Applying Conjoint Analysis to Social Advertisements

Moshe Engelberg, Stanford University
Rosalind M. Pierson, Stanford University
Hiroshi Kashio, Stanford University
ABSTRACT - A pilot field study (N=30) explored the feasibility of using conjoint analysis to examine important attributes of mass-mediated messages. Conjoint analysis is a technique typically used for measuring consumer trade-offs among multi-attributed products or services; however, little research has been conducted as to whether this type of analysis is appropriate for the study of more complex stimuli, like social advertisements. Using a fractional factorial design, 16 advertisements (copy only) promoting recycling were developed, each comprised of three factors: spokesperson, message/setting, and sponsor/format. Results revealed that the message/setting factor was most important to respondents, with a relative importance of 65.4%. Congruency between setting and message was highly valued. Spokesperson was the second most important factor at 31%; however, the most motivating spokesperson varied depending on the audience segmentation strategy used. Sponsor and format was the least important message attribute, with a relative importance of only 4.1%. Overall, conjoint analysis did an adequate job of depicting both individual and group preferences for particular message attributes. These results lead us to conclude that conjoint methodology is an interesting, creative and effective way to gather information not only about traditional products and services, but for designing advertisements as well. Research and managerial implications of extending conjoint analysis to the domain of advertisements are discussed.
[ to cite ]:
Moshe Engelberg, Rosalind M. Pierson, and Hiroshi Kashio (1992) ,"Applying Conjoint Analysis to Social Advertisements", in NA - Advances in Consumer Research Volume 19, eds. John F. Sherry, Jr. and Brian Sternthal, Provo, UT : Association for Consumer Research, Pages: 696-705.

Advances in Consumer Research Volume 19, 1992      Pages 696-705

APPLYING CONJOINT ANALYSIS TO SOCIAL ADVERTISEMENTS

Moshe Engelberg, Stanford University

Rosalind M. Pierson, Stanford University

Hiroshi Kashio, Stanford University

[The authors gratefully acknowledge the comments of V. Srinivasan.]

ABSTRACT -

A pilot field study (N=30) explored the feasibility of using conjoint analysis to examine important attributes of mass-mediated messages. Conjoint analysis is a technique typically used for measuring consumer trade-offs among multi-attributed products or services; however, little research has been conducted as to whether this type of analysis is appropriate for the study of more complex stimuli, like social advertisements. Using a fractional factorial design, 16 advertisements (copy only) promoting recycling were developed, each comprised of three factors: spokesperson, message/setting, and sponsor/format. Results revealed that the message/setting factor was most important to respondents, with a relative importance of 65.4%. Congruency between setting and message was highly valued. Spokesperson was the second most important factor at 31%; however, the most motivating spokesperson varied depending on the audience segmentation strategy used. Sponsor and format was the least important message attribute, with a relative importance of only 4.1%. Overall, conjoint analysis did an adequate job of depicting both individual and group preferences for particular message attributes. These results lead us to conclude that conjoint methodology is an interesting, creative and effective way to gather information not only about traditional products and services, but for designing advertisements as well. Research and managerial implications of extending conjoint analysis to the domain of advertisements are discussed.

BACKGROUND

The purpose of this pilot study was to determine whether conjoint analysis methodology would be useful in the development of television advertisements, an area in which it has received minimal application. Conjoint analysis is a set of techniques typically used for measuring consumer trade-offs among multi-attributed products or services (Green & Srinivasan, 1990). It decomposes overall judgements about a set of complex alternatives such as brand name, price and various product features into utility scales which reflect the relative importance of various attributes (Green & Wind, 1975). This study used social advertisements as stimulus material to provide new ground for the application of conjoint analysis techniques, while investigating conditions under which message elements common to most television advertising influenced motivation to comply with the behavioral request in the advertisement.

THEORY

Social Advertisements

Advertisements for consumer goods and social concerns each constitute a class of messages which compete within and across classes for public attention and support. Social advertisements include those that deal with issues like pollution, population growth, homelessness, AIDS, unemployment, and recycling (Flora & Maibach, 1990). The use of social advertisements in this study heeds the call for attention to what has been called "the dark side of consumer behavior" (Hirschman, 1991), and more generically reflects an aspect of social marketing (Kotler & Roberto, 1989).

In terms of message design considerations, these two classes of messages are similar on some dimensions and different on others (Slater & Flora, in press). The similarities apply to most television advertising; e.g. a short, unsolicited message imbedded in selected programming, choices around what is said, who says it, and the context and manner in which it is said. Ads for social issues and consumer goods often differ in terms of objectives, style of appeal, portrayed benefits, motives of message source, and production value. Some distinctions between these two classes of ads can blur, as in a campaign for beverages built around the product=s environmentally sound packaging. In this study, we investigated message elements that can be found in many classes of ads, though the relative utilities of these elements may be a function of the class of ad.

Elements of Advertisements

A first step towards building research-based design strategies for advertisements is to assess the relative influences of various message attributes. The tradition of involvement research in consumer behavior and communication suggests that salient features of advertisements can be conceptualized in terms of central or peripheral cues, which in turn affect the way information is processed and used (e.g. Petty & Cacioppo, 1986).

Research concerned with effects of persuasive messages often divides message attributes into source factors, message factors, and context factors (O=Keefe, 1991). Source factors may include the presenting source (spokesperson), and the sponsoring source. Message factors encompass a multitude of both content and structural characteristics. Context factors can include the viewing situation, type of programming surrounding the ad, and the packaging of the ad. Typically, attention to message content reflects a central route to persuasion, while attention to source and context cues lead to a peripheral route to persuasion (Petty & Cacioppo, 1986).

Application of Conjoint Analysis to Advertisements

This study conceptualizes the advertisement as the product, and uses conjoint analysis to determine consumers= utilities for various message components. By doing so, we are assuming both theoretically and methodologically that ads consist of reasonably independent and additive components, and that an ad=s overall utility can be decomposed and represented by the sum of the utilities for its separate components. We do not claim that the ad elements investigated in this study represent the gestalt of the ad, any more than the factors in any conjoint study represent everything about the product or service under examination. Rather, our intent is to decompose ads by looking at reasonably independent and additive elements which have received previous empirical attention, albeit with different methods, and are critical in terms of message design considerations.

The well-documented role of involvement in message effects (Petty, Cacioppo, & Schumann, 1983; Salmon, 1986) led us to focus on motivation, an involvement construct, as the criterion variable. This reflects our belief that for campaign managers and ad agencies to make maximum use of research concerned with effects of involvement, some research needs to focus on when and how various message elements generate involvement. Thus, we asked respondents to rank the advertisement scenarios in terms of how motivating they were, realizing that ad liking may figure into their judgement processes.

METHODS

Overview

Respondents were told that the Stanford Recycling Project (hypothetical) needed their input to design a motivating television spot aimed at increasing recycling among young adults in the Bay Area. They then rated and ranked sixteen profiles of such spots, which had different combinations of spokesperson, message and setting, sponsor and format. Concepts, procedures, and stimulus materials were thoroughly pretested, and results were cross-validated with self-explicated rankings.

Sample

A total of thirty Stanford graduate and undergraduate students participated in this study. Demographically, our sample was 57% male, 63% graduate students, and 87% single. Additionally, our sample represented 14 majors, which we collapsed into two categories known at Stanford University as "fuzzies" and "techies?. "Fuzzies" (53%) are students who are primarily liberal arts majors, whereas "techies" (40%) are students who are classified as science or engineering majors (7% were undecided).

Experimental Design

We utilized a three factor fractional factorial design (Addelman, 1962) with four levels of each factor. Source, message, and context factors were represented.

Two source factors were manipulated; the spokesperson and the ad sponsor. Spokespersons were chosen to be highly credible or not, and highly likable or not. Ad sponsor was intended to represent for-profit motives or public service motives. Ad sponsor was used in combination with the format of the message, a context factor, that is whether it was described as an advertisement or public service announcement. Selection of message factors drew on framing principles (Tversky & Kahneman, 1986) by using a positive message (gain), or mathematically equivalent negative message (loss), in combination with a positive or negative setting. The positive and negative settings were intended to qualitatively represent a gain and loss respectively. This led to the construction of 16 advertisement profiles, as follows:

FACTOR                   LEVELS

Spokesperson             Bill Cosby, actor and philanthropist

                                 William K. Reilly, head of EPA

                                  Bo Jackson, professional athlete

                                  Ron Smith, Bay Area graduate student

Message and               Positive message, pristine lake

setting                         Positive message, dirty landfill

                                 Negative message, pristine lake

                                 Negative message, dirty landfill

Sponsor and               Danville Corporation, PSA

format                        Danville Corporation, advertisement

                                 Recycling Cooperative, PSA

                                 Recycling Cooperative,advertisement

A complete description of the messages and settings can be found in Table 1.

The message/setting factor and the sponsor/format factor are both combinations of two two-level factors. Making each into a four level factor allowed us to examine interactions visually, and with minimum constraints in our model, as well as avoiding the inflation of relative factor importance due to unequal number of levels. A sample profile is illustrated below.

Spokesperson:  Bill Cosby, Entertainer & Philanthropist

Setting: Beautiful, pristine mountain lake with gentle breeze blowing.

Message: (with disdain) This year alone, two-thirds of our glass, aluminum, and newspapers were dumped in landfills. We are destroying our planet now and it may not survive past this generation. Save our planet...recycle!

This advertisement is sponsored by the Danville Corporation

Respondents first read a cover sheet which briefly described the scenario of our recycling message design project, and their task. Because the profiles included a substantial amount of information, direct rankings of sixteen profiles would have been too difficult. Therefore, respondents first rated the cards on a seven point Likert scale (1=not at all motivating to 7=extremely motivating), and then ranked each stack of cards.

TABLE 1

SETTING AND MESSAGE DESCRIPTIONS

Respondents were given four practice cards to rate and rank, before working with the actual profiles. After completing the primary task, respondents completed a questionnaire with items on demographics, recycling involvement and behavior, and television use. Lastly, respondents provided self-explicated ratings of factor importance (1=not at all important to 7=extremely important) for use as cross-validation data.

Pretesting

Because application of conjoint analysis to messages is relatively unexplored, we conducted three levels of pretesting:

1) During the concept development phase of this project, we ran a focus group whose feedback led us to include both positive and negative messages and settings, and suggested certain levels for our other factors.

2) 21 students rated 20 potential spokespersons on the dimensions of credibility and likability, which led us to select one spokesperson for each possible combination of high or low credibility crossed with high or low likability.

3) Six pretest subjects went through our whole experiment, which allowed us to identify procedural and methodological problems.

RESULTS

Validity and Reliability of Data

Rank order results were analyzed using the LINMAP program (Srinivasan & Shocker, 1973). Overall, validity of results across respondents was fairly consistent, with all respondents having less than 10% of pairs violated. Multiple regression analysis was used to validate the LINMAP analysis of the ranked data, and to compare results obtained from the rankings data and ratings data. In both cases, regression analysis and LINMAP results were consistent. Self-explicated ratings of factor importance corroborated the relative importance percentages obtained through LINMAP.

Aggregate Results

Results for the average part worth utilities for the entire sample are shown in Figure 1. The combination setting/message factor was clearly the most important factor across respondents with a relative importance of 65%. Messages that had congruent setting and message were the most motivating. Respondents most preferred the negative setting/negative message combination, followed by the positive setting/positive message combination. Respondents least preferred the negative setting/positive message combination.

Spokesperson was the next most important attribute with a relative importance of 31%. Bill Cosby was the most preferred spokesperson, followed by William K. Reilly, head of the Environmental Protection Agency (EPA). Bo Jackson came in third; thus Bo may be more adept at promoting "sneakers?, than motivating people to recycle. Ron Smith, a hypothetical Bay Area graduate student, was the least motivating spokesperson.

FIGURE 1

AVERAGE PART WORTHS BY FACTOR

Not surprisingly, sponsor/format came in as least important, with a relative importance of 4.1%. During the debriefing phase of our study, many respondents stated that they either failed to really notice the sponsor/format, or that it only played a role when they went back and ranked profiles within a stack. The first attribution suggests the need for counterbalancing order of factors. However, we purposely put sponsor/format at the bottom of the profile, as these features are typically not explicitly germaine to most social advertisements.

The following analyses illustrate that the patterns described above generally held up for the message/setting factor and the spokesperson factor, across segments for all four segmentation strategies. The lack of consistent patterns on the sponsor/ format factor is interesting in its own right. These results are organized by factor in Figures 2 - 4. Figure 5 displays relative factor importance across segmentation strategies.

FIGURE 2

SEGMENTATION ANALYSES BY PART WORTHS: SPOKESPERSON FACTOR

Demographic and Psychographic Segmentation Analyses

Segmentation analysis was performed to examine whether meaningful differences between groups existed for each factor and level. Our findings revealed the following pertinent information.

1) Males versus Females: Results from the gender segments indicated that males found the spokesperson to be more important than did females (42% vs 19%). Also, males chose Bill Cosby as their preferred spokesperson, whereas females selected William K. Reilly.

2) "Fuzzies" versus "Techies?: Techies were more sensitive to spokesperson than were fuzzies (48% vs 22%). Fuzzies preferred William K. Reilly and techies preferred Bill Cosby. Fuzzies were more sensitive to the setting/message factor than techies (71% vs 40%). Also, fuzzies most preferred the negative setting/message combination, whereas techies equally preferred the two congruent setting/message duos.

FIGURE 3

SEGMENTATION ANALYSES BY PART WORTHS: SETTING/MESSAGE FACTOR

3) High versus Low Involvement: Highly involved respondents were those who reported extreme concern about recycling efforts. Their most preferred factor was setting/message (69%) with less involved respondents finding this factor a bit less motivating (60%). Less involved respondents were more sensitive to the spokesperson (35% vs. 28%), with the spokesperson of choice being Bill Cosby. William K. Reilly was most preferred by the high involvement group.

Segmentation Based on Cluster Analysis

Respondents were also segmented empirically, based on results from cluster analysis. Two, three, and four cluster solutions were examined. We selected the three cluster solution because its distinctions were richer and more interpretable. However, upon further review, we discarded the third of the three segments because of the lack of cohesion among the three cases which comprised the cluster, and retained the two larger segments for final analysis.

Key differences between the segments were: The sponsor/format factor was approximately twice as important to Segment 1 as to Segment 2, with Segment 1 preferring the Recycling Cooperative as sponsor, and Segment 2 preferring the Danville Corporation. Preference for the negative setting/message combination was much stronger for Segment 2 versus Segment 1. Segment 1 found Bill Cosby most preferable, while Segment 2 selected William K. Reilly.

FIGURE 4

SEGMENTATION ANALYSES BY PART WORTHS: SPONSOR/FORMAT FACTOR

To help us better understand the two segments derived from cluster analysis, regression analysis was used, with demographics, involvement and television use as predictors of group membership. (This technique is mathematically equivalent to discriminant analysis, when the criterion variable is two-category nominal). Segment 1 was characterized by respondents who were undergraduates (p>.01), fuzzies (p=.02), and less involved (p=.11), while Segment 2 tended to be more involved, graduate students, and techies.

DISCUSSION

Summary

This study was an attempt to assess the feasibility of using conjoint analysis to examine complex stimuli such as social advertisements. Overall, subjects reacted positively to the conjoint task and had little trouble understanding or completing the task. Our results revealed that both individual and group part-worth functions adequately depicted individual and group preferences. For example, although both males and females believed the setting/message combination was the most important factor, a high percentage of males reported that the spokesperson was also fairly important.

These results have important implications for message design issues, like spokesperson importance and selection, in relation to audience segmentation strategies. Individuals who reported being highly involved in the recycling issue, were less likely to care about the spokesperson than individuals who reported low involvement in the issue, as is consistent with the elaboration likelihood model (Petty & Cacioppo, 1986). Furthermore, the spokesperson of choice for highly involved individuals was the little known expert William K. Reilly, compared to the popular Bill Cosby for the low involvement group.

FIGURE 5

% IMPORTANCE BY SEGMENTS

Finally, although it is clear that the setting and tone of a message plays an important role in how motivating it is perceived to be, it seems that other factors, like spokesperson, and to a smaller extent sponsor and format, can either contribute to or detract from the message, depending on the audience being targeted.

Both the results gathered in this study and our own intuitive insights indicate that conjoint analysis is an interesting, creative, and effective way to gather information which can aid the development of advertised messages. If the technique is applicable in the more abstract domain of social issues, it seems likely that it will be at least as useful for advertisements for consumer products.

In the future, we hope to develop actual messages based on results of conjoint analysis and test them with audiences similar to the research participants. This will allow us to determine the conditions under which consumers can essentially design their own messages, which we believe to be an important extension of the fundamental marketing concept of consumer sovereignty (Blaug, 1990).

Limitations

Our extension of conjoint to a new domain of inquiry, mass media messages, is limited by a number of methodological considerations. The validity of this study rests on our presumption that advertisements can be decomposed into reasonably independent constituent parts in order to justify use of an additive main effects model, as well as the notion that individuals can predict their response to those constituent parts (see Kahneman & Snell, 1990 for a discussion of predicting utility). However, we do not presume that these processes would remain valid under all conditions, e.g. investigating preferences for structural features like cuts and zooms through stimulus material consisting of written copy only.

The use of written copy only here accurately reflects our preference for experimental control over true message complexity at this stage in the game. We would expect this preference to gradually shift if the marriage of conjoint to messages garners further attention.

Finally, the lack of a holdout sample, lack of covariates in group level analyses, and only one order of factor presentation limit validity and generalizability, and should be addressed in subsequent work.

Recommendations

Recommendations for future research applying conjoint analysis techniques to the study of messages in general, are as follows:

1) Larger scale studies should be implemented in order to gather more reliable and statistically significant data.

2) Conjoint techniques should be applied to other advertising design issues, both in the domains of social issues and consumer goods, in order to more fully test the applicability of this research methodology to mass media messages.

3) In further testing of this technique on advertisements, it may be worthwhile to actually develop storyboards or audio-visual profiles that consumers could actually view. This way, consumer preferences would not need to be inferred from evaluations of written descriptions of messages, thus leading to more accurate and meaningful results.

4) Before any commercial study of this type is undertaken, the actual cost/benefit of the study should be accurately assessed to determine if the initial outlay of resources is worth the time, energy and money. Agreeing on what constitutes "benefit" should be an early consideration.

REFERENCES

Addelman, S. (1962), Orthogonal main-effect plans for asymmetrical factorial experiments. Technometrics, 4, 21-46.

Blaug, M. (1990), The History of Economic Thought. Brookfield, VT: E. Elgar Publishing.

Dominick, J. (1990), The Dynamics of Mass Communication. New York: McGraw-Hill.

Flora, J. & Maibach, E. (1989), Cognitive responses to AIDS information. Communication Research, 17, 759-774.

Green, P. & Srinivasan, V. (1990), Conjoint analysis in marketing research: New developments and directions. Journal of Marketing, 54, 3-19.

Green, P. & Wind, Y. (1975), New ways to measure consumer judgements. Harvard Business Review, 53, 107-17.

Hirschman, E. (1991), Secular mortality and the dark side of consumer behavior: Or how semiotics saved my life. Advances in Consumer Research, 18, 1-4.

Kahneman, D. & Snell, J. (1990), Predicting utility. In R. Hogarth (ed.), Insights in Decision-Making: A Tribute to Hillel J. Einhorn, Chicago, University of Chicago Press, 295-310.

Kotler, P. & Roberto, E. (1989), Social Marketing: Strategies for Changing Public Behavior. New York: The Free Press.

O=Keefe, D. (1990), Persuasion: Theory and Research. Beverly Hills, CA: Sage.

Petty, R. & Cacioppo, J. (1986), Communication and Persuasion: Central and Peripheral Routes to Attitude Change. New York: Springer-Verlag.

Petty, R., Cacioppo, J., & Schumann, D. (1983), Central and peripheral routes to advertising effectiveness: The moderating role of involvement. Journal of Consumer Research, 10, 135-146.

Salmon, C. (1986), Perspectives on involvement in consumer and communication research. In B. Dervin & M. Voight (eds.), Progress in Communication Sciences, Norwood, NJ: Ablex, 243-269.

Slater, M. & Flora, J. (in press), Is health behavior consumer behavior? Health behavior determinants, audience segmentation, and designing media health campaigns. In E. Clark, T. Brock, & D. Stewart (eds.), Advertising and Consumer Psychology, Hillsdale, NJ: Erlbaum.

Srinivasan, V. & Shocker, A. (1973), Linear programming techniques for multidimensional analysis of preferences. Psychometrika, 38, 337-69.

Tversky, A. & Kahneman, D. (1986), Rational choice and the framing of decisions. The Journal of Business, 59, 251-277.

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