When Does Source Credibility Exert It’S Influence on Beliefs: At Encoding Or At Retrieval?

ABSTRACT - This study was designed to investigate the impact of source credibility on the believability of an advertising claim. While earlier research had suggested that source credibility influences truth ratings, it was not clear when the credible source exerts its influenceBis it when one is exposed to an advertising claim (at encoding), or, when it is subsequently remembered (at retrieval), or, at both times? Results show that while the credibility of source at encoding as well as the source remembered at retrieval influenced truth ratings of a claim, the source at retrieval influenced truth ratings the most. Marketing implications of the results are discussed.

Citation:

Sharmistha Law (1999) ,"When Does Source Credibility Exert It’S Influence on Beliefs: At Encoding Or At Retrieval?", in E - European Advances in Consumer Research Volume 4, eds. Bernard Dubois, Tina M. Lowrey, and L. J. Shrum, Marc Vanhuele, Provo, UT : Association for Consumer Research, Pages: 45-50.

European Advances in Consumer Research Volume 4, 1999      Pages 45-50

WHEN DOES SOURCE CREDIBILITY EXERT IT’S INFLUENCE ON BELIEFS: AT ENCODING OR AT RETRIEVAL?

Sharmistha Law, University of Toronto at Scarborough, Canada

[This research was supported by funding from the Social Sciences and Humanities Research Council of Canada.]

ABSTRACT -

This study was designed to investigate the impact of source credibility on the believability of an advertising claim. While earlier research had suggested that source credibility influences truth ratings, it was not clear when the credible source exerts its influenceBis it when one is exposed to an advertising claim (at encoding), or, when it is subsequently remembered (at retrieval), or, at both times? Results show that while the credibility of source at encoding as well as the source remembered at retrieval influenced truth ratings of a claim, the source at retrieval influenced truth ratings the most. Marketing implications of the results are discussed.

INTRODUCTION

Consider the following statement: "Peanut butter is low in cholesterol." Do you believe it? Past research in marketing has demonstrated that if you have seen this item before this article you are likely to believe it more this time, regardless of whether it is true or not. Mere repetition increases belief (the truth effect). People are naturally prdisposed to believe that which is familiar and familiarity increases with repetition (Hawkins & Hoch, 1992).

If this does seem familiarBwhere did you hear it before? As a consumer, one could have learned this "fact" from numerous sources: a radio or television commercial, a close friend, a research report, a nutrition expert. Let’s say this item is from an ad by Kraft. If you remember this source correctly, then your natural skepticism regarding commercially motivated claims may temper your belief in this claim. On the other hand, if you misattribute this claim to a dietitian friend you hold in high regard, then one would expect that your beliefs in the claim would be higher. Investigating the mechanism which underlies the effect of source credibility on consumer judgments is of considerable contemporary interest because marketers routinely spend millions of dollars in garnering the endorsement of credible spokespeople (e.g., Nike and Gatorade rely on Michael Jordan; Candice Bergen sells Sprint’s long-distance phone service).

Furthermore, there is a history in the marketing literature of examining source effects (McCracken, 1989; Petty & Cacioppo, 1986; Petty & Cacioppo, 1981; Sternthal et al., 1978) and this study builds on and extends this body of knowledge. For instance, the impact of the personal characteristics of a source (such as, his or her credibility, expertise, trustworthiness, attractiveness) on the persuasiveness of a communication has been examined by a substantial number of researchers in social and cognitive psychology as well as in marketing (Homer & Kahle, 1990; Hovland, Janis & Kelley, 1953; Hovland & Weiss, 1951; Petty & Cacioppo, 1986; Pratkanis, Greenwald, Ronis, Leippe & Baumgardner, 1986). Reviews have summarized the impact of source credibility on immediate attitude change (Andersen & Clevenger, 1963), on delayed measures of persuasion (Pratkanis et al., 1986), and on the interaction of source characteristics with other variables such as the timing of source identification and level of personal involvement with the communicator on subsequent attitudes (e.g., Petty & Cacioppo, 1986; Sternthal, Phillips & Dholakia, 1978). In general, it has been found that a credible source facilitates change in consumer evaluations. What is not clear from previous studies, however, is when source credibility influences rated truth. Is the influence of source credibility on beliefs exerted at the time when the message is encoded? Or, is the effect of the source on the credibility of the message exerted when the message and its source are retrieved just prior to making a decision?

The next section reviews evidence from the truth effect literature concerning the relationship between source memory and beliefs. The article then discusses the rationale and hypothesis of the present study. Following this, the design, procedure, and results obtained are discussed.

Source memory and beliefs: A brief review

Subjects’ memory for the credibility of a message source influences truth rating. Several studies have found that claims presented during the study but attributed to an extra-experimental source received the highest truth ratings (Arkes et al., 1991; Arkes et al., 1989, Experiments 1 and 1A; Law, Hawkins, & Craik, 1998; Sitton & Griffin, 1980). For instance, Arkes et al. (1989) report that the rated truth of repeated claims when identified as seen in the experimental session was 3.87 (on a 7 point scale where 7 was "definitely true"), whereas those attributed to a source outside the study were rated as 5.03 (an increase of 1.16 scale points). Presumably, subjects considered the extra-experimental source(s) to be more trustworthy and hence trusted claims attributed to outside sources more than claims attributed to the experimental setting.

Results reported by Law (1998) are also consistent with the view that a spokesperson’s credibility influences truth ratings and also raises the possibility that source credibility has greatest impact at the time the truth judgment is being made. Subjects saw adertising claims at study, some paired with credible sources (such as Better Business Bureau, Consumer Reports) others paired with doubtful sources (such as National Enquirer, Commercial messages). After a brief distractor task, the studied claims were repeated together with similar, new claims and subjects were asked to (a) assess the truth of each claim, and (b) indicate whether they recognize having encountered the claim before. For the claims they recognized, subjects were required to circle the source of prior exposure from a list which had two broad options, First session and Outside of Experiment (e.g., commercial material; word-of-mouth; personal experience; News/ research reports). The primary objective was to examine the impact on truth ratings when the experimentally provided source was forgotten, and when it was misremembered. It was observed that subjects’ memory for the source of a claim had a significant impact on truth ratings even when those memories were false. Not only did participants frequently commit false memory errors (i.e., they reported having seen an ad claim before, outsideBwhich was not possible since the claims were fictitious and created for experimental purposes), but they also showed a tendency to trust these false judgments the most.

These data seem to converge on the notion that source remembered when making a judgment is an important determinant of truth ratings. However, because source memory was tested by a general "seen in first session/seen outside experiment" measure, it is not clear from the Law (1998) results whether differences in truth ratings were a result of poor learning at encoding (when claims were paired with credible or doubtful source) or result from the subjects’ inability to remember source information at retrieval. In order to more fairly evaluate memory mechanisms underlying truth ratings, source should be manipulated at exposure and memory for both the claim, and its source must be evaluated at retrieval.

Thus, the issue of when source memory exerts its influence on consumer beliefs is investigated here within a more general framework of memory which involves the processes of encoding and retrieval.

Based on a popular framework of memory (Craik & Tulving, 1975; Tulving, 1983; Cole & Houston, 1987; Park, Smith, Dudley & Lafronza, 1989; Park & Hastak, 1994; Rabinowitz, Craik & Ackerman, 1982), one can postulate two ways in which the credibility of a source will affect future decisions. It is possible that subjects make decisions regarding their belief in a statement at the time of encoding. That is, the subject evaluates all relevant factors (plausibility of a claim, credibility of its source) at encoding and tags the claim with a credibility assessment as soon as it is processed at encoding (called the encoding-tag hypothesis). This hypothesis therefore predicts a main effect of spokesperson credibility at encoding on truth ratings and can be formally stated as thus:

H₁:Encoding Tag Hypothesis: Claims presented with a credible source at encoding will be perceived as being more truthful than claims presented with a doubtful source at encoding, regardless of the source to which they are attributed at retrieval.

Alternatively, the assessment of the truthfulness of a claim may occur at the time of retrievalBfor instance, at the time of purchase. In this case, the subject stores the claim and source information, but does not evaluate its credibility since there is no express need to do so. At the time of purchase, or in the case of the experiment, at the time of retrieval, the subject retrieves all information about the claim, the fact and the source, and forms a judgment regarding its credibility. This pattern was obtained by Begg et al. (1992, Experiment 3) and Law (1997) where rated truth was found to be ifluenced solely by subjects’ memory for the credibility of presenter. Hence, the retrieval-evaluation hypothesis predicts that source credibility exerts an influence on beliefs when a task or circumstance requires a truth judgment to be made:

H₂:Retrieval Evaluation Hypothesis: Claims accorded to a credible source at retrieval will be perceived as being more truthful than claims accorded to a doubtful source at encoding, regardless of the source with which they were presented at encoding.

Investigating whether the impact of source credibility on beliefs occurs at the time of learning (encoding) or, whether such evaluations occur when a purchase decision is being made (retrieval), is not only important for theoretical reasons, but also has important marketing implications. If validity assessments are tagged to a message at the time of encoding, then information gleaned from a credible source would be believed, regardless of whether one remembers the source from which the message was learned. In a marketing context, this means that the critical mediator of a consumer’s belief in an ad claim would be her impression of the credibility of the communicator of the message at the time of learning, and it should not matter whether the source of the claim is remembered at point of purchase.

However, consumers do not have perfect memory. They may forget or misattribute the source. If so, validity assessments will depend upon the credibility implications of the source retrieved rather than source presented at encoding. Indeed, if validity assessment is made during retrieval, then we would expect people to believe more in those messages which they (correctly or incorrectly) attribute to a credible source at the time of retrieval. Here, the failure to remember a credible source, or mistakenly assigning it to an incorrect source at retrieval, could incorrectly inflate or deflate the consumers’ belief in a claim. The present study investigates this question by manipulating source credibility at presentation and measuring subjects’ memory for item and source information, and truth ratings of claims at retrieval.

It is important to recognize, however, that the encoding-tag and retrieval-evaluation may not occur exclusive of each other. It is plausible that both effects operat simultaneously, though the numerical size of their effects may differ. It is conceivable that the presented source has a certain quantitative effect, and that this may be added to or discounted depending upon the source to which the statement is accorded at retrieval. Also, in addition to their main effects, the final outcome may depend on an interplay between the two. If this is so, then an investigation of the encoding-retrieval interaction will be necessary for a complete understanding of belief formation.

Participants were presented with a list of claims taken from the same pool as reported by Law, Hawkins and Craik (1998), half of which were attributed to a high credible source while the other half were attributed to a low credible source. After an interval of 20 minutes, participants were again shown the original claims mixed with similar, but new claims and instructed to evaluate how true each claim was, indicate whether they recognized the claim, and indicate which source they remembered the claim to be from.

Several aspects of this experiment should be highlighted. Unlike the majority of investigations of source credibility, this study required participants to evaluate more than one advertising claim. By presenting subjects with multiple claims, the study mimics the everyday consumer context more closely. Furthermore, consumers are frequently confronted with endorsements from credible and dubious sources at once, a situation in which memory errors may have graver consequences. Thus t make the laboratory experiment more reflective of the real-life scenario, each subject saw both credible as well as doubtful sources (i.e. credibility manipulation was within subject).

In addition to manipulating the credibility of a claim’s source at encoding, this study also involved a manipulation of levels of encoding. The reason for including a level of encoding manipulation was twofold: First, the two levels of encoding provide the opportunity of exploring the hypotheses over more than one task condition. This would strengthen the robustness of the results. Second, the shallow encoding condition is expected to result in poorer source memory relative to the deeper encoding condition and thus generate a wider range of retrieval outcomes (i.e., higher incidence of source forgetting and source misattributions). Hence, by providing more than one encoding context and providing a wider range of source retrieval outcomes, this manipulation will allow a more precise investigation of the mechanism through which source memory influences rated truth.

The two levels of encoding were operationalized in the following manner. Different groups of subjects were given either syllable counting or typicality rating instructions prior to the presentation of the claims (discussed under Procedure). According to the levels of processing (LOP) framework, the syllable counting task is data-driven and encourages little elaboration of the to-be-learned material, which represents a shallow encoding task. In contrast, the typicality rating task, by orienting the subjects to the meaning of the claim and hence encouraging elaborative processing of its meaning, constitutes a deeper task (Craik & Lockhart, 1972). Therefore, item and source memory performance was expected to be relatively poor for syllable counting instructions compared to the typicality rating instructions.

In summary, the general hypothesis investigated in the present study was that validity judgements about advertising claims are moderated by memorial processes at encoding and retrieval. This hypothesis was tested first by examining the relative influences of these two memory processes on validity ratings, and second, by manipulating source credibility at exposure (encoding), and third, by examining the relative influences of encoding and retrieval on validity ratings.

METHOD

Subjects. 74 undergraduate students at a large north western university participated in the study in exchange for course credit. Data from two subjects were excluded because they failed to follow the instructions.

Design and Stimuli Lists. The experiment was a 2 (Source at presentation: Credible versus Doubtful) x 2 (Source at retrieval: Credible versus Doubtful) x 2 (Level of processing: Deep versus Shallow) mixed-factorial design, with encoding task (shallow vs. deep level of processing) manipulated between subjects and source credibility (credible vs. doubtful) varied within subjects. There were 36 subjects in the deep encoding condition and 38 in the shallow condition. Credibility was manipulated at presentation by dividing the list of claims into two sets of equal length: for one set, subjects were told that the claims are from research and investigative reports appearing in the Globe and Mail (a Canadian national newspaper) while for the other set, subjects were informed that the claims were reprinted from the National Enquirer.

Pretesting was conducted to ensure that the sources chosen would be sufficiently different in credibility ratings for adequate investigation of the influence of credibility on truth ratings. Sixteen students drawn from the same population as the main experiment participated in the pretest. Each participant was asked to indicate whether they were familiar with each of 14 local and other publications (The Toronto Star, Consumer Reports, Advertising material, Th National Enquirer, etc.) and rate how credible they perceived them to be. Globe and Mail, a Toronto based national newspaper, received the highest ratings (M_Globe=1.27 on a continuous scale from 0 [very credible] to 7 [very doubtful]). In contrast, the National Enquirer, a publication equally familiar to the respondents, received a mean rating of 6.72 on the same scale. This finding was the basis for selecting the sources used in the study and was further confirmed in a post-test survey using the subjects in the main experiment.

The experimental stimuli were 73 statements selected from the list described in Experiments 1 and 2. These 73 items included a set of 64 target items (32 to serve as old claims and 32 to serve as new claims) and 9 items to serve as non-tested buffer items in the study and test lists. Two versions of the study list (List A and List B) were prepared and rotated across groups of subjects such that old items shown to about half of the subjects would be the new items shown to the remaining participants and vice versa.

To create the study list, the 32 "old" target items were randomly divided into two sets of 16 items. For each set of 16 target items, three buffer items were added to the beginning and served as practice items. An additional four obviously true or obviously false filler items were randomly inserted in the appropriate claim set in order to reinforce the contrast between the credible and doubtful blocks (example of obvious fillers: The current federal government of Canada is LiberalBtrue; sitting too long makes you shorterBfalse). Thus, one set of 23 claims was presented as the credible set and the other presented as the doubtful set. As in the case of repetition status, the order of presentation of credibility was balanced across subject groups such that about half of the participants were told that the first block of claims were from the Globe and Mail while the others were told that the first block was culled from the National Enquirer.

The test list contained 67 items, 32 of which were repeated from the study list and were thus "old", 32 of which were "new" claims never seen before, and 3 buffer items were added at the beginning of the list to allow subjects to become familiar with the task.

Procedure. Participants were tested in groups of three to six. At the start of the study phase, subjects were given a rating sheet and informed that they would be asked to evaluate the communication styles of different publications. This was followed by information about the source of the to-be-presented claims (Globe and Mail or National Enquirer), and a description of the task. In the syllable counting condition, subjects were told that their primary task was to indicate the number of syllables in each claim. Here, the purpose of the study was stated as assessing whether some publications had a higher proclivity towards using polysyllabic words. For the typicality condition, subjects were required to judge whether the claim was typical of those seen in the Globe and Mail (National Enquirer). To maintain the cover story the instructions in this condition stressed that the research interest was to assess whether readers perceived differences in the communication styles of different publications. See the Appendix A for the experimental protocols.

The study items were presented via an overhead projector at the rate of a claim every 10 seconds. As discussed, the presentation of the claims was blocked by source credibility (credible vs. doubtful source). After a 20 minute distracter task, subjects were re-presented with the 32 "old" claims together with 32 "new", but similar, fictitious claims. The subjects were informed that half of the claims had not been presented in the study list. Each statement was presented for 12 seconds, and subjects were asked to indicate: (a) how credible each statement was on a Truth scale (a continuous Likert scale, 16 cm. long, and anchored at Definitely False on the left and Definitely True on the right with higher scores reflecting greater validity); (b) whether they had seen the claim before (a forced choice Yes/No: i.e., recognition memory for item); and (c) where they had seen the claim before from a list of sources (Experimental source: Globe and Mail news, National Enquirer; Outside source: Globe and Mail news, Other news/research, National Enquirer, Advertising material, Word of Mouth, Other source; Don’t Remember). See the Appendix for details on dependent measures. Following this, subjects were required to fill out a post-test version of the source rating task. This questionnaire was very similar to the pretest version: Subjects were given a list of 14 sources and asked to indicate how credible they perceived each to be on the 0 to 7 continuous scale with high values indicating low credibility. A debriefing session was conducted at the end to receive input from participants about the instructions and experimental demand effects.

RESULTS

The dependent variable was the truth-rating that the subjects accorded to a given statement. There were several independent factors. The first was the source attached to a statement at presentation, it had two levels: Globe and Mail (credible) vs. National Enquirer (doubtful). The second factor was the source to which the subjects accorded the statements on retrieval. This was, by definition, a post-hoc assignment depending upon the subjects response and could be accorded to: Globe and Mail (credible), National Enquirer (doubtful), a source outside the experiment, or unable to recall a source. A cross between these two factors (encoding and retrieval) provides a total of eight cells, and in principle a subject could provide data for each of these eight cells. The levels of processing manipulation was a between subject manipulation which subsumed the above two factors. For each subject, the average of the truth-ratings for statements in each cell was used as the dependent variable for analysis instead of analyzing ratings for each claim. If the raw statements are taken as the unit of analysis, it leads to more than 4,000 degrees of freedom for the Error sums of squares and does not control for the within-subject nature of the study. This aggregation of data is in keeping with previous research in the truth effect (Hawkins & Hoch, 1992).

TABLE 1

ANALYSIS OF VARIANCE OUTPUT TESTING THE MAIN EFFECTS OF LEVELS OF PROCESSING (LOP), ENCODED SOURCE, RETRIEVED SOURCE, AND THEIR INTERACTIONS ON RATED TRUTH

The results are presented in three sections. First, data regarding manipulation checks and counterbalancing procedures are reported. Second, the results relating to the replication of the truth effect are presented. The final section contains the analysis of the impact of encoding and retrieval of source information on truth ratings.

Counterbalancing of Booklet and Source Credibility. A two-way ANOVA including the booklet type (List A or B) and the order of presentation of sources (credible first vs. doubtful first) on truth ratings showed no significant main effect of study list, F(1,70)=1.33, p>.2, no significant main effect of credibility order, F(1,70)=1.6, p>0.2, and no reliable interaction between the two, F(1,70)=0.61, p>0.4. These results confirm that the process of stimulus and task order randomization had their desired effects. As a result data across booklet types and order of presentation were pooled for further analysis.

Levels of Processing. These instructions had the expected effect in that subjects’ ability to discriminate between old and new claims (as measured by d’) systematically increased as instructions changed from shallow to deep elaboration (M_shallow=1.3 vs. M_deep=1.9). An ANOVA of these data indicated that the benefits of deep encoding instructions on memory were reliable, F(1, 69)=14.9, p<.001, and thus the levels of processing manipulation had its desired effect.

Truth Ratings. As in previous studies, the data were analyzed separatly for actual and perceived repetition. An ANOVA showed that repeated claims were rated higher in perceived truth relative to claims seen for the first time, (M_old=9.9, M_new=9.1, F(1,71)=38.5, p<.001), thus confirming previous results of the positive impact of actual repetition on beliefs. Also in line with previous findings, claims thought to be repeated were given higher truth ratings compared to claims thought to be new, regardless of their true repetition status, (M_{perceived old}=10.8, M_{perceived new}=8.2, F(1,71)=200.0, p<.001).

To assess the impact of LOP on rated truth and the encoding-tag and retrieval-evaluation hypotheses, a repeated measures ANOVA was conducted on the truth ratings for items recognized as being old using LOP, source presented at encoding (credible or doubtful), and the source identified at retrieval (credible or doubtful) as factors. The dependent variable was the average truth rating accorded by each subject to claims in a given condition. Since the encoding manipulation was administered between-subjects, the data were analyzed using a mixed factor model (within-subjects for credibility of encoded and retrieved source, and between subjects for level of encoding). Table 1 gives the appropriate degrees of freedom used to test each main effect and interaction in this analysis, and the respective F values and significance levels obtained.

As revealed in Table 1, there was a significant main effect of LOP on rated truth, F(1, 70)=7.29, p<.01. Inspection of the mean truth ratings showed that shallow processing instructions led to higher belief ratings than deeper processing instructions (M_shallow=10.8 vs. M_deep=9.7). However, the two way interactions between encoding task and source credibility (F(1,68)=0.46, p=0.5) and the three way interaction among encoding task, credibility at study, and retrieved source was not reliable (F(1,58)=0.43, p=.5).

Encoding Tag Hypothesis. The encoding tag hypothesis proposed that source presented at learning will influence subsequent ratings regardless of what source comes to mind at retrieval. From Table 1 it is evident that truth rating of claims showed a significant main effect of encoded source, F(1, 68)=9.69, p<.01. The average truth ratings showed that regardless of subjects’ accurate memory for encoded source, claims actually paired with a credible source were rated more true (10.8) than those paired with a doubtful source at presentation (9.8). This difference in mean ratings corresponds to a partial eta squared of 0.03; and the effect size index (mean difference divided by pooled standard deviation, computed as per Cohen, 1987) reveals that the encoding effect size index is 0.28 which is small according to Cohen.

TABLE 2

EFFECT OF ENCODING AND RETRIEVAL ON TRUTH RATINGS

If subjects assess truthfulness of claims at the time of learning, then claims paired with a credible source at encoding will be more likely to be believed than claims paired with a doubtful source even when subjects fail to remember the source at retrieval. Thus, a comparison of the truth ratings for claims across the two credibility conditions in cases where the claims are correctly recognized as being repeated, but for which subjects do not remember the source, could be viewed as a better test of the encoding-tag hypothesis, since there is no retrieved source that could potentially influence the ratings. A separate analysis of this data (a contrast analysis of the #source forgot’ column presented in Table 2 revealed a significant difference in truth ratings based on whether the claims were initially paired with a credible or doubtful source (M_credible=11.39, M_doubt=7.9, F(1, 71)=19.8, p<.001).

In sum, truth ratings of claims show evidence of being influenced by the credibility of the source presented at encoding. The fact that claims initially paired with a doubtful source were believed less than those presented with a credible source was obtained even in cases where the original source was forgotten at the time of testing attests to the strength ofthe influence of encoding on beliefs.

Retrieval Evaluation Hypothesis. This hypothesis predicted that source remembered at retrieval would influence rated truth. This expectation was confirmed in the repeated measures ANOVA presented in Table 1 shows that the main effect of retrieved source was significant, F(1,67)=67.89, p<.001). Subjects were most skeptical of claims remembered (correctly or incorrectly) as being learned from doubtful sources (9.0) compared with those remembered as being learned from credible sources (11.5), the difference between the two is large (partial eta squared=0.27; effect size as per Cohen is 0.93) and the difference is reliable.

Encoding versus Retrieval. The foregoing results suggest that encoding has an effect on beliefs and so does retrieval. Which of these two is the more important determinant of beliefs? This issue can be answered in several forms. First, numerically the effect size of the source at presentation was smaller than the effect size of the source at retrieval (partial eta squared 0.03 versus 0.27; Cohen’s effect size index 0.28 vs. 0.93).

A more intuitive comparison can be made by referring to data on encoding-retrieval reciprocal mistakes. Comparison of ratings for claims presented with a doubtful source but accorded to a credible source at retrieval (E_D-R_C) with those presented with a credible source but accorded to a doubtful source (E_C-R_D) gives valuable insight about the relative impact of encoding versus retrieval on beliefs. If encoding is the more important determinant of beliefs, then we would expect claims presented with a credible source even if attributed to a doubtful source, would be rated higher than claims presented with a doubtful source but accorded to a credible source (that is, E_C-R_D>E_D-R_C). Conversely, if retrieval has a greater impact on beliefs, then one would expect claims presented with a doubtful source but attributed to a credible source would be rated higher in validity compared to claims presented with a credible but attributed a doubtful source (that is, E_D-R_C>E_C-R_D). The data show that E_D-R_C had a mean value of 11.3 (SD=2.5) versus a E_C-R_D value of 9.6 (SD=2.9).

The comparison of E_D-R_C to E_C-R_D constitutes neither a main effect nor an interaction. It is a comparison of diagonal cells to each other. To get a formal statistical comparison of the two cells, we directly compared them using a one-way ANOVA, with the two cells as two levels of the factor. There is a significant difference between E_D-R_C and E_C-R_D in favor of the former (F(1,61)=13.07, p=0.006). The difference between E_C-R_D and E_D-R_C, which reflects the superiority of retrieval over encoding is a large effect size (effect size index of 0.62 as per Cohen, a partial eta squared=0.18). These results indicate that the retrieved source tends to have a bigger impact on subsequent truth ratings than the presented source. However, it is important to bear in mind that both source at presentation and retrieved source have significant influences on belief.

DISCUSSION

Both encoding and retrieval processes impact beliefs about advertising claims. However, the retrieval evaluation has a more prominent effect. Although one must always be cautious when inferring the nature of mental representation from behavioral responses (Braitenberg, 1984), the results suggest that some kind of validity is tagged on to the statement at encoding, but that this is modified by the source which is retrieved at the time of decision. If the source at retrieval is a credible one, it inflates belief, whereas if the retrieved source is a doubtful one it may negate any benefit of having actually presented the statemen with a credible source.

One has to be cautious whenever extrapolating from results in the laboratory to real life purchasing situations. But, two limitations of our model should be particularly highlighted. First, the interval between encoding and retrieval was of the order of about twenty minutes. In real life scenarios it is usually much longer. Since there are suggestions in the marketing literature that with increasing duration there may be a qualitative change in consumers evaluative responses (sleeper effect studies), one must be careful when extrapolating from these data. Second, while the subjects had no indication regarding the true purpose of the experiment at the time of encoding, a reasonable person may fathom the purpose of the experiment from the dependent measures at retrieval. Thus, there is the potential that the bigger effect size accorded at retrieval may be influenced by the demand effects of the experiment. However, a post-experimental questionnaire asking for subjects’ guesses regarding the hypotheses showed that only four of seventy two subjects could say that the study had something to do with source credibility and belief. Besides, withholding these subjects from the analyses did not produce any significant changes in our conclusions. Hence, demand effects do not appear to be a major concern in this experiment.

In the marketing context these findings suggest that while marketers should do their best to design advertising copy and campaigns to ensure that consumers associate their chosen endorser with their product, they should also expend resources to enhance the accurate retrieval of the endorser at the point of purchase. In other words, the fee marketers pay to their spokesperson will have the greatest facilitatory impact on consumer beliefs when (a) the same spokesperson is repeatedly paired with the same product information, and (b) when the advertising copy and identity of the product’s endorser is reinstated at the point of purchase.

REFERENCES

Andersen, K., & Clevenger, T. (1963). A summary of experimental research in ethos, Speech Monographs (pp. 59-78). New York.

Arkes, H. R., Boehm, L. E., & Xu, G. (1991). Determinants of Judged Validity. Journal of Experimental Social Psychology, 27, 576-605.

Arkes, H. R., Hackett, C., & Boehm, L. (1989). The generality of the relation between familiarity and judged validity. Journal of Behavioral Decision Making, 2, 81-94.

Begg, I. M., Anas, A., & Farinacci, S. (1992). Dissociation of processes in belief: Source recollection, statement familiarity, and the illusion of truth. Journal of Experimental Psychology: General, 121(4), 446-458.

Cohen, J. (1987). Statistical Power Analysis for the Behavioral Sciences. (Revised ed.). Hillsdale, NJ: Lawrence Erlbaum Associates.

Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684.

Craik, F. I. M., & Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General, 104(September), 268-294.

Hawkins, S. A., & Hoch, S. J. (1992). Low-involvement learning: Memory without evaluation. Journal of Consumer Research, 19, 212-225.

Homer, P. M., & Kahle, L. R. (1990). Source expertise, time of source identification, and involvement in persuasion: An elaborative processing perspective. Journal of Advertising, 19(1), 30-39.

Hovland, C., & Weiss, W. (1951). The influence of credibility on communication effectiveness.Public Opinion Quarterly, 15, 635-650.

Law, Sharmistha (1998), Do we believe what we remember, or do we remember what we believe? Advances in Consumer Research, v. 25, 221-225.

Law, Sharmistha, Scott A. Hawkins, and Fergus I. M. Craik (1998), Repetition induced belief in the elderly: Implications for rehabilitating memory deficits, Journal of Consumer Research, v. 25, 91-107.

McCracken, G. (1989). Who is the celebrity endorser? Cultural foundations of the endorsement process. Journal of Consumer Research, 16, 310-321.

Park, D. C., Smith, A., D., Dudley, W. N., & Lafronza, V. N. (1989). Effects of age and a divided attention task presented during encoding and retrieval on memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15(6), 1185-1191.

Park, J.-W., & Hastak, M. (1994). Memory-based product judgments: Effects of involvement at encoding and retrieval. Journal of Consumer Research, 21(December), 534-547.

Petty, R. E., & Cacioppo, J. (1986). The elaboration likelihood model of persuasion. In L. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 19, pp. 123-205). New York: Academic Press.

Petty, R. E., & Cacioppo, J. T. (1981). Attitudes and Persuasion: Classic and Contemporary Approaches. Dubuque: Brown.

Pratkanis, A. R., Greenwald, A. G., Ronis, D. L., Leippe, M. R., & Baumgardner, M. H. (1986). Consumer-product and sociopolitical messages for use in studies of persuasion. Personality and Social Psychology Bulletin, 12(4), 536-538.

Rabinowitz, J. C., Craik, F. I. M., & Ackerman, B. P. (1982). A processing resource account of age differences in recall. Canadian Journal of Psychology, 36(2), 325-344.

Sitton, S., & Griffin, S. (1980). The sleeper effect in reconstructive memory. The Journal of General Psychology, 103, 21-25.

Sternthal, B., Phillips, L., & Dholakia, R. (1978). The persuasive effect of source credibility: A situational analysis. Public Opinion Quarterly, 42, 285-314.

Tulving, E. (1983). Elements of Episodic Memory. Oxford: Oxford University Press.

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