Effects of Involvement on On-Line Brand Evaluations: a Stronger Test of the Elm

Jong-Won Park, Korea University
Manoj Hastak, American University
ABSTRACT - The Elaboration Likelihood Model predicts differences in the processes mediating persuasion effects for involved and uninvolved audiences C but only in those situations where persuasion occurs on-line, i.e., during message exposure. However, previous tests of the ELM have failed to provide compelling evidence for on-line persuasion, and hence have failed to create the strongest possible test for the ELM. Such a test was created in a pilot and a main study. Results from the pilot study showed that involvement influenced the response time for on-line brand evaluations and subsequent recall of product information. In the main experiment, typical ELM manipulations and procedures were replicated, but all subjects were instructed to evaluate the advertised brands while they were viewing the ads. The results were consistent with ELM predictions. Implication for future research on the ELM are discussed.
[ to cite ]:
Jong-Won Park and Manoj Hastak (1995) ,"Effects of Involvement on On-Line Brand Evaluations: a Stronger Test of the Elm", in NA - Advances in Consumer Research Volume 22, eds. Frank R. Kardes and Mita Sujan, Provo, UT : Association for Consumer Research, Pages: 435-439.

Advances in Consumer Research Volume 22, 1995      Pages 435-439

EFFECTS OF INVOLVEMENT ON ON-LINE BRAND EVALUATIONS: A STRONGER TEST OF THE ELM

Jong-Won Park, Korea University

Manoj Hastak, American University

ABSTRACT -

The Elaboration Likelihood Model predicts differences in the processes mediating persuasion effects for involved and uninvolved audiences C but only in those situations where persuasion occurs on-line, i.e., during message exposure. However, previous tests of the ELM have failed to provide compelling evidence for on-line persuasion, and hence have failed to create the strongest possible test for the ELM. Such a test was created in a pilot and a main study. Results from the pilot study showed that involvement influenced the response time for on-line brand evaluations and subsequent recall of product information. In the main experiment, typical ELM manipulations and procedures were replicated, but all subjects were instructed to evaluate the advertised brands while they were viewing the ads. The results were consistent with ELM predictions. Implication for future research on the ELM are discussed.

INTRODUCTION

The effects of issue involvement on the processes mediating issue evaluation are of interest to social psychologists as well as consumer researchers. A particularly well accepted model in this area is the Elaboration Likelihood Model (ELM) proposed by Petty and Cacioppo (1983, 1986). The ELM suggests that brand involvement (defined as degree of perceived personal relevance of the advertised brand) is one of the key determinants of the way in which audiences process an ad message for the brand. High brand involvement leads to a "central route" to persuasion in that the ad recipients carefully examine and process those ad message elements that they believe are central to a meaningful and logic evaluation of the brand (e.g., brand attribute information). By contrast, low brand involvement induces a "peripheral route" to persuasion whereby recipients evaluate the brand based on superficial analysis of readily available and salient cues in the ad regardless of whether these cues are meaningfully related to the brand (e.g., background music). These ELM predictions have received support in several studies (see Petty, Unnava, and Strathman 1991 for a review). Also, several models of advertising effects have suggested processing differences between high and low involvement audiences similar to the ELM (Batra and Ray 1985; Greenwald and Leavitt 1984; MacInnis and Jaworsky 1989; Mitchell 1986).

AN IMPLICIT ASSUMPTION IN THE ELM

Extensive critiques of the ELM on conceptual as well as methodological grounds have been reported in the persuasion literature (MacInnis and Jaworsky 1989; Areni and Lutz 1988; Miniard and Dickson 1988; Andrews 1988). However, one implicit but important assumption that forms the basis of the ELM framework has been virtually overlooked by researchers in this area. Specifically, it is assumed in the ELM that all audiences, be they involved or uninvolved with an advertised brand, form brand evaluations on-line, i.e., while they are exposed to the ad message. Stated differently, while the ELM hypothesizes different persuasion routes for involved versus uninvolved audiences, it implicitly assumes that persuasion occurs during ad exposure under both routes.

We should note here that we do not view the assumption of on-line persuasion as a general limitation of the ELM model. Rather, the assumption limits the persuasion contexts in which predictions based on the ELM can be reasonably expected to hold. The ELM says nothing about persuasion process that occur sometime after exposure to stimulus information, nor does it predict differences in on-line versus delayed (i.e., memory-based) persuasion. Predictions based on the ELM are only germane to situations where audiences form brand evaluations during exposure to advertising messages.

Implications for Empirical Tests of the ELM

The above discussion suggests that a compelling test of ELM predictions can only be conducted in contexts where on-line brand evaluation processes are occurring. This can be achieved in a laboratory study simply by giving subjects a brand-evaluation goal, i.e., requiring them to evaluate the advertised brand on-line. However, a careful examination of the processing goals and/or orienting instructions given to subjects in ELM studies shows that a brand evaluation goal is almost never explicitly given. Goals typically given subject include evaluating the sound quality of audio messages (Petty, Cacioppo and Goldman 1981; Petty and Cacioppo 1981), evaluating or forming an impression of the speaker (Chaiken 1980, experiment #2; Petty and Cacioppo 1984), evaluating the ad (Petty, Cacioppo, and Schumann 1983), viewing the ad in a natural manner (Celsi and Olson 1988), general comprehension (Chaiken 1980, experiment #1), evaluating the background program (Batra and Ray 1985) etc.. The intent of these instructions is likely to mask the true purpose of the study. However, an undesirable side effect of such instructions may be to inhibit or even discourage subjects from engaging in brand evaluation processing during message exposure. Furthermore, these studies do not even provide any (post-hoc) evidence to suggest that the obtained persuasion effects occurred on-line.

Failure to enforce a brand evaluation goal or otherwise provide evidence supporting on-line brand evaluation clearly opens the possibility that the quality of empirical tests of the ELM may be compromised. Furthermore, we believed that this may be a particularly serious problem for prediction concerning uninvolved audiences. An individual who is involved with an advertised brand will likely spontaneously evaluate the brand on-line because forming a brand evaluation is relevant to either his/her short-term goals or his/her enduring interest with the brand/product category. However, uninvolved audiences do not see the brand as personally relevant, and may therefore refrain from evaluating the brand because such an evaluation serves no objective. Indeed, there is considerable evidence in the literature on memory-based judgement and evaluation to show that individuals who do not have an evaluation objective when they are exposed to stimulus information do not spontaneously form evaluations on-line (Hastie and Park 1986; Lichtenstein and Srull 1985, 1987; Wyer and Srull 1989).

The preceding discussion suggests an alternative explanation for effects due to peripheral ad cues (e.g., an attractive source) on brand evaluation for uninvolved audiences that are obtained in empirical tests of the ELM. Rather than a peripheral cue serving as a low effort heuristic for brand evaluation on-line, the cue may influence brand evaluation later (when the evaluation is measured) in a memory-based manner. Since peripheral cues used in ELM studies are usually salient and vivid (e.g., pictures of celebrities in Petty, Cacioppo, and Schumann 1983), theses cues are likely more accessible in long term memory than are brand message arguments. As a consequence, peripheral cues could have strong effects on uninvolved audiences regardless of whether these effects were on-line or memory based. Thus, peripheral cue effects for uninvolved audiences do not constitute a compelling test of the ELM since they potentially confound on-line and memory-based persuasion processes.

In sum, there is an implicit assumption in the ELM which needs to be empirically validated: persuasion occurs during ad message exposure regardless of which route (central or peripheral) is followed. We report on a pilot study and a main experiment which evaluated this critical assumption. In both studies subjects were explicitly instructed to evaluate the advertised brand during ad exposure. The pilot study examined the effect of involvement on response times for on-line brand evaluations and on recall. The main study attempted a replication of typical ELM tests to see if results found in previous studies could be obtained even when all subjects were given an explicit brand evaluation goal.

PILOT STUDY

As we noted earlier, The ELM suggests that involved audiences form evaluations that are carefully reasoned (central processing), whereas uninvolved audiences form evaluations that are relatively superficial (peripheral processing). Since central processing involves a more careful assessment of information in an ad, response times for generating brand evaluations should be longer for involved audiences than for uninvolved audiences. Also, involved audiences should recall ad information better than uninvolved audiences. These possibilities were examined in this study.

Procedure

Ball point pen was used as the target product. Print ads for fictitious brands of six different products including the target product were created for the study. The ad for the pen contained a simple headline, a picture of the product, and ten verbal product claims.

Fifty four subjects participated in the study. Upon arrival, subjects were seated in individual computer booths and were told that their task would be to evaluate each advertised product in the booklet they were about to receive. After a short computer practice session, they were assigned to either high or low involvement for the pen brand. Involvement was manipulated via two instructions given to subjects just before they examined the ads in the booklet. First, subjects were informed that during a subsequent session they would be asked to choose one of several brands of a product, and that an ad for one of these brands was in the booklet they were about to examine. Second, subjects were told that they would win a cash prize if their evaluation score for this same brand was similar to the overall rating given to the brand by Consumer Reports. The product for the choice task and the cash prize was varied: it was the ball point pen for the high involvement condition and a different product for the low involvement condition (see Celsi and Olson 1988; Maheswaran and Sternthal 1990 for a similar procedure).

Subjects examined the ads in the booklet one at a time at their own pace, and reported their brand evaluation on a 9-point scale (very poor-very good) by pressing the appropriate number on their keyboard. The computer automatically timed each evaluation. This procedure was carefully controlled to minimize any extraneous effects on the response times.

Next, after a five minute distraction task, subjects responded to three measures of "felt" involvement (Celsi and Olson 1988) on 9-point scales (not at all interested-very much interested in the product; not at all attentive-very much attentive to information; not at all concerned-very much concerned about the accuracy of the evaluation). Since these measures were highly inter-correlated (alpha=.82), the responses were averaged to construct a composite measure of felt involvement. Finally, subjects were asked to recall the original information for the pen brand. Recalled items were scored by two independent judges and the inter-judge agreement was 88.8%. Disagreements were resolved through discussion.

Results

Subjects in the high involvement conditions reported a higher level of felt involvement than those in the low involvement conditions (M=6.32 versus 5.19, F(1,53)=5.51, p<.05), indicating that the involvement manipulation was successful. An ANOVA on response times for evaluations revealed a significant effect for involvement (F=6.32, p<.05). As expected, involved subjects took longer to evaluate the brand than did uninvolved subjects (M=35.3 versus 26.7 seconds). Furthermore, involved subjects recalled more than did uninvolved subjects (M=3.41 versus 2.63, (F(1,26)=4.19, p<.10). In sum, the response time and recall data suggest that involved subjects processed product information more carefully in making on-line evaluations than uninvolved subjects.

MAIN STUDY

A key prediction of the ELM is that "central" message cues (e.g., message quality) have a stronger effect on involved audiences, while "peripheral" cues (e.g., source credibility) are more impactful on uninvolved audiences. We tested these predictions under conditions where subjects were clearly forming on-line brand evaluations. Results consistent with the ELM predictions should provide stronger support for the ELM than has been achieved in previous tests.

Method

Subjects and Design. A total of 144 male and female undergraduate students participated in the experiment for course credit. Of these, 138 provided usable responses. The design was a 2 (high/low involvement) by 2 (strong/weak ad message) by 2 (high/moderate source credibility) factorial.

Procedure. Upon arrival, subjects were randomly assigned to one of the eight experimental conditions. All subjects received a folder containing: (a) an introduction booklet, which described the general purpose of the study, experimental tasks, and lottery procedures designed to manipulate involvement, (b) an ad booklet, which contained eight mock ads including an ad for a running shoe brand (target ad), and (c) a dependent measure booklet.

The introduction booklet informed subjects that their task was to examine several product ads at their own pace, and to form an overall evaluation of each of the advertised brands. In addition, subjects were asked to indicate this overall evaluation on response scales that were provided on a separate sheet. Thus, all subjects were given a brand evaluation goal, and these evaluations were measured on-line, i.e., while the pertinent ads were in front of the subjects. Next, after a brief distracting task, subjects were asked to complete the dependent measure booklet. The entire experimental procedure took about 30 minutes, and concluded with a debriefing of subjects.

Manipulation of Involvement. Involvement was manipulated through the use of lotteries. The introduction booklet informed subjects that several lotteries, each for a different product category, would be run as a compensation for their participation, and that they had been randomly assigned to one of these lotteries. Subjects in the high involvement condition were told that they would participate in a lottery for the target product (running shoes). In contrast, subjects in the low involvement condition were informed that they would participate in a lottery for another product (boombox). All subjects were told that lottery winners would be chosen at random after the experiment, and would be allowed to select their preferred brand from among those available within the product category. To bolster the involvement manipulation, subjects were told that they would see an ad for one brand of the lottery product during the study, and that this brand would be one of those available in the lottery.

Manipulation of Message Quality. A variety of message arguments for running shoes were pretested in terms of their convincingness and persuasive strength. The results of the pretest were used to create a strong and a weak version of the running shoe ad. Both versions contained eight arguments, and were approximately equal in length. However, the strong ad version contained relatively compelling and persuasive claims about the advertised brand, while the weak ad version contained relatively uncompelling and vacuous brand claims.

Manipulation of Source Credibility. The running shoe ad contained a headline featuring a personal testimonial for the advertised brand. In the high credibility condition, the endorser was described as a special columnist for the Runner's World magazine. In the moderate credibility condition, the endorser was introduced as a political consultant.

Dependent Measures. Subjects' evaluation of the running shoe brand was measured on two 9-point scales (not at all-very likable; very unsatisfactory-very satisfactory). Since the two measures were highly intercorrelated (r=.81), responses were averaged to construct a composite measure. Other dependent variables included subjects' ratings of message argument quality and source credibility, unaided recall of ad message content, self-reported levels of felt involvement with the advertised brand, and (enduring) product category involvement. As an instruction check, subjects were asked to recall the product lottery to which they had been assigned.

Results

Manipulation Checks. To assess the effectiveness of the message quality manipulation, subjects were asked to rate the brand information in the running shoe ad on two 9-point scales (not at all-very convincing; very weak-very strong). A three-way ANOVA on the average of these ratings (which were highly correlated, r=.86) yielded only a significant main effect for the message quality manipulation (F=44.82, p<.001). As expected, subjects who had received the strong version of the ad rated the message quality stronger (M=5.79) than those who had received the weak version (M=3.46).

As a check on the source credibility manipulation, subjects were asked to rate the person who recommended the running shoe brand on two 9-point scales (not at all-very credible; not at all-very knowledgeable). A three-way ANOVA on the average of these ratings (which were highly correlated, r=.87) yielded significant main effects for the source credibility manipulation (F=72.92, p<.001) as well as the message quality manipulation (F=10.17, p<.01). As expected, subjects in the high credibility condition rated the source as more credible than did subjects in the low credibility condition (M=5.79 versus M=3.50). Also, subjects who had received the strong version of the ad rated the source as more credible than did subjects who had received the weak version (M=5.34 versus M=4.30). Importantly, none of the interactions were significant (F<1). In short, the source credibility manipulation was successful.

The effectiveness of the involvement manipulation was assessed in two ways. First, we examined whether or not subjects correctly recalled the product lottery to which they had been assigned. Only one subject incorrectly recalled the lottery. This suggests that the lottery manipulation successfully influenced personal relevance (i.e., subjects know whether or not they would be making a short term decision regarding the running shoe brand). Note, however, that the ultimate goal of this manipulation was to influence the degree to which subjects actually felt involved with the brand message during exposure to the ad. Therefore, as a more direct check, we examined the effects of the lottery manipulation on three 9-point scales designed to measure how involved subjects actually felt, and how attentively and carefully subjects processed the running shoe brand message (paying a little-paying a lot of attention; not at all carefully-very carefully read; not at all-very involved). A three-way ANOVA on the average of these scales (average r=.66) did not yield any significant main or interactive effects. Contrary to expectations, the main effect due to the involvement manipulation was not significant (M=5.63 versus M=5.23, F=2.04, p>.10), although the means were in the expected direction. Thus, subjects in the high involvement condition did seem to recognize the personal relevance of the advertised running shoe, but did not actually feel more involved, or process brand message information more intensely than did subjects in the low involvement conditions.

Tests of the ELM. Failure to successfully manipulate involvement limits our ability to conduct strong tests of the ELM. Since subjects in the high/low involvement conditions did not differ on intensity and degree of brand message processing, we would not expect differential effects of the message quality and source credibility manipulations on these subjects. This proved to be the case. A three-way ANOVA (involvement by message quality by source credibility) on overall evaluations revealed significant main effects due to the source credibility manipulation (F=8.16, p<.01) and the message quality manipulation (F=52.71, p<.001), but no significant interactions. The high credibility source led to more positive evaluations than the moderate credibility source (M=5.47 versus 4.59), while the strong quality message led to more positive evaluations than the weak quality message (M=6.19 versus 3.89). Importantly, neither the involvement by source credibility interaction nor the involvement by message quality interaction was significant (F=1.73, p<.19, and F<1, respectively). Note that while these results are consistent with the ELM, they do not provide compelling support since they require acceptance of null hypotheses.

Two sets of analysis were done in an attempt to create somewhat stronger tests for the ELM than those reported above. First, a median split on our measure of felt involvement (based on the involvement manipulation check measures) was used to create two groups that differed significantly on actual levels of involvement experienced during the ad viewing episode, and all analyses were conducted using this blocking factor. Second, a median split on our measure of product class involvement (i.e., involvement with running shoes in general) was used in all analyses. These two approaches generated similar results. To conserve space, we report only on the results based on the felt involvement blocking factor. As a result of the median split, 69 (68) subjects were assigned to the high (low) felt involvement conditions. These groups differed significantly on reported felt involvement (M=6.75 versus 3.98, F=323.99, p<.0001).

A three-way ANOVA on overall evaluations as a function of felt involvement (blocking factor), message quality, and source credibility revealed significant main effects for the source credibility and message quality manipulations (F=8.99, p<.005; F=55.45, p<.0001, respectively). More importantly, and as hypothesized in the ELM, these effects were qualified by significant involvement by source credibility and involvement by message quality interactions. The involvement by source credibility interaction (F=7.76, p<.01) indicated that source credibility effects on evaluations were different for high versus low involvement conditions (see Table 1).

TABLE 1

BRAND EVALUATIONS AS A FUNCTION OF INVOLVEMENT AND CREDIBILITY

TABLE 2

BRAND EVALUATIONS AS A FUNCTION OF INVOLVEMENT AND MESSAGE QUALITY

Planned comparisons revealed that the highly credible source produced more positive evaluations than did the source of moderate credibility when felt involvement was low (F=20.89, p<.001) but not when it was high (F<1). This result is consistent with the ELM.

The involvement x message quality interaction (F=6.14, p<.02) indicated that message quality effects on evaluations also differed across the two involvement conditions (see Table 2).

Planned comparisons revealed that although message quality affected evaluations under both high involvement (F=41.56, p<.001) and low involvement (F= 15.29, p<.001), the impact of message quality was much greater in the high involvement conditions (R2=.39) than in the low involvement conditions (R2=.15). This result is also consistent with the ELM.

DISCUSSION

The ELM predicts the process mediators of persuasion only in on-line persuasion contexts. Unfortunately, previous tests have failed to provide compelling evidence for on-line persuasion, and hence have not generated the strongest possible tests of the ELM. We sought to achieve such a test by giving all our subjects clear instructions to engage in on-line brand evaluations. Also, we measured these evaluations as they were formed (rather than after subjects had viewed all our ads) to ensure that subjects followed our instructions. Results from our two studies provide support for the ELM. Specifically, reaction times for evaluations and recall of information were greater for involved subjects than for uninvolved subjects. Also, message quality had a stronger effect on brand attitude for involved subjects while source credibility had a stronger effect on brand attitudes for uninvolved subjects.

There is, of course, one important caveat to our results, and that concerns our failure to manipulate involvement in the main study. We used a (lottery) procedure similar to one that has been successfully used in past research (Celsi and Olson 1988). Our subjects seemed to recognize the implications of the lottery, but did not process product information with differential intensities as a result.

One possibility for why the involvement manipulation failed may be that subjects' involvement with the product class (running shoes) had a strong impact on how involved they felt with the running shoe ad (felt involvement), and this overpowered our lottery manipulation. This argument is bolstered by the fact that our measures of product class involvement and felt involvement were significantly correlated (r=.45), and both measures yielded similar results when introduced as blocking variables in our analyses. Interestingly, Celsi and Olson (1988) also found that the lottery manipulation produced much weaker (although significant) effects on felt involvement than did product class involvement in their study.

In sum, while our results are consistent with the ELM, they do not provide unambiguous evidence for the casual influence of involvement on on-line evaluation processes. Constructive replications that create successful manipulations of involvement are needed to produce such evidence. Also, future research should examine the effects of brand involvement on the processing goals that subjects adopt during ad exposure. Finally, the effects of "central" and "peripheral" cues on persuasion that is on-line versus memory-based, and mediators of these effects are important and unresolved issues that should be investigated.

REFERENCES

Andrews, J. C. (1988), "Motivation, Ability, and Opportunity to Process Information: Conceptual and Experimental Manipulation Issues," in Advances in Consumer Research, 15, M. J. Houston, ed., Association for Consumer Research, 197-203.

Areni, C. S. and R. J. Lutz (1988), "The Role of Argument Quality in the Elaboration Likelihood Model," in Advances in Consumer Research, 15, M.J. Houston, ed., Association for Consumer Research, 197-203.

Batra, R and M. L. Ray (1985), "How Advertising Works at Contact," in Psychological processes and Advertising Effects: Theory, Research, and Applications, eds., L. Alwitt and A. A. Mitchell, Hillsdale, NJ: Lawrence Erlbaum, 13-43.

Celsi, R. L. and J. C. Olson (1988), "The Role of Involvement in Attention and Comprehension Processes," Journal of Consumer Research, 15 (2), 210-224.

Chaiken, S. (1980), "Heuristic versus Systematic Information processing and the Use of Source versus Message Cues in Persuasion," Journal of Personality and Social Psychology, 39 (5), 752-766.

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

Hastie, R. and B. Park (1986), "The Relationship Between Memory and Judgment Depends on Whether the Judgment Task is Memory-Based or On-Line," Psychological Review, 93 (July), 258-268.

Lichtenstein, M. and T. K. Srull (1985), "Conceptual and Methodological Issues in Examining the Relationship Between Consumer Memory and Judgment," in Psychological Processes and Advertising Effects: Theory, Research and Applications, eds., L. Alwitt and A. A. Mitchell, Hillsdale, NJ: Lawrence Earlbaum, 113-128.

Lichtenstein, M. and T. K. Srull (1987), "Processing Objectives as a Determinant of the Relationship Between Recall and Judgment," Journal of Experimental Social Psychology, 23 (March), 93-118.

MacInnis, D. J. and B. J. Jawroski (1989), "Information Processing from Advertisements: Toward an Integrative Framework," Journal of Marketing, 53 (October), 1-23.

Maheswaran, D. and B. Sternthal (1990), "The Effects of Knowledge, Motivation, and Type of Message on Ad Processing and Product Judgments," Journal of Consumer Research, 17 (June), 66-73.

Miniard, P. W., P. R. Dickson, and K. R. Lord (1988), "Some Central and Peripheral Thoughts on Routes to Persuasion," in Advances in Consumer Research, 15, M. J. Houston, ed., Association for Consumer Research, 197-203.

Mitchell, A. A. (1986), "Theoretical and Methodological Issues in Developing an Individual- Level Model of Advertising Effects," in Advertising and Consumer Psychology, eds., J. C. Olson and K. Sentis, 3, 172-196.

Petty, R. E. and J. T. Cacioppo (1981), "Issue Involvement as a Moderator of the Effects on Attitude of Advertising Content and Context," in Advances in Consumer Research, 8, K. B. Monroe, ed., Association for Consumer Research, 20-24.

Petty, R. E. and J. T. Cacioppo (1983), "Central and Peripheral Routes to Persuasion: Application to Advertising," in Advertising and Consumer Psychology, L. Percy and A. Woodside, eds., Lexington, MA: Heath, 3-23.

Petty, R. E. and J. T. Cacioppo (1986), "The Elaboration Likelihood Model of Persuasion," in Advances in Experimental Social Psychology, ed., L. Berkowitz, 19, Orlando, FL: Academic Press, 123-205.

Petty, R. E., J. T. Cacioppo, and R. Goldman (1981), "Personal Involvement as a Determinant of Argument-Based Persuasion," Journal of Personality and Social Psychology, 41, 847-855.

Petty, R. E., J. T. Cacioppo, and D. M. Schumann (1983), "Central and Peripheral Routes to Advertising Effectiveness: The Moderating Role of Involvement," Journal of Consumer Research, 10 (September), 135-146.

Petty, R. E., R. Unnava, and A. J. Strathman (1991), in Handbook of Consumer Behavior, eds., T. S. Robertson and H. H. Kassarjian, Englewood Cliffs, NJ: Prentice-Hall, 1-49.

Wyer, R. S. Jr. and T. K. Srull (1989), Memory and Cognition in Its Social Context, Hillsdale, NJ: Erlbaum.

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