Studies in Imagery, Styles of Processeing, and Parallel Processing Need Realtime Response Measures

G. David Hughes, University of North Carolina
[ to cite ]:
G. David Hughes (1990) ,"Studies in Imagery, Styles of Processeing, and Parallel Processing Need Realtime Response Measures", in NA - Advances in Consumer Research Volume 17, eds. Marvin E. Goldberg, Gerald Gorn, and Richard W. Pollay, Provo, UT : Association for Consumer Research, Pages: 461-466.

Advances in Consumer Research Volume 17, 1990      Pages 461-466

STUDIES IN IMAGERY, STYLES OF PROCESSEING, AND PARALLEL PROCESSING NEED REALTIME RESPONSE MEASURES

G. David Hughes, University of North Carolina

The three papers in this session provide stimulating contributions to our knowledge of consumer information processing. Ways that each paper could be strengthened are discussed. But these papers have the same limitation: they focused on a static concept, recalled, processed information, not the dynamics of information processing. An example from realtime response research illustrates how these limitations can be overcome.

IMAGERY EFFECTS ON BEHAVIORAL INTENTIONS

Bone and Ellen (1990) make a genuine contribution to the literature in their finding that the degree of self imagery determines attitude toward a radio advertisement, which, in turn, determines subjects' behavioral intentions (BI) regarding the purchase of the advertised hypothetical brand of popcorn. The original hypotheses are well grounded in the current literature to explore the question, "Why should imagined sensory experiences result in more positive BI?" The availability/valence hypothesis and the semantic network theory are very relevant. Their design would have been strengthened, however, by including Heider's balance theory in the semantic network theory (Hughes 1971). It would have provided a very testable model for the examination of attitudes toward self, attitudes toward advertisements, and attitudes toward a brand.

They state their hypotheses clearly and use appropriate rigor in their statistical analysis. But, between these two points, there is considerable ambiguity regarding their working definitions and their measurement procedures, as will be detailed below.

When the authors found that their treatments did not induce the expected imagery effects, they abandoned their causal design and sorted subjects into groups according to their degree of imagery. They were careful to limit their conclusions to those of correlation, not causation.

At this point, however, it would have been helpful to know why individuals differed in their ability to create mental images. Do the authors have any demographic or psychographic profile data that would reveal insightful significant differences between high and low imagers? Perhaps this distinction is a matter of right hemisphere or left hemisphere information processing. The concepts of visual and verbal processors (Gould 1990) could contribute to an investigation of high and low imagers. Or perhaps the low imagers watch more television and have never developed imaging skills. An older group of subjects, raised on radio, may have higher imaging skills. Knowing more about people with low and high degrees of imaging skills is a prerequisite to applying the authors' findings to advertising strategies.

AMBIGUITIES IN DEFINITIONS AND MEASUREMENTS

Sorting out the definitions, variables, scaling techniques, and the hypothesized relationships was difficult. A flow diagram would have been very helpful. The original design was modified halfway into the paper with the addition of the concept of attitudes toward the advertisement. This addition was made without the benefit of relevant citations, so it is not clear why it was added. Some of the ambiguities and recommendations for improvements follow.

Message Elements. The authors examined the impact of three message elements--self relatedness, plausibility, and distinctiveness--on the imagery-behavioral intentions relationships. Self relatedness (not to be confused with self congruity) deals with whether the main character in the imagined scene is the consumer him/herself or someone else. Advertisements that build on self imagery were found to create more positive behavioral intentions (BI).

Plausibility is defined as an imagined scenario that is consistent with the consumer's lifestyle. This concept is different from believability, which is frequently used to test television advertisements. It would be useful to do a study to see if subjects understand the difference and if the difference is relevant to communications strategies .

Distinctiveness is defined as the imagined scenes' uniqueness, which lies between the outer bounds of mundane and bizarre. This definition overlaps their definition of plausibility, and caused problems. These two concepts were then combined into a single treatment, with insignificant results. Because the similar concepts of believability and comparative advantage have produced significant results in commercial testing, it is unfortunate that these concepts were combined. Perhaps a definition of distinctiveness that required a brand choice would have yielded significant results. The authors' design gave no choices of brand, only whether the subject would buy the test brand or not. An information processing design would measure before-and-after changes in brand choice.

Independent Variables Image processing was measured by first asking subjects if they experienced imagery. If they did, they then answered two semantic differential questions dealing with ease and vividness of the imagery. They also answered five Likert-type scales on the difficulty of imagining. No reasons were given for using two scaling techniques.

Attitude toward the ad was the sum of seven items--good, happy, cheerful, pleased, amused, soothed, and warm--that had been measured along a four-point scale. Even though some of these items have been studied by other researchers, no citations were given in support of their inclusion. It would have been useful to see how these affect dimensions and their cognitive dimension of plausibility were linked to imagery, to attitude toward the ad, and then to behavioral intentions. The findings could support or refute the arguments that cognition does not need to precede affect (e.g, Zajonc, 1980).

Self relatedness was measured with an openended question that asked who the main character was in their imagery. Plausibility used two seven-point scales that asked if the event could happen. Distinctiveness was the sum of two scales that asked if the event was ordinary or special. No reasons were given for the variety in scaling techniques. One can only wonder if the variations in measuring techniques contributed to the insignificant findings regarding plausibility and distinctiveness.

Behavioral intentions were the sum of three standardized measures. One was an eleven-point scale of the probability of purchasing the advertised brand. The second scale used a five-point likelihood of purchasing. The third scale used a seven-point Likert-type scale in response to "The next time I purchase popcorn, I will buy the brand..." No reason for the three scales or their summation was given. It is not clear why attitudes toward purchasing the brand were measured, how these differed from behavioral intentions, and where they fit into the analysis.

The authors could enhance their contribution to the literature by reanalyzing their data with more precisely defined variables. Instead of combining measures that used various scales, they should examine them independently so that they can recommend to other researchers the most appropriate scales and then use them across all concepts. Using a uniform scaling technique can facilitate the selection of statistical methods and reduce respondent confusion.

Because plausibility and distinctiveness were combined in their design, these concepts cannot be separated. But the authors did measure plausibility and distinctiveness for manipulation checks. Perhaps the data from the subjects who were not in the plausibility and distinctiveness cells could be analyzed with these manipulation check variables to determine if future research would benefit from a sharper definition of these two concepts.

The authors could enhance their contribution first by explaining why they used so many different scaling techniques and then by developing more parsimonious measurement methods to facilitate replication by them and other researchers. Such replications are necessary for a field of study to claim that it is a science.

But none of these recommendations overcomes the fact that this research is not measuring consumer information processing. It is, at best, measuring recalled information that was processed. And recall is subject to errors in memory, variations in attending to the stimuli, and a problem that has been called backward filtration, which occurs when a dominant stimulus appears at the end of the stream of stimuli (Hughes, 1990). This late, dominant stimulus overshadows earlier stimuli. (This effect should not be confused with recency effects.) Finally, subjects' attempts to recall information processes can be influenced by the experimental design either by interrupting the process with a questionnaire or revealing the purpose of the study. The latter could have occurred in this study because the subjects completed imagery, distinctiveness, plausibility, and self-related manipulation checks before completing the behavioral intentions scales. Perhaps the subjects guessed the intent of the study.

To measure consumer information processing we must have methods for realtime response research. A procedure with examples will be shown briefly at the end of this paper.

STYLE OF INFORMATION PROCESSING

Gould (1990) explores the relationships between styles of information processing and involvement with selected products, shopping visualization, and self-consciousness. While Gould's design does not assure directions of causality, the findings provide new insights into processed consumer information and suggest ways to answer questions raised in the Bone and Ellen paper.

For his independent variable Gould used the mental imagery literature and an existing scale that identified verbal and visual processors. He assigned subjects to one of four cells--low processors (low verbal and low visual), high verbals (high verbal and low visual), high visuals (low verbal and high visual), and high processors (high verbals and high visuals). The median scores were used to assign the 138 subjects to one of the cells. The median values and the number of subjects were not reported, which makes it difficult to replicate the study.

The dependent measures included involvement with seven products (television, radio, magazines, books, clothes, cameras, and music), shopping visualization (making shopping lists and visualizing what will be bought), and self consciousness (private-public self consciousness and social anxiety). Gould used existing measuring instruments for these variables.

Because the hypotheses are not stated, the results and discussion are guided by the relationships that were significant. Consequently, the combined "Discussion and Conclusions" section is very difficult to follow and sometimes is not well supported by the findings. But the results are interesting. For example, high verbals and high processors report being less involved with television. Perhaps these individuals like to create their own images. Bone and Ellen may want to add a visual-verbal processing measure to help explain why some persons image and some do not.

There is considerable face validity in the findings. High verbals were more involved in books. High visuals and high processors were more likely to visualize or plan a shopping trip. High visuals and high processors also may tend more to visualize themselves when they focus on their self concept. While the support for this conclusion is not clear, it could be relevant to the conclusions of Bone and Ellen that self-related imagery creates a more positive behavior intention. The finding that women tend to be high processors could be relevant to copy and media strategies.

This study can also be criticized for not being a study of information processing, but a study of recalled processed information.

PARALLEL PROCESSING

Martin and Kiecker (1990) step back from the details of information processing research and conclude that current approaches assume that the consumer thinks like an old computer, in a serial or sequential fashion, while evidence from neurobiology suggests that the human brain is a massive parallel processor. Evidence of parallel processing requires new definitions of memory structures and knowledge representation, which, in turn, require new means for measuring consumer information processing.

Parallel Distributed Processing (PDP) models stress the concept that information processing amounts to changes in connections between networks of neurons. Information will change the weight and/or sign of these connections as well as create new connections. Thus, information is stored in the form of connections throughout the network, not at a single location to be addressed by a central processor. This model helps to explain how a consumer can combine previous experience and new external information to reach a decision. The authors conclude that information processing research that uses methods such as protocol analysis, information monitoring, and eye movement, restrict the subject to the limited serial task.

Those of us who have been dissatisfied with the present information processing research would like to consider testing parallel processing models, but the paper is limited in its suggestions for implementing the concept. It refers to a study on vision, one on analysis of response times, and one on reading research, all of which assumed a parallel model. It suggests displaying multiple attributes of various brands on a video terminal for a short duration and then asking for a purchase decision. This seems to be a modern variation of threshold research that used a tachistoscope. This design could be implemented very easily with realtime response research technology. It could be done on individuals and groups, which would make it easier and cheaper than the eye movement design that is proposed by the authors. In realtime response research the computer would measure each individual subject's response time, which would be a proxy for the amount of processing that was done.

REALTIME RESPONSE RESEARCH

These three papers share a common need for realtime measures while the subjects are processing the information. The first realtime system was developed at CBS in the mid 1930s to measure audience reactions to programs. It consisted basically of a rheostat for each subject. When dialed, these rheostats controlled pens over a continuous paper strip, like an EKG. Within the last few years new developments in microcomputer and video technology have made it possible to have portable systems that measure responses in a fraction of-a second and superimpose the results over the stimulus videotape. This discussant has developed four generations of these systems since 1984 and has conducted hundreds of realtime research projects. His second and third generation response devices were a seven-position dial that had many limitations (Hughes, 1990). The fourth generation subject response device has a 1-100 dial, a full alpha numeric keypad, and a screen that gives subjects immediate feedback as they dial. The system is called SpeedBack(sm). An example will illustrate how realtime measurements can improve on the research that has been reported in this session.

Bone and Ellen had difficulty in defining and measuring variables that reduced to affect and cognitions. Can subjects distinguish between these concepts? Figure 1 shows the results of a small test that was conducted just for this session.

Twelve MBA students watched six television commercials twice. The first time they dialed for affect along a scale of unfavorable(0)favorable(100). The second time they dialed for useless(0)-useful(100). The California Raisin commercial that featured Ray Charles and used the claymation technique went above 80 in favorability, but dropped to 30 in uselessness. The next commercial, "Living Well," promoted a St. Patrick's Day Special at a local health club. The responses were straight down for both measures. (The company went bankrupt.) Sudafed, a decongestant medicine, hardly moved off neutral (50) for favorability, but ended on a strong up trend in usefulness when clinical proof was mentioned. The First Response Pregnancy Test had a downward trend in favorability, was slightly above neutral in usefulness at the midpoint, but ended on a sharply negative trend at the end. The Pontiac and local Ford dealer advertisements were downward sloping on both dimensions, but the usefulness curve was higher because both ads discussed discounts and some of the subjects were planning to buy a car in the near future.

Each of these patterns can provide useful insights-for generating strategies. We are studying the relationships between these types of patterns and successful advertisements to answer questions such as, "Should an advertisement end on an upward trend?" We have noted that the most successful advertisements are not always rated favorably, but they do need to have useful or believable information. When these patterns were played back, the subjects had reasonable explanations. For example, the Sudafed commercial ended with a reference to clinical studies showing that it was effective in 30 minutes, so the useful information measure went upward. Some subjects were about to buy a car, so they found information about discounts useful even though they did not like the commercial.

FIGURE 1

FAVORABILITY VS. USEFULNESS (12 MBAs)

Because we measure in realtime, and in very fine and fast measures, we are able to see patterns of individual styles of processing. We see several different response patterns. We call some subjects "chunkers" because they wait for a chunk of information before making a decision, while the "averagers" respond to every stimulus when it comes. An example of each style of processing appears in Figure 2. The upper curve for the California Raisin commercial is the averager who made small incremental changes with each new stimulus. The chunker happened to go negative, so that is the lower curve. We do not presently know why these two types occur. In some cases we have noted that people who have no need for the product advertised chunk negative quickly and then go flat. Thus, they seem to tune out the commercial. We plan to examine levels of attending in future tests.

We are just beginning to study the individual differences between these two response styles. We will certainly consider the verbal and visual scales that Gould used.

The Martin and Kiecker paper stimulates us to examine the time between the stimulus and the subject's response to see if we can identify parallel processing. This will be a very easy task with the SpeedBack(sm) system. The response time of the system is very fast because of the smooth, continuous 1-100 dial and the sampling rate of 1/5 of a second. Realtime response research would also make their information display board design a very simple task.

Figures 1 and 2 illustrate another advantage of fast, realtime response research. It is an excellent tool for small group diagnostic studies. A researcher can start with small groups and, if the results are stable, add additional groups, combine data, and use analytical techniques that require a larger sample. We have replicated the shapes of the curves in these figures with other groups of 12 to 20 subjects .

Realtime response research is like clinical studies in medicine in that it should be used early in the research process. It is not instant polling or a substitute for survey research (Hughes 1990). Conversely, in clinical research the subject becomes his/her own control over time, so we know that differences are due to the stimuli. The fine and fast measures allow us to detect subtle, small differences, such as a single word. We have found that we can get stable results with sample sizes as small as 40 subjects.

FIGURE 2

PROCESSING STYLES--AVERAGER/CHUNKER

Persons who are concerned about small samples for diagnostic research should refer to articles in the medical journals. During the period from October, 1987, to October, 1989, there were 35 citations in the Drug Newsletter that used a single test group of fewer than 20 subjects. Eleven had 5 or fewer subjects, 11 had 6 to 10 subjects, 7 had 11 to 15 subjects, and 6 had 16 to 20 subjects. Included in these publications were prestigious journals such as the British Medical Journal, the Journal of Clinical Psychiatry, and the New England Journal of Medicine

In conclusion, while there are many ways that these authors could refine their research, we must acknowledge that theirs were exploratory studies. I hope that these comments will help them to further analyze the data from these studies and to develop more precise approaches in future studies. Realtime research measures can help them in that regard. Their ideas and findings will stimulate new directions for those of us who are researching consumer information processing.

REFERENCES

Bone, Paula F. and Pam S. Ellen (1990), 'The Effect of Imagery Processing and Imagery Content on Behavioral Intentions," in press.

Gould, Stephen J. (1990), "Style of Information Processing Differences In Relation to Products, Shopping, and Self-Consciousness," in press.

Hughes, G. David (1971), Attitude Measurement for Marketing Strategies, Glenview, IL: Scott, Foresman and Company, pp.50-56.

Hughes, G. David (1990), " Diagnosing Communications Problems with Continuous Measures of Subjects' Responses: Applications, Limitations, and Future Research," Current Issues and Research in Advertising, in press.

Martin, Dawne and Pamela Kiecker (1990), "Parallel Processing Models of Consumer Information Processing: Their Impact on Consumer Research Methods," in press.

Zajonc, R. B. (1980), "Feeling and Thinking: Preferences Need No Inferences," American Psychologist. 35: 151-175.

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