions of cognitive structure which could be investigated, e.g., degree of differentiation (Scott, 1968). However, the present discussion will be limited to the ExV representation of cognitive structure.] is seen as a causal influence on attitude, both before and after exposure to a persuasive communication. There is ample evidence (e.g., Lutz, 1976) which suggests that this causal order is operating in at least some persuasion situations, although Rosenberg (1960), in his classic study involving the hypnotic reversal of attitudinal affect, demonstrated that the causal order may be reversed in some cases. Nevertheless, the important point is that by comparing pre- and post-communication measures of cognitive structure, the communicator can determine if the message had its basic intended effect at the cognitive level. Only if cognitive change is observed can attitude be expected to change. More critically, if post-communication attitude is no different than pre-communication attitude, the cognitive structure index can be used to determine what went wrong with the campaign.

In recent years, there have been a number of modifications to the basic cognitive structure model. These modifications have been directed at obtaining a more precise delineation of the cognitive effects of a communication, while preserving the basic ExV representation of cognitive structure.

Kaplan (1972) introduced an attitude change model which extended the basic attitude structure model postulated by Fishbein (1963):


where A is attitude (affect) to toward an object; B. is the belief that the object possesses some attribute i; ai is the individual's evaluation of the ith attribute; and n is the number of salient attributes. Fishbein's theory states that those beliefs about an object which are salient to the individual, together with their related evaluative aspects (ai), constitute the underlying determinants of the attitude toward the object. While Fishbein focused his work primarily on attitude formation, it is clear that his basic orientation is essentially equivalent to Cartwright's earlier analysis of attitude change. Accordingly, the basic conceptualization of the Fishbein model fits into the paradigm suggested by Figure 1, with the index of cognitive structure being represented by the right-hand side of Equation 1.

Kaplan's (1972) extension of Fishbein's model essentially consisted of a partitioning of post-communication cognitive structure into two "mechanisms", labeled acceptance and impact. The acceptance mechanism models the increment in cognitive structure which results from a message introducing (i.e., making salient) new cognitive elements. The impact mechanism, on the other hand, is simply a post-communication measure of the original cognitive structure. Kaplan is assuming that the effects of introducing new beliefs into cognitive structure may not be purely additive, but may rather be interactive in the sense that the newly salient elements may induce changes in previously existing elements, even though these latter elements were not explicitly attacked in the persuasive message. It follows, then, that any change in attitude may be mediated not only by acceptance of a set of explicitly communicated beliefs but also by unanticipated impact on previously salient, but non-communicated beliefs.

Noting that Kaplan's model dealt only with the situation in which new beliefs were included in the persuasive message, Lutz (1975b) posited a model more general than Kaplan's in that it can be applied to situations where previously held beliefs are attacked, as well as those instances where new beliefs are communicated. Basically, the model postulates three types of cognitive effects which may result from a persuasive message. First-order effects consist of post-communication measures of the belief(s) attacked in the persuasive message, together with the associated evaluative aspects (al). These beliefs may be either new or previously existing; the key feature is that they are mentioned explicitly in the message. [An exception to this would be when the persuasive message is using an implicit conclusion with respect to the desired belief change. Here, even though the belief may not be mentioned explicitly, it would still constitute a first-order effect in that it is a specific goal of the message.]

When a persuasive appeal is directed only at adding new beliefs, the first-order effect is identical to Kaplan's acceptance mechanism. Under a belief modification appeal, the first-order effect is made up of that subset of previously salient beliefs which is being attacked in the message.

Second-order effects are those observed on elements of cognitive structure previously salient but not attacked in the persuasive message. In a situation where new cognitive elements are being added hut old beliefs are not under attack, the second order effect is equivalent to Kaplan's impact mechanism. When old beliefs are under attack, however, the second-order effect will consist of only the unattacked subset of pre-communication beliefs.

Finally, the third-order effect consists of those newly-formed beliefs which were not previously salient nor mentioned in the persuasive message. These beliefs may be formed through attributional or syllogistic processes initiated by belief statements within the message. The unanticipated formation of these beliefs may have considerable impact on the overall success of a persuasive message, either enhancing its effectiveness if the third-order effect is favorable, or attenuating it if unfavorable.

Ideally, the communicator would prefer large first-order effects and negligible second- and third-order effects, as this would afford maximum control over the effects of the campaign. It is unlikely that this situation exists often in reality, however. Kaplan (1972) and Lutz (1975b) found substantial impact and second-order effects, respectively, which suggests that non-communicated beliefs may be controlling non-trivial proportions of variance in post-communication attitudes.

The primary advantage that the Kaplan and Lutz models offer over the basic cognitive structure model is the partitioning of cognitive effects into meaningful categories for the purpose of diagnosing the reasons underlying communications effects. A more recent elaboration of the cognitive structure model has been undertaken by Fishbein and Ajzen (1975). Their analysis is considerably more complex than either of the two preceding ones and would appear to offer considerable promise in tracking communications effects even more precisely.

The Fishbein and Ajzen model is shown in Figure 2.



The essential modification preferred by this model is the treatment of beliefs other than those seen as direct determinants of attitude. Previous ExV models have focused only on salient beliefs about the attitude object per se. In their latest model, Fishbein and Ajzen examine the relationships among beliefs communicated in the message (source beliefs), corresponding beliefs held by the audience (proximal beliefs), determinant (or primary beliefs,) and so-called external beliefs.

In brief, the communications process is conceived of as follows. The receiver, characterized by an initial cognitive structure and attitude, is exposed to a message consisting of information items, or source beliefs. These source beliefs exist only as perceived by the receiver and are subject to distortion and other selective processes. Depending upon the degree of discrepancy perceived between the source's belief and the receiver's own belief, acceptance of the source belief may occur. Corresponding directly to the communication beliefs are the so-called proximal beliefs. These are simply the receiver's own initial beliefs (represented by subjective probabilities) which may or may not be altered by the influence attempt. Proximal beliefs may be further disaggregated into primary beliefs, i.e., those that are determinants of the eventual dependent variable, and non-primary, which exert no direct causal impact on the dependent variable. Beliefs which are not mentioned in the message but which become salient (or were previously salient) are designated external beliefs. External beliefs can also be partitioned into primary and non-primary categories, depending upon their causal relation to the dependent variable. Primary external beliefs would encompass the second-order and third-order effects in the Lutz model discussed above. Both the non-primary proximal and external beliefs may serve as determinants of primary beliefs, but not the ultimate dependent variable.

The principal advancement of the Fishbein and Ajzen model over previous ExV models is the explicit treatment of belief variables other than the primary (i.e., determinant) beliefs. As yet, the model awaits rigorous empirical investigation, and as the multiplicity of arrows in Figure 2 indicates, there is considerable ambiguity regarding the precise interrelationships among the model's constructs.

Critique of the Cognitive Structure Model

The cognitive structure approach to monitoring communication effectiveness has some clear strengths which make it an attractive tool for policy makers. First, its emphasis on determinant cognitive structure both before and after the communication enables the communicator to both derive strategy and monitor results within the same theoretical framework. Second, due to the close-ended measurement procedures for the model's constructs, it is amenable to group administration and survey methods, an important feature for the policy maker and often the researcher. Finally, the measurement procedures are, by and large, well-established with reasonable reliability and validity. The approach can be fairly confidently applied, with little danger of serious error.

On the negative side, the cognitive structure model may be seen as overly structured, to the point where structure may overcome substance and actually obscure relevant data regarding communications effects. For instance, the scaling procedure used to measure all constructs in the model may unnecessarily constrain respondents to an unnatural form of reaction to the message. Too often experimental subjects may dutifully fill out the semantic differential scales, while harboring a much different reaction internally, but one which the method will not allow them to express. Second, there has been a tendency to rely too heavily on only pre-determined salient beliefs in monitoring communications effects. More attention must be paid to the "external" beliefs generated spontaneously by message recipients. Finally, the cognitive structure model tends to ignore cognitive process in its construction of communications effects. While the recent model proposed by Fishbein and Ajzen begins to move away from a total reliance on structure, there is as yet no empirical work which has been generated by that model.

In sum, both the strengths and weaknesses of the cognitive structure model seem to center around its own structure: pre-determined, scaled variables, combined in a pre-specified fashion.


The purpose of this section is to briefly review cognitive response research. Those studies which explicitly deal with cognitive responses generally include some or all of the following steps:

(1) the subject is rehearsed in verbalizing or writing his cognitive responses--"thoughts," "reactions," "ideas," etc.,--which occur during exposure to a message.

(2) the subject's cognitive responses are collected by tape recorder or in written form either during or after the message, and under either limited or unlimited response time constraints.

(3) the cognitive responses are then categorized on various criteria by the subject or the experimenter. These criteria may include: "critical" vs. "favorable" orientation to the message, importance of thought to overall opinion, perceived origin of the thought, or whether or not the response meets a prescribed definition of a class of cognitive responses. For example, to be classified as counterarguing, the cognitive response may have to clearly state a position against the advocated message (Brock, 1967), whereas a source derogation classification may require an expression of distrust or derogation of advertisements or the advertiser (Wright, 1973). As noted by Wright (1974), these coding schemes and definitions have varied widely.

(4) the fourth step is to compute category scores for each subject. This may simply be the number of responses in each category or possibly an importance-weighted index for each cognitive response category. These scores are then used as the operationalization of the cognitive response postulated by the theory under investigation.

Areas which have utilized cognitive responses include fear appeals, distraction, source credibility and advertising effectiveness. Janis and Feshbach (1954) hypothesized that the anxiety drive state created by fear appeals would motivate the individual to engage in different types of internal responses to avoid anxiety-arousing cues. They reasoned that responses of an "interfering nature---inattentiveness, perceptual distortions, defensive efforts to deny or minimize the threat ..." would occur. (Janis and Feshbach, 1954, p. 162). In a subsequent paper, Janis and Terwilliger (1962) directly investigated these proposed internal responses and found that subjects receiving the high threat communication expressed more explicit rejection statements and fewer responses indicating acceptance. This was interpreted as support for the original Janis and Feshbach hypothesis.

In the source credibility area, Festinger and Maccoby (1964) and McGuire (1965) postulated the mediating role of counterarguing as an explanation for the impact of perceived source competence. Cook (1969) tested McGuire's hypothesis by manipulating source credibility and reception and measuring subjects' counterarguing. Basically, Cook's results were in accord with McGuire's hypothesis. He found that attitude change and counter-arguing were negatively correlated and affected by competence in the expected direction. His analysis of whether or not counterarguing preceded attitude change was based on three facts: (1) Using counterarguing as a covariate eliminated the treatment effects on attitude, (2) Counterarguing was measured before attitude and (3) Other studies (Festinger and Maccoby, 1964; McGuire, 1965) were predicated on low counterarguing causing attitude or belief change (Cook, 1969, p. 355). While his conclusion is plausible and intuitively appealing, it may be somewhat overstated since a hypothetical model treating counterarguing and attitude as correlates (see Figure 3) would yield the same empirical results.

In the distraction effects research area, Osterhouse and Brock (1970) provided additional evidence for the mediating role of counterarguing by demonstrating that the effect of distraction on attitude change became insignificant when the number of counterarguments was used as a covariate.

Insko, Turnbull and Yandell (1974) also incorporated measures of counterarguing into the distraction paradigm. A distinctive aspect of this design was a "set" factor which represented the subject's orientation to the message or the distraction task. Using an adaptation of Blalock's (1964) procedure for testing causal models with correlational data, the authors suggested that the set factor (message vs. task orientation) led to different mediation factors of distraction's effect on attitude change. For the message set, they hypothesized that distraction led to increased "communication-favorable'' thoughts through decreased counterargumentation and finally to attitude change. In the task set the suggested mediators were a devaluation of the communication and a decrement in message recall. It should be noted, however, that these proposed causal models were based on a post-hoc analysis and that additional exogenous variables are needed before a time test of the causal order can be made.

The importance of the set factor demonstrated by Insko, Turnbull and Yandell coincides nicely with the earlier results of Wright (1973) concerning information processing involvement. This research was one of the first direct applications of cognitive response in marketing. In his research, Wright manipulated "content-processing involvement" and analyzed how well various combinations of weighted and unweighted cognitive response variables predicted attitude. [Also included as major variables in his analysis, but not of direct interest here, were measures of message reception, origin of thoughts, thought "importance", and behavioral intentions.] (See Figure 4). Using measures of source derogation, counterarguing, and support arguing, between 35 and 60% of the variance in message acceptance could be explained. Counterarguing was found to be the best single predictor. The impact of involvement was evidenced by the increase in predictive power of the unweighted models in the high involvement situations.



This finding is given further support by the research of Calder, Insko and Yandell (1974), who found in a simulated jury trail experiment that an unweighted index of the number of prosecution thoughts minus the number of defense thoughts predicted final belief as well as a weighted importance index.



This brief summary of research employing cognitive response models was necessarily selective. More exhaustive reviews of this literature have been provided by Roberts and Maccoby (1973) and Wright (1974). The intent here was to provide an overview of the cognitive response method and to raise some key issues for future research in this area.

Critique of the Cognitive Response Model

Wright (1974) and Ward (1974), in an excellent exchange, highlight the positive features of the cognitive response model:

1) The use of cognitive responses makes no pre- judgments about what responses are possible, thus allowing the subject to exhibit a wider range of responses than simply attitude change;

2) The cognitive response methodology allows for the study of subjective reactions during message exposure, thereby enabling researchers to move more closely to the study of process rather than inferring process mechanisms from pre- and post-communication measures; and,

3) Cognitive responses allow for the study of communications effects at both the individual and aggregate levels.

On the other hand, cognitive response research has several seri6us limitations including:

1) Demand characteristics of the data collection process which may artificially produce "cognitive'' activity and thus eliminate low involvement processing;

2) The potential effect of differences in when and how the responses are obtained (e.g. "during" vs. "after", limited vs. unlimited response time, verbal vs. written responses); [See Roberts and Maccoby (1973) for an empirical comparison of "during" and "after" cognitive response measurement.]


3) The wide variation of coding schemes for cognitive responses, reflecting different conceptual views of cognitive response (e.g., origin of the thought [Greenwald, 1968] vs. type of thought---support arguments, counterarguments, source derogations, etc.).

As Wright (1974) notes, the cognitive response method is still in its infancy and no doubt many of the research issues will be handled as additional studies are conducted. However, it appears that there is a major question regarding the construct validity of cognitive response variables. Current support for the assumed causal relationship shown in Figure 3 (a) is based solely on correlational data and "conceptual" appeal.

Both Cook (1969) and Osterhouse and Brock (1970) relied on a covariance analysis to support the model and found that the treatment effect was eliminated when counter-arguing was removed. Unfortunately they did not report whether or not covarying attitude removed the effects of the independent variable(s) on counterarguing. Their model assumes that this analysis would show no effect on counterarguing. In the absence of this analysis, the two models illustrated in Figure 3 cannot be distinguished. Alternatively, additional exogenous variables could be introduced into the system, thus allowing the use of more powerful causal models (Duncan, 1975). Thus, despite the predictive validity of past cognitive response research, the construct validity issue is seriously clouded. [For example, Miller (1971) claims that counterarguing is merely a rationalization process.]


It is clear from the preceding discussion of the cognitive structure and cognitive response models that each model has some unique strengths and some inherent weaknesses. Interestingly, in several instances the strength of one model coincides with a weakness of the other model. This suggests that the two models may serve as useful complements to one another in pursuing the general goal of tracing communications effects. In particular, cognitive structure-models could be applied to the problems of mediator coding and construct validity within the cognitive response model, while the latter model can be useful in broadening response modalities and moving the cognitive structure models along a more process-oriented path.

A first step toward integrating the two models might be to use cognitive structure models for representing the various cognitive response mediating variables, such as the one shown in Figure 4. This idea is one which was first explicitly stated by Calder, Insko and Yandell (1974) and mentioned again by Calder (1975). Basically, what this proposed modification entails is a two-step measurement procedure, wherein subjects' free responses would be recast into belief statements of the form used by Fishbein and his associates. In previous cognitive response research, all thoughts or associations mentioned by a subject have been treated as equivalent in belief strength. By allowing the subject to express his/her degree of agreement with the various thoughts that become salient, a more precise estimate of belief strength (Bi) can be obtained.

In a similar vein, subjects would be asked to provide ai (goodness-badness) ratings for the beliefs that they first mention spontaneously. This is similar to a procedure used by Greenwald (1968). Thus, it should be possible, by following "free" cognitive responses with more precise measurement procedures, to characterize each belief along both the belief strength and evaluative dimensions. This procedure would obviously be quite "labor-intensive" and would require individual administration of experimental materials. Nevertheless, it would appear to be manageable, and in fact, not that much more involved than many of the experimental procedures reported previously in the cognitive response literature.

Following the experimental session, all belief statements would be judgmentally classified into the various cognitive response categories, perhaps those shown in Figure 4, or some other classification. Using the Bi and ai measures, then, separate indices like that shown in Equation 1 would be constructed to represent all the cognitive response variables. Analysis would then proceed with these cognitive structure indices representing the cognitive response mediators.

One observation that becomes apparent in moving to cognitive structure representations of cognitive response mediators is that the source derogation mediator should not be directly causally linked to the post-communication attitude toward the object or issue being treated in the persuasive message. The types of beliefs which would be categorized as source derogations would be those which express a belief about the source of the communication rather than the object of the communication. According to cognitive structure theory, these beliefs about the source should combine with their related ai components to form an attitude toward the source, but not toward the attitude object in question. In order for source derogation to have a causal impact on attitude, it must first have an effect on the primary beliefs underlying the attitude. Thus it appears that source derogation beliefs may be what Fishbein and Ajzen (1975) call external beliefs, which are not directly linked to attitude, but which may influence attitude through other (primary)beliefs. In contrast, it would appear that both the counterarguing and support arguing mediators contain beliefs directly relevant to the attitude object and hence should be seen as direct causal influences on post-communication attitude. In sum, it seams clear that counterarguing, support arguing and source derogation are not parallel processes directly mediating attitude change; the cognitive structure modeling of these response mechanisms vividly demonstrates the qualitative differences between the latter construct and the former two.

By integrating the cognitive structure and cognitive response approaches, a clearer picture begins to emerge regarding the conceptual status of some of the more common cognitive response variables. It is not immediately evident what some of the rather ad hoc indices used in past research (e.g., number of counterarguments) may have really meant. For example, does a large number of counterarguments indicate only counterarguing, or does it also pick up a general propensity to write down (or mention) thoughts in general? The expectancy-value formulation proposed here has a clear conceptual status as a summary measure of cognitions underlying attitudinal affect.

Using the ExV representation of cognitive measures also clarifies the nature of causal relationships among the variables. It seems apparent that a simple linear representation of these constructs is inadequate. Therefore, a more elaborate model is dictated, one which attempts to specify more precisely the possible causal relationships among the major mediating variables.

Figure 5 portrays one possible form that a joint cognitive structure/cognitive response model might take. Basically it represents a blend of the most recent cognitive structure model (Fishbein and Ajzen, 1975) and a common form of the cognitive response model (Wright, 1973, 1974). The principal goal of this joint model is to begin a more thorough investigation of causal relationships among the constructs in the original models.

Moving from left to right through Figure 5, it can be seen that the first four constructs in the model are identical to those in Figure 2, a cognitive structure model. In contrast to the cognitive response model, pre-communication cognitive structure is treated explicitly, as is the perception of the message. Wright (1973) included measures of message reception which closely parallel the perception variable, but he did not treat these measures as mediating variables as proposed here, instead entering them into a linear regression equation along with his cognitive response variables. Thus, Figure 5 posits message perception as a major mediating variable which can have direct causal impact on five subsequent variables. Of these five variables, three are the basic cognitive response measures, while the other two come from the cognitive structure research tradition.



Acceptance is identical to the first-order effect in the Lutz (1975b) model discussed earlier, and deals with post-communication belief levels with respect to beliefs attacked in the message. This variable may or may not be a cognitive response variable, depending upon whether subjects volunteer a statement regarding their belief or disbelief, of do not mention it spontaneously but simply respond to that item on the post-communication questionnaire. It seems likely that, in many cases, these beliefs will be mentioned in the free-response phase, and Acceptance will assume the status of a cognitive response mediator.

The Impact construct, on the other hand, is not viewed as a cognitive response variable; rather, it is made up of those beliefs which were salient prior to the message, but were not mentioned in the free-response phase of the experiment. Instead, they are simply re-measured following the communication and exist as a portion of post-communication cognitive structure. This Impact construct is not seen as being identical to Kaplan's (1972) impact mechanism nor Lutz's (1975b) second-order effect. The reason for the discrepancy is that, under the present model, certain previously salient but non-communicated beliefs may be spontaneously mentioned by the subject and thus enter the Counterarguing or Support Arguing constructs. Thus the Impact variable models that portion of pre-communication cognitive structure (n beliefs) that is not mentioned in the message (k beliefs) nor spontaneously by the message recipient (r beliefs).

Support Arguing and Counterarguing are modeled here by the cognitive structure indices discussed above. These two constructs, together with the Acceptance and Impact variables, constitute the postulated direct causal mediators of post-communication attitude. Source Derogation is no longer seen as being directly linked to attitude, but rather is conceptualized as operating on attitude through one or a combination of the direct cognitive mediators. Therefore, in Figure 5, Source Derogation is not shown as being modeled via Equation 1, as are the other cognitive response variables. The dashed arrow in Figure 5 illustrates that Source Derogation is expected to have direct causal impact on perceptions of the source, e.g., credibility and attractiveness. This effect would be of obvious importance in subsequent communications situations involving the same source.

The two-way arrows between certain pairs of constructs in the proposed model indicate ambiguities in the present knowledge about the nature of cognitive response variables. For instance, does acceptance of message beliefs lead to or follow from counter-arguing and support arguing? This would appear to vary situationally and with respect to the numerous possible independent variables which may be manipulated in a persuasion campaign. In a similar vein, impact on cognitive structure may either precede or follow attitude change. Whichever causal order is likely to prevail would again appear to be variable across situations. Empirical research utilizing the proposed model could shed light on the dominant direction of the causal flow along the two-way arrows.

Critique of the Proposed Joint Model

The joint model, as shown in Figure 5, would appear to offer a number of advantages over either of the original models taken singly. First, it combines the strengths of the free-response format with the more precise measurement procedures of cognitive structure models, thus not constraining subjects', spontaneous responses but using them as a starting point for subsequent measurements. Second, both structure and "process" are considered in the joint model; in particular, the inclusion of the pre-communication cognitive structure measure is an important difference from most cognitive response research given the diagnostic capabilities offered by that measure. And the emphasis on cognitive response mediators goes beyond most past cognitive structure research in moving toward an answer to the nagging question of the origin of the beliefs in cognitive structure.

Third, and a very important point, is that the combination of the two models has already helped to clarify the conceptual status of the Source Derogation mediator. This was due to the focus of the cognitive structure model on the ultimate dependent variable (throughout the present analysis, attitude). Cast in cognitive structure terms, it becomes obvious that source derogation should not have a direct causal effect on attitude. Future cognitive structure analyses of other potential cognitive response variables may help to clarify their roles in the communications process.

The proposed joint model is not without its disadvantages; the major ones center around its complex and quite possibly reactive measurement procedures, which include a free-response phase in which beliefs are elicited, following which the free-response beliefs must be placed in appropriate format to quantitatively assess belief strength (Bi) and evaluation (ai). A second related disadvantage is the virtual necessity of individual level administration with the experimenter and subject face to face. Here the method almost seems to take on the character of a depth interview in response to a persuasive communication. Finally, the proposed model really is of little help in solving the coding problems of cognitive response research. Judgmental classification of thoughts into response categories is still required, though the cognitive structure measures may be of marginal assistance in this task.



The issue of reactivity of measurement procedures has plagued both the cognitive structure and cognitive response research traditions. While it certainly seems likely that some reactivity is present under both approaches, and with the joint model proposed above, the magnitude of the effect is not certain. In view of the purpose and seeming promise of these cognitive models for examining communications effects, it would seem reasonable to directly investigate the reactive nature of these models to determine just how bad they really are.

One way to approach this problem would be to trace the effects of using the cognitive models on overall attitude change, the major dependent variable of concern. Starting with the simple black-box model as a baseline, where no cognitive measures of any kind are employed, one could view cognitive response measurement and cognitive structure measurement as treatment variables in an ANOVA sense. This suggests the design shown in Figure 6. By using this 2 x 2 design, the black-box model could serve as a control group for the testing of reactive effects on attitude of the other three models, including the joint model proposed in this paper. With this evidence in hand, the researcher could, on a more informed basis, make the explicit trade-off between reactivity and the value of the increased information about communications effects.



Causal Ordering Among Mediators

One of the supposed virtues of the joint model proposed here is its emphasis on teasing out the causal ordering among cognitive mediators. [Obviously, causal relations could be investigated within either of the original models and have been in some instances (Cook, 1969; Osterhouse & Brock,1970; Lutz,'76.)] A system as complex as the one portrayed in Figure 5 is not easily amenable to traditional experimental manipulation. It uses mostly response rather than stimulus variables and therefore cannot be directly manipulated as can the independent variables in the communications process. Accordingly, causal correlational analysis becomes an attractive tool for examining this model (Blalock, 1964; Duncan, 1975).

Causal correlational analysis allows the researcher to examine the correlations and partial correlations among a set of theoretical constructs in order to determine if the pattern of these correlations coincides with that which would be observed if the proposed causal ordering were true. When two or more theoretical models are competing explanations of a phenomenon, that model which most closely fits the data is supported, while the competitors are ruled out. Causal correlational analysis works best with fairly "rich" theoretical networks such as that in Figure 5 because alternative hypotheses are more easily rejected. This is not the case with simpler models such as the three-variable systems tested by Cook (1969) and Osterhouse and Brock (1970).

One assumption underlying the application of causal models is that the theory being tested is reasonably well established on conceptual grounds. This avoids endless iterations of the variables in the model in order to achieve better "fit". The model proposed in Figure 5 consists of a number of relationships that rest on a firm theoretical base and others that are in question, as indicated by the two-way arrows. These latter relationships could be treated as alternative models in empirical tests in order to ascertain which direction the arrow should be pointing in most instances.

Of greater concern in the testing of the joint model is the problem of identification within the model. The model as depicted in Figure 5 is underidentified, which is to say that certain parameters cannot be unambiguously estimated due to confounding with other relationships in the model (Duncan, 1975). While a more thorough discussion of identification is beyond the scope of this paper, the best remedy for underidentification is to introduce additional exogenous (i.e., independent) variables into the model. These exogenous variables should not be global in nature, but should be expected to affect the various endogenous variables in the model differentially. While it is not clear at this point exactly what set of exogenous variables would allow the model to be identified for purposes of estimation, examples of such variables taken from previous research include involvement (Wright, 1973), distraction (Osterhouse and Brock, 1970), source credibility and attractiveness (McGuire, 1968), self-confidence (Cox and Bauer, 1964), and mode of presentation (Wright, 1973). It appears likely that other communications, situational, and individual difference variables could be introduced into the system to allow empirical testing. However, this would constitute a demanding task at both the theoretical and data collection levels.

By increasing the number of constructs in the model to achieve identifiability, the burden on the individual subject is increased as more and more data are required of each person. Secondly, in order to provide enough degrees of freedom to allow meaningful estimation of the parameters in the model, the sample size would have to be reasonably large. Given the individual level administration required by the model, this becomes an imposing burden on the researcher.


The proposed integration of the cognitive structure and cognitive response models into a joint communications model appears to capture the power of both approaches, to the mutual benefit of both the original models. In particular, the joint model corrects some of the weaknesses inherent to the models when taken singly, but is unable to correct some of the other weaknesses, especially the reactivity and coding issues.

The joint model has provided some further insights into the nature of communications effects, most notably in the area of measurement of cognitive response variables and their status as mediating constructs. The identifiability issue, while difficult, points to the need for further elaboration of the communications model to include important exogenous variables.

Due to the increased data collection problems implied by the joint model and the causal methodology outlined, there is likely to be a great deal of resistance to applying the proposed model. Perhaps it can be partitioned to ease initial investigations.

Despite the troublesome obstacles to empirical research, the proposed model would appear to offer promise for the continuing study of communication effects. The adoption of a causal process perspective introduces the technology necessary for evolutionary model building in the pursuit of increased understanding of the communications process. At the policy level, it would appear that after appropriate testing in the academic environment, a streamlined version of the joint cognitive structure/cognitive response model may be of great practical utility in communication pretesting.

For example, a unified advertising pretesting and monitoring system could be developed along these lines to generate message tactics, pretest the various cognitive impacts of the message, and then monitor its effects in the marketplace. Clearly this sort of system, to be workable, would have to be more parsimonious than that portrayed in Figure 5, but that would not be an impossible adjustment. The goal of the present paper was to suggest an approach for increased understanding of the communications process. Once that process is more clearly understood, then the basic core can be applied to more practical problems.


Joel N. Axelrod, "Induced Moods and Attitudes Toward Products," Journal of Advertising Research, 3 (1963), 19-24.

Hubert M. Blalock, Jr., Causal Inference in Non-experimental Research (Chapel Hill: University of North Carolina Press, 1964).

Timothy C. Brock, "Communication Discrepancy and Intent to Persuade as Determinants of Counterarguing," Journal of Experimental Social Psychology, 3 (1967), 296-309.

Bobby J. Calder, Chester A. Insko and Ben Yandell, "The Relation of Cognitive and Memorial Processes to Persuasion in a Simulated Jury Trial," Journal of Applied Social Psychology, 4 (1974), 62-93.

Earl R. Carlson, "Attitude Change Through Modification of Attitude Structure," Journal of Abnormal and Social Psychology, 52 (1956), 256-261.

Dorwin Cartwright, "Some Principles of Mass Persuasions," Human Relations, 2 (1949), 253-267.

Thomas D. Cook, "Competence, Counterarguing, and Attitude Change," Journal of Personality, 37 (1969) 342-358.

Donald F. Cox and Raymond A. Bauer, "Self Confidence and Persuasibility in Women," Public Opinion Quarterly, 28 (1964), 453-466.

Otis Dudley Duncan, Introduction to Structural Equation Models, (New York: Academic Press, 1975).

Leon Festinger and Nathan Maccoby, "On Resistance to Persuasive Communication," Journal of Abnormal and Social Psychology, 68 (1964), 359-367.

Martin Fishbein, "An Investigation of the Relationships Between Beliefs about an Object and the Attitude Toward That Object," Human Relations, 16 (1963) 233-240.

Martin Fishbein and Icek Ajzen, "Attitudes and Opinions," Annual Review of Psychology, 23 (1972), 487-544.

.Martin Fishbein and Icek Ajzen, Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research, (Reading, Mass.: Addison-Wesley, 1975).

Anthony G. Greenwald, "Cognitive Learning, Cognitive Response to Persuasion, and Attitude Change," in Anthony G. Greenwald, Timothy C. Brock and Thomas M. Ostrom (eds.), Psychological Foundations of Attitudes (New York: Academic Press, 1968), 147-170.

Carl I. Hovland, Irving L. Janis, and Harold H. Kelley, Communication and Persuasion (New Haven: Yale University Press, 1959).

Chester A. Insko, William Turnbull, and Ben Yandell, "Facilitative and Inhibiting Effects of Distraction on Attitude Change," Sociometry, 37 (1974), 508-528.

Irving L. Janis and Seymour Feshbach, "Personality Differences Associated with Responsiveness to Fear Arousing Communications, Journal of Personality, 23 (1954), 154-166.

Irving L. Janis and Robert F. Terwilliger, "An Experimental Study of Psychological Resistances to Fear Arousing Communications," Journal of Abnormal and Social Psychology, 65 (1962), 403-410.

Kalman J. Kaplan, "From Attitude Formation to Attitude Change: Acceptance and Impact on Cognitive Mediators," Sociometry, 35 (1972), 448-467.

Robert Lavidge and Gary A. Steiner, "A Model for Predictive Measurement of Advertising Effectiveness," Journal of Marketing, 25 (1961), 59-62.

Richard J. Lutz, "Changing Brand Attitudes Through Modification of Cognitive Structure," Journal of Consumer Research, 1 (1975), 49-59.

Richard J. Lutz, "First-Order and Second-Order Cognitive Effects in Attitude Change," Communication Research, 2 (July 1975), 289-299.

Richard J. Lutz, "An Experimental Investigation of Causal Relations Among Cognitions, Affect, and Behavioral Intentions," Journal of Consumer Research, 1976, forthcoming.

William J. McGuire, "Inducing Resistance to Persuasions," in L. Berkowitz (ed.), Advances in Experimental Social Psychology, (New York: Wiley, 1964).

William J. McGuire, "Personality and Susceptibility to Social Influence" in E.G. Borgatta and W.W. Lambert (ads.), Handbook of Personality Theory and Research, (Chicago: Rand McNally, 1968), 1130-1187.

N. Miller, "On Measuring Counterarguing," Paper presented at American Psychological Association, Washington, D.C. (1971).

Robert A. Osterhouse and Timothy C. Brock, "Distraction Increases Yielding to Propaganda by Inhibiting Counter-arguing," Journal of Personality and Social Psychology, 15 (1970), 344-358.

Helen Peak, "The Effects of Aroused Motivation on Attitudes," Journal of Abnormal and Social Psychology, 61 (1960), 463-468.

Donald F. Roberts and Nathan Maccoby, "Information Processing and Persuasion: Counterarguing Behavior," in Peter C. Clarke (ed.), New Models for Mass Communication Research, (Beverly Hills: Sage Publications, 1973), 269-307.

Milton J. Rosenberg, "Cognitive Reorganization in Response to Hypnotic Reversal of Attitudinal Affect," Journal of Personality, 28 (1960) 39-63.

William A. Scott, "Attitude Measures," in G. Lindzey and E. Aronson (ads.), Handbook of Social Psychology, Vol. II (2nd Ed.), (Cambridge, Mass.: Addison-Wesley, 1968).

Scott Ward, "A Discussion of Wright's Paper on Direct Monitoring," in G.D. Hughes and M.L. Ray (ads.), Buyer/Consumer Information Processing (Chapel Hill: University of North Carolina Press, 1974), 249-255.

William L. Wilkie and Edgar A. Pessemier, "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, 10 (1973), 428-441.

Peter L. Wright, "On The Direct Monitoring of Cognitive Response to Advertising," in G.D. Hughes and M.L. Ray (ads.), Buyer/Consumer Information Processing (Chapel Hill: University of North Carolina Press, 1974), 220-248.

Peter L. Wright, "The Cognitive Processes Mediating Acceptance of Advertising," Journal of Marketing Research, 10 (February, 1973), 53-62.