Can discounts look even better?
An experimental extension of the scope of the Group
Attractiveness effect
Daan C. van Houten 10352635 Master’s Thesis Graduate School of Communication, University of Amsterdam Master’s programme Communication Science, Persuasive Communication Supervised by dr. Stephan Winter
I would like to thank dr. Stephan Winter for sharing his enthusiasm for the subject and his support in supervising my thesis.
Mijn dank gaat uit naar mam en pap, voor hun onvoorwaardelijke steun en voor de vrijheid om een uitgebreide studie te genieten, met als sluitsteen deze scriptie.
TABLE OF CONTENTS Abstract 3 Introduction 4 Theoretical framework 5 Method 16 Results 21 Discussion 24 References 29 Appendix 32
ABSTRACT
This study was an investigation of the question whether a group attractiveness effect (GA-effect) occurs when evaluating discount items. The GA-effect refers to the empirically observed phenomenon that a group of people is perceived as more attractive than the average attractiveness of its individual members. Of multiple proposed explanations for the effect, selective attention has the most convincing theoretical and empirical support and is therefore the interpretation most focused on in this study. This thesis is an attempt to extend these findings to another type of stimulus instead of humans: discount items. This way, it is attempted to break into new territory for marketing communication, discovering new ways to improve the perception of discount items. The study also extends the scholarship on group attractiveness, as until now it has only been researched with regard to human physical attractiveness. However, the proposed underlying mechanism, selective attention, can also be applied to other stimuli. An experiment was carried out comparing the evaluations of discount items presented in a group (either rated as a whole or one by one) and presented separately. In addition, the effect of a goal prime to direct attention was investigated. No significant differences were found between the grouping and priming conditions. It is concluded that the GA-effect does not carry over as easily to discount items. It is proposed this might be due to the comparative nature of discount evaluation, which cancels out the attention bias occurring when judging human faces.
INTRODUCTION
Among the wide range of sales techniques consumers encounter daily, perhaps the most familiar one is offering discounts. Discount items (e.g. the weekly selection on sale in supermarkets) are intended to attract customers, and therefore it is important to make them as attractive as possible. The amount of discount one can offer is limited, but the ways of presenting it are endless, and thus, from a marketing communications perspective, so are the ways to improve the receiver’s perceptions.
In the psychology field, some findings on improving perception are of particular interest for marketing communication. Van Osch, Blanken, Meijs and Van Wolferen (2015) found an interesting phenomenon concerning perception they call the group attractiveness effect (GA-effect, as coined by Van Osch et al.). They empirically demonstrate that the physical attractiveness of a group is perceived as greater than the average attractiveness of the group’s members. This effect is known colloquially as the cheerleader effect, which refers to the general observation that people seem more attractive (physically) when they are in a group than when they are alone. Van Osch et al. (2015) exposed their participants to still images of human faces, which is a rather basic stimulus. The experimental conditions differed only in the way these stimuli were presented: separately or in a group. This raises the question how much of this effect is due to purely the mode of presentation, and, more importantly, whether the effect then also occurs with other types of (still image) stimuli. Van Osch et al. proposed that the effect might be due to attention being drawn more towards the attractive faces in a group, thus weighing heavier in the evaluation of that group than when assessing everyone separately, with equal attention.
Looping back to discount items (often presented as still images in a discount booklet or on posters), it would be valuable if GA can be used to make discount items seem more attractive as well. GA comes from the cognitive psychology field, but potentially has much
applicability in marketing communications, since it deals with perception bias dependent on presentation. Hence, this study addresses the question: does a group attractiveness effect occur when discount items are presented in a group, as opposed to being presented separately? In addition, to emphasize the influence of selective attention, this study introduces goal priming as an extra factor. A primed goal can steer visual attention (Higgins et al. 2014). If a goal to look for certain elements enhances the group attractiveness effect, this is of added value to its overall applicability as well as our understanding of the effect.
First, the GA-effect will be discussed elaborately, interpreting the Van Osch et al. (2015) study and comparing it to another GA study carried out by Walker and Vul (2014). It will be explained why the selective attention explanation for the GA-effect seems most plausible, and how this can subsequently be applied to formulate expectations about GA-effects on discount items. This includes the introduction of goal priming as a tool to stimulate the GA-effect. An experiment to test these expectations is reported and its findings are discussed.
THEORETICAL FRAMEWORK
Group attractiveness: constituents and explanations
To apply GA to other stimuli, it must first be understood. The notion of the cheerleader effect has led to two interpretations in two different studies (i.e. Van Osch et al., 2015 and Walker & Vul, 2014). Van Osch et al. carried out nine studies which all evolved around the general observation that the perceived attractiveness of a group as a whole is higher than the average perceived attractiveness of its individual members. They compared three conditions, between subjects: group-rating, group-member and individual-ratings (study 1). In the group-rating condition, participants were shown a photograph of a group of people, and asked to evaluate the attractiveness of the group as a whole. In the group-member condition, the participants
saw the same photograph, but were asked to evaluate each member individually. In the individual-ratings condition, the participants were sequentially shown cropped pictures of all group members, and also evaluated each member individually. They found that the group-ratings were significantly higher than the average group-ratings in the group-member and individual-ratings condition. They replicated and further investigated the effect in the other experiments. They found that people across conditions all judge the same type of attractiveness (study 2), and that the GA-effect occurred with female, male and mixed-gender stimuli (study 1, 3, 4). The GA-effect also arose with a within-subjects design, albeit attenuated, yet significant, when participants were asked to make the group judgements first (study 5). When participants were forced to pay close attention to their group evaluations, their ratings were lower, but still produced a significant GA-effect (study 7). These findings show the GA-effect’s robustness against different setups. Excluding the within-subjects experiment, they calculated a mean effect size which was medium to large, Cohen’s d=0.60.
Van Osch et al. (2015) pose that the effect can be explained by selective attention. The authors theorize that, when judging a group, attention is drawn towards its most attractive members, who subsequently weigh heavier in the overall assessment. They substantiate this with a brief review of literature pointing out the evolutionary benefits of attending to the more attractive persons in a group (e.g. associating attractiveness with mating success, and other positive individual traits, see p. 560). When the members are judged individually (group-member and individual-ratings), the participants are forced to pay attention to all members equally, thus eliminating the GA-effect. It was expected that, if more attention is payed to the attractive group members, they would also be remembered better. They found that in the group-rating condition, the least attractive members were remembered less (study 8), thus supporting the selective attention account. An eye-tracking study revealed that participants looking at a group fixated more on the most attractive faces, however no
significant relation with GA-effect size was found (study 9). Another account considered by Van Osch et al. is based on the Gestalt-principle: the perception of a group is more than the sum of its parts. Put differently: the mere fact that it is a group increases, for example, the members’ attractiveness. Although the authors do not conclude from their findings that such dynamics are at the core of the GA-effect, they do admit there must be some at play. In some instances, they observed group attractiveness ratings (albeit non-significantly) higher than the individual rating of the most attractive member. Their results do not show any apparent differences in mean ratings between the group-member and individual-ratings condition. The Gestalt account provides a dynamic but no explanation for the observed GA-effect specifically. Walker and Vul (2014, experiment 4) investigated whether there was a difference if the group stimulus consisted of a photo (wherein all faces would have a coherent context) or just a composed set of portrait photos. The GA-effect was observed regardless.
Walker and Vul (2014) explain the GA-effect, that they call the cheerleader effect, differently. Their study notably only included a group-member and individual-ratings condition. They observed a GA-effect, with the attractiveness ratings in the group-member condition being significantly higher. This contrasts with the findings of Van Osch et al. (2015), who observed no such differences in their experiments. This issue is addressed further on. Walker and Vul theorize that the GA-effect arises from the phenomenon that observers create an ‘ensemble representation’ of the faces in a group. Put differently: the observer mentally constructs a face that is the average of all the group members’ faces. Then, the perception of the individual faces in the group is biased towards this ensemble representation, so all individual faces are perceived as slightly more like the group average. According to the authors, average-looking faces are found more attractive. Hence, presenting the faces in a group (vs. separately) makes each individual face appear more attractive.
To emphasize the difference between the accounts of Walker and Vul’s (2014) and Van Osch et al. (2015), I propose a mathematical interpretation of both. We first define group attractiveness as a certain value G, which is the overall perceived attractiveness of a group of n elements; the attractiveness rating in the group-rating condition. Every individual element i has its own perceived attractiveness value, a, and an assigned weight, w. G, then, is the average of all these element’s a multiplied by w:
(1)
𝐺 =∑ 𝑎𝑖𝑤𝑖
𝑛 𝑖=1
𝑖
Let there be another value Ma, which is the average of all elements’ individually
perceived attractiveness values a, thus equals the average attractiveness rating in the separate condition. If all elements weigh equally in the assessment of G, then w=1 for all elements, resulting in G=Ma.
(2)
if for all elements i, w = 1, then 𝐺 =∑𝑛𝑖=1𝑎𝑖𝑤𝑖
𝑖 = ∑𝑛𝑖=1𝑎𝑖
𝑖 = 𝑀𝑎
The group attractiveness effect, however, predicts G>Ma. Van Osch et al. (2015) let
their participants rate photos of (groups of) people, and argue that the effect is caused by attention being selectively divided when looking at a group. Importantly, the participants in the group-rating condition were asked to give a rating of the attractiveness of the group as a whole. Supposedly, attention bias towards the more attractive elements in a group makes those weigh heavier in the participant’s overall assessment of the group’s attractiveness than
the less attractive elements. This produces a higher average attractiveness rating than when exposed to each group member individually, in other words: w>1 for the more attractive elements, where w=1 for the average or low attractive elements, constituting a certain value for G. When exposed to each element individually, attention is forced to be spread evenly, so that for all elements w=1, resulting in a value Ma, and thus, G>Ma.
Walker and Vul (2014) approached the group-rating condition differently. They make the stronger claim that not the group as a whole, but each individual in that group appears more attractive due to the bias towards the ensemble representation of the group’s faces. That means they assume that the difference between G and Ma arises from increases in a when
calculating G, again resulting in G>Ma. To support this assumption, Walker and Vul
hypothesized that the GA-effect would be stronger when group size increases (experiment 4) and when the faces in the group are blurred (experiment 5). These manipulations would improve conceiving an ensemble representation, thus increasing the bias towards this in perceiving the individual faces. Their results demonstrated no effects of these manipulations, leaving their account unsubstantiated.
The findings of Van Osch et al. (2015) contradict Walker and Vul’s (2014). In their first experiment, they include three group-rating conditions: group-rating, individual-ratings and member. The member condition is the same as Walker and Vul’s group-rating condition, in that the participants rate the attractiveness of an individual shown in a group picture. The individual-ratings condition is the same in both studies: participants rate the attractiveness of an individual pictured alone on a cropped photo. The group-rating condition is unique to the Van Osch et al. study, wherein participants rate the average attractiveness of the people in a group photo. Walker and Vul found that the group-member condition produces higher attractiveness ratings than the individual-ratings condition. However, Van Osch et al. find no significant difference; the individual-ratings condition even
produced a slightly higher rating. The group-rating condition gave significantly higher attractiveness ratings than both other conditions. This strengthens the assertion that when attention is forced to be spent equally on every face, the GA-effect vanishes, in other words, that the bias arises from increases in w.
This does not mean that there are no fluctuations in a across conditions. Geiselman, Haight and Kimata (1984) found that the attractiveness rating of a target face assimilated towards the attractiveness ratings of two context faces. They carried out an experiment where subjects were exposed to triads of faces, wherein the target face, presented in the middle, was always of average attractiveness, and the two context faces beside the target were either of low, high, average or mixed (one low, one high) attractiveness. They found a significant effect of (a priori) context attractiveness on (a posteriori) target attractiveness rating. Target faces were perceived as more attractive when surrounded by high- or average- attractiveness faces, and as less attractive when surrounded by low-attractiveness faces. Geiselman et al. do not report the a posteriori assessments of the context faces’ attractiveness. The target faces did assimilate in both directions, so it is probable that the context faces’ attractiveness perceptions also assimilated. Rodway, Schepman and Lambert (2013) did find that the perceived attractiveness of a face positioned in the middle between four more faces assimilates more strongly to the other faces’ attractiveness than when positioned elsewhere. However, if the position of the elements in a group is varied, this assimilation bias would be of course eliminated, as all elements will be at some point displayed in a privileged position. This would mean that, even though a assimilates, Ma would remain constant and therefore a
does not affect the overall GA-effect. Anderson, Lindner and Lopes (1973) in demonstrated that group attractiveness only differs from the average attractiveness of its members when one member is designated as the group leader. When the leader was less or more attractive, respectively, than the other members, the group attractiveness was lower respectively higher
than the average of the members’ attractiveness. The findings of Geiselman et al., Anderson et al., and Van Osch et al. all contradict the idea that the GA-effect stems from an increase in a, as Walker and Vul propose. This again leads us to assume that the GA-effect arises from an increase in w, then, be it extra weight based on one person’s characteristics such as leadership, or evolutionary benefits associated with attractiveness. Why the GA-effect was observed in the Walker and Vul study and not in Van Osch et al.’s replication (group-member vs. individual-ratings) remains unanswered but may as well be due to an unidentified extraneous factor that distributed w unevenly in Walker and Vul’s study, and evenly in the Van Osch et al. study. Van Osch et al. propose that perhaps their sample sizes were too small to find a difference between the group-member and individual-ratings conditions. They also note that there might be a “qualitative difference between within- and between-subjects elicitations of attractiveness subjects” (p. 570). To speculate, and stick with the selective attention hypothesis: perhaps the procedure in the Van Osch et al. study forced the participants more strongly to divide their attention equally than the procedure carried out by Walker and Vul.
Group attractiveness in discount items
For this study, the selective attention account is most useful. Selective attention is a process that is easily transferable to hypothesize certain effects for discounts. Even if the construction of an ensemble representation of all the group’s elements takes place, there is no reason to assume a ‘more average looking’ discount would seem more attractive, like Walker and Vul (2014) propose regarding human faces. Whereas the attractiveness of a face lies in arguably subjective characteristics, the ‘attractiveness’ of a discount item is of a more concrete nature (e.g. the fixed number of a discount percentage). Thus, it is not expected that a GA-effect would arise from a biased increase in a in a group condition. Therefore, in terms of
interpreting the group attractiveness effect it will be mainly considered as described by Van Osch et al.. Hence this in this study the GA-effect will be understood as G>Ma, where G is
defined as in equation (1) and Ma is the average of each individual element’s perceived
attractiveness (i.e. w=1 for all elements). The difference between grouping conditions is expected to arise from increases in w for the more attractive elements in the group-rating condition, when the viewer has the freedom to unevenly distribute attention to the elements in a group (that would be in the group-rating condition). Thus it is hypothesized:
H1: The evaluation of discount items is more positive when presented in a group (group-rating) than when presented individually, one by one (individual-ratings).
Although Van Osch et al. (2015) observed from their eye-tracking data that the least attractive group members were fixated less upon, they did not find any correlation between the GA-effect and the proportion of visual attention paid to different faces in a group. In other words, the visual attention did not account for the variance in GA-effect size. This raises the suspicion that perhaps attention is not the selective activity. If visual attention differences do not account for the differences in attractiveness ratings, but we do assume that some elements weigh more strongly in the final assessment, then perhaps it is in a later phase of processing that the selection takes place. To test this idea, a measure should be implemented that assesses whether the memory of the discounts is (more) biased dependent on grouping condition. Therefore, the hypothesis:
H2: When relying on memory, the average discount of all items is estimated higher when the items were presented as a group (group-ratings) than when presented one by one (individual ratings).
In the Van Osch et al. (2015) interpretation, the difference between grouping conditions arises from attention being spread unevenly in the group-rating condition, with a bias towards more attractive elements. In the group-member condition, attention should be forcibly divided evenly across the whole group, just as in the individual-ratings condition. In that respect, the group-member condition is expected to produce similar results to the individual-ratings condition. However, Walker and Vul (2014) did demonstrate otherwise. To hopefully gain more understanding of the group-member presentation, it will be included in the current study.
RQ: Does the group-member condition, with regard to H1 and H2, produce evaluations and discount estimations similar to the group-rating condition, the individual-ratings condition, or otherwise?
Perception and memory of discount items
Whereas with human faces there are some natural explanations at hand for increases in w of the attractive members when observing a group, this does not come as simply for discount items. Nonetheless, the selective attention account requires for a GA-effect to occur only a reason why w would increase for the more attractive elements. Hence, it is important to establish what could bring this about with discount items, so that this can be taken into account and controlled or manipulated where needed.
With regard to mental processes in consumer behavior, especially involving selection and choosing, three more or less sequential levels are important to consider: the environmental features that can catalyze an automatic process, the process itself and the (behavioral) outcomes of this process (Chartrand, 2005). This study is concerned with the
first two: the environment (grouped or separate presentation of discounts) and the automatic process (perception of the discounts’ attractiveness), which are very much interlinked. To return to the GA-effect: it is within these two phases where factors may bring about fluctuations in w.
Environment
The perception of discount levels is influenced by many environmental factors. It matters what items are discounted and how much. Discounts may be concrete digits, yet it has been well-established that their perception can be influenced by the way they are framed (e.g. with gain vs. loss frames, Tversky & Kahneman, 1981; or which items in a set are discounted, Janiszewski & Cunha, 2004). Visual attention to advertisements competing with surrounding advertisements is steered by color (preferred over black and white ads), size (larger ads are seen more often than smaller ones), image-text ratio (ads with more image area, instead of text, get more attention) and position (the more at the end of a page, the less attention) and possibly countless other factors (for a review, see: Higgins, Leinenger & Rayner, 2014). The current study’s experiment will be designed so any conceivable environmental characteristic such as looks, lay out and discount assignment are either as constant as possible or randomized. The environmental factor of interest is manipulated: grouped (vs. separate) presentation. Keeping the visual and semantic characteristics of the stimuli constant we can continue to assume the individual elements’ w is influenced otherwise.
Goals and perception
A mental process often steering behavior is a goal. Goals can emerge from a prime (in the environment) or may be pre-existing. A primed goal directs behavior for many activities (Dijksterhuis, Smith, Van Baaren & Wigboldus, 2005). This also applies to visual attention.
When a viewer is goal-driven while looking at a visual stimulus, visual attention is directed accordingly (Higgins et al., 2014). Assuming a viewer is motivated to look for greater discounts (either because people love discounts, or they have been primed with a goal to do so), their visual attention, then, will be driven by this motivation. That could mean, for example, that the viewer would dwell longer on the higher discounts when looking at a poster full of discount items. Nordfang, Dyrholm and Bundesen (2013) found in their study, where participants had the task to seek out letters from a set of letters and digits, that, indeed, more visual attention is paid to elements that are relevant to a task. People who are more price conscious, tend to perceive discount value as lower than less price conscious people, presumably because they are more motivated to calculate the discount value more precisely (Alford & Biswas, 2002). Janiszewski, Kuo and Tavissoli (2013) found that selective attention may increase the likelihood of choosing the attended item over other items. One possible explanation is that viewers fail to perceive (and thus do not consider) stimuli they are not attentive to (Mack & Rock, 1998). These findings steer us directly towards the argument that a GA-effect has the potential to occur with other types of stimuli than photos of humans – in this case, discount items. Whether task-irrelevant information is at all processed or not, is still an unanswered question with contradictory findings across studies (Wood & Cowan, 1995). However, as was just established, the increased attention to the higher discounts may at least induce a preference for the higher discounts and decrease the chances of even perceiving the other ones. Through this process, it seems likely that the higher discounts have greater chances of being included in the calculations when the viewer would assess the overall appreciation of the discount deals on the poster, or estimate the average discount of all items. Thus, a goal of seeking the highest discounts should cause an increase in w of the higher discounts in a group of items.
Whereas people might have a natural tendency to pay more attention to physically attractive humans, for perhaps evolutionary reasons, this goal might not be naturally present when looking at discount items. However, goals can be primed with something as easy as a task instruction. Goal primes are often found in the environment, and exert an unconscious influence on the subsequent behavior, but one may as well be aware of the environmental factors (Chartrand, 2005). In fact, Chartrand and Bargh (1996) found that an unconscious goal prime produces the same behavioral outcomes as does an explicit task instruction. Therefore, it would be useful to prime the participants with a goal to look for the ‘most attractive’ discounts, such that they are (implicitly) given a task to find the greatest discount rates among the supplied set of discount items. This task would imply that the greater discounts are more important, and thus, hopefully, steer attention towards these discounts, providing them with a higher value for w than the lower discounts. This may have a main effect on the perception of the discount attractiveness, because (discount) search intention negatively affects the perceived value of a discount (Alford & Biswas, 2002). The presumed interaction remains unaffected, however, which is that the difference in perceived discount attractiveness is greater when primed with a goal to look for discounts.
H3a: The difference in evaluation of discount items between grouping conditions is greater when the participant is primed with a goal to look for discounts.
H3b: The difference in evaluation of average discount between grouping conditions is greater when the participant is primed with a goal to look for discounts.
METHOD
The hypotheses were tested using a 2 (prime vs. no prime) 3 (group vs. group-member vs. separate) between-subjects experimental design. The experiment was set up by distributing an online questionnaire, built with Qualtrics, wherein participants were randomly assigned to one of the six conditions.
Sample
A total of 185 responses were recorded. However, having acquired the raw dataset, it turned out 63 responses were invalid due to being unfinished or never even started. Of the remaining 122 participants, 59 (48.4%) reported they were male, 61 (50%) female, and two reported “rather not say” or “other”. The ages ranged from 18 to 65, M=31.41, SD=12.32. 81.10% of the participants reported Dutch as their native language. The convenience sample was acquired by distributing the online questionnaire through social media and by e-mail. As an incentive, they were informed about a raffle for a 25 euro gift voucher among the participants.
Stimuli and manipulations
Discount items
The stimulus material consisted of fake discount items consisting of a product image, the original price (crossed out), the discounted price and the discount percentage. The discount design was modeled after the discount items on Dutch supermarket websites, to contain the same information but to not look like they came from a known store (see Appendix A). All items were cleaning and household products; this was chosen because it seemed a rather neutral product category and one that would naturally be found in discount booklets. Each
participant was exposed to the same six items, either presented altogether simultaneously, or one after another.
The six products were chosen so that for each product, one other product had the same original price (prices were made-up but set close to the original price in the Albert Heijn web shop). For each of these price-pairs, one product was assigned a lower discount, and the other one a higher discount, so that in total there were three products with a high discount (35%, 40% and 45%), and three with a low discount (10%, and twice 15%). This was done so that the original prices would not interfere with the discount perception. Two versions were created of the set of products and their discounts, so that of each price-pair, in the first version, one would be assigned the high discount and one the low discount, and vice versa in the other version. Participants were randomly exposed to either one, to cancel out any perceived value differences between the products in a price-pair (see Appendix B).
To eliminate order- and positioning effects, the presentation order of the items was randomized. For the group and group-member conditions, three random orders were created and chosen so that no item was in the same position twice. Participants in these conditions were randomly shown one of these three orders. In the individual-ratings condition, the order in which the items were displayed was randomized completely.
Note that, across all variations, the actual values of the individual offers and overall remain constant.
Prime conditions
The participants that were randomly assigned to the prime condition (n=60), were presented an additional sentence on the instruction page, stating: “This study focuses on people’s ability to locate the largest discounts. Try to look for the best offers.”. In the control condition (n=62), this sentence was omitted.
Grouping-presentation conditions
Each participant was randomly assigned to one of the three conditions: group-rating (n=44), group-member (n=36), or individual-ratings (n=42). In the group-rating condition, the participants were presented randomly with one of the six order-discount variations of the discount items, displayed altogether (see Appendix A). Simultaneously, they were asked to answer three questions about the group of items they saw.
In the group-member condition, the participants were presented as well with one of these six order-price variations. However, they were asked to answer the questions about one product pointed out by a black arrow (see Appendix A). Next, the arrow moved and the questions were asked about the next item, repeated until the questions about all six items were completed.
In the individual-ratings condition, the participants were presented with one item at a time, in random order, and asked to answer the questions about that item. This was repeated until they saw the whole set of six items. They were randomly presented with one of the two sets of discount variations.
Measures
Discount appreciation
The dependent variable measuring the evaluation of discounts was measured with three questions, presented alongside the stimulus material. The first question was 1) “How attractive do you find these/this discount(s)?” (plural in the group-rating condition, singular in the group-member and individual-ratings condition), to be answered on a 7-point Likert scale labeled at the extremes with “Very unattractive” and “Very attractive”. This was directly adapted from the attractiveness measure as employed by Van Osch et al. (2015).
For the other two questions participants were asked to what extent they agreed with the following statements: 2) “These/this item(s) are/is an excellent buy for the money.” and 3) “The advertised item(s) represent(s) an extremely fair price.”, to be answered on a 7-point Likert scale labeled at the extremes with “Strongly disagree” and “Strongly agree”. These two questions are adapted from Alford and Biswas (2002) measure for perceptions of offer value.
In the group-rating condition, the participants saw all discount items simultaneously and answered the three questions once for the whole collection. In the group-member condition, the participants saw all discount items simultaneously, but were instructed to answer the questions about the item pointed out by a black arrow, repeated six times. In the individual-ratings condition, participants saw one item at a time, and had to answer the questions six times, once for each item.
From the responses to these three questions, a reliable scale was constructed to analyze as dependent variable and named Discount Appreciation ( = .81; for a detailed description of this construction, see Appendix C).
Discount Estimation
To measure the participants’ estimation of the average discount of all items they were shown, they were asked: “Of all the discount items you have just been shown, could you estimate the average discount of these items?”. To reply, they could fill in a number (M=24.15, SD=6.54,), providing the dependent variable Discount Estimation. To ensure the participants would have to rely on their memory to answer this question, some demographical questions with the purpose of distraction were implemented between the stimuli and this memory question.
Procedure
Participants could participate online. After an introduction and an informed consent, they were shown written instructions stating that the research topic was supermarket shopping, they would be shown some discount items and be asked questions about them, emphasizing there are no wrong answers and not to overthink it. This section also included the prime. Subsequently they were presented with the discount items and the accompanying questions. After that followed a prime manipulation check and a debriefing explaining the purpose of the study. At this point participants also had the opportunity to give their e-mail address to participate in the gift voucher raffle. After closing the questionnaire, one participant was selected by generating a random number and finding the corresponding item number in SPSS. After the winner selection procedure, the email addresses were erased from the data set.
RESULTS
Manipulation check
The manipulation check for the prime consisted of a 7-point Likert scale for agreement with the statement “While looking at the discounts, I really tried to find the best offers.”. It was unaffected by the priming condition, F(1, 116)=0.29, p=.593. This is peculiar, but does not necessarily mean that the people in the primed condition were not primed with a goal. They could simply have been unaware of this goal (Chartrand, 2005). The manipulation check measures only awareness of this goal.
Effects on Discount Appreciation
Firstly, the main- and interaction effects of grouping condition and priming condition on Discount Appreciation were analyzed in a two-factorial ANOVA, with grouping and priming
condition as factors and Discount Appreciation as dependent variable. The Discount Appreciation in the group-rating condition (M=4.03, SD=1.12, n=44) was slightly lower than in the group-member condition (M=4.35, SD=0.86, n=36) and the individual-ratings condition (M=4.42, SD=0.81, n=42). Note that the latter two are rather close to each other. No significant main effect of group condition on Discount Appreciation was found, F(2, 116)=2.11, p=.126, 2p=.04. A Fisher’s least significant difference (LSD) post hoc
comparison showed that only between the group-rating and individual-ratings condition, there was a marginally significant difference, p=.060, see Figure 1. These findings provide no support for H1, and the slightly lower average in the group-ratings condition even contradicts H1. Thus, H1 is rejected.
The prime condition had no significant main effects on discount appreciation, F(1, 116)=0.03, p=.868, The prime condition did not show any interaction effects with the grouping condition on discount appreciation, F(2, 116)=0.22, p=.804. Hence, H3a is rejected.
Figure 1
Note. The lines indicate the 95% confidence intervals for the mean.
Effects on Discount Estimation
The effects on Discount Estimation were analyzed with two-factorial ANOVA with group condition and prime condition as factors and Discount Estimation as the dependent variable. In the group-rating condition, the average Discount estimation was 23.27 percent (SD=6.24), group-member condition 23.49 percent (SD=5.02), and in the individual condition as 25.64 percent (SD=7.76), yielding no significant effect of group condition on Discount Estimation, F(2, 116)=1.48, p=.232, 2p=.03. LSD post hoc comparisons indicated no significant
differences between means. Hence, H2 is rejected. In addition, a one-sample t-test showed that the average guess was significantly lower than the actual average discount, which was 26.67 percent, M=24.15, SD=6.54, t(121)=-4.25, p<.001.
The prime condition did not have a significant main effect on Discount Estimation, F(1, 116)=0.119, p=.730, nor a significant interaction effect with group condition, F(2, 116)=0.191, p=.826. Thus, H3b is rejected.
Additional analyses
The ANOVA with the prime manipulation check as the dependent variable revealed a peculiarity: in the group-rating condition the participants reported a lower effort to look for the best offers (M=3.86, SD=2.08) than in the group-member condition (M=4.61, SD=1.68) and the individual-ratings condition (M=4.93, SD=1.44), F(2, 116)=4.47, p=.014, 2p=.07. A
post-hoc Bonferroni comparison indicated that only the difference between the group-rating and individual ratings condition was significant, p=.018.
To further scrutinize the marginal effect of group condition on Discount Appreciation, three one-way ANOVAs were carried out to assess the separate effects of group condition on the three questions that Discount Appreciation was constructed from:
1) “How attractive do you find these/this discount(s)?” 2) “These/this item(s) are/is an excellent buy for the money.” 3) “The advertised item(s) represent(s) an extremely fair price.”
Group condition had no significant effects on the responses to question 1, F(2, 119)=1.39, p=.239, nor question 3, F(2, 119)=0.31, p=.738. Interestingly, group condition did have a significant, medium-size effect on the response to question 2, F(2, 119)=4.20, p=0.017,
2
=.066. LSD post hoc comparison indicated a significant difference between the means in the group-rating (M=3.93, SD=1.45) and individual-rating conditions (M=4.61, SD=0.86),
p=.006, and a less pronounced difference between the group-rating and group-member condition (M=4.41, SD=0.88) or marginal significance, p=0.060.
DISCUSSION
By investigating the GA-effect in a stimulus other than human faces, this study served two purposes. Firstly, to deepen the understanding of the GA-effect so that it can be considered in more detail, providing scholarship on the subject with more direction. This was in part accomplished by the theoretical review, in which the two main studies on the subject were compared (i.e. Van Osch et al., 2015 and Walker & Vul, 2014). From this comparison, it was inferred that selective attention is the mechanism most likely underlying the GA-effect, which led to the expectation that a GA-effect could be induced by steering attention towards the relevant elements within a group. The experiment put this thesis to the test. The other purpose of this study relates to marketing communication, and was to find a new way to improve the perception of discounts through group presentation.
The GA-effect was not observed with judging discount items. The results did not produce any confirmation of the hypotheses. The differences in discount appreciation and discount estimation found between the grouping conditions were not significant. If anything, the discount appreciation was actually lower in the group-rating condition than in the individual-ratings and group-member condition, albeit by a small margin. The explanation for this lies likely in the differences between the stimulus material. Haberman and Whitney (2009) found that observers are able to produce summary statistical representations of human faces. They showed that people were able to determine the mean emotion in a set of faces. They relate this ability to ensemble coding, which is the ability of people to produce a summary statistical representation of a group of similar objects, such as a set of dots. However, their findings with regard to human faces are exceptional, since a face contains
more complex visual information than a dot. They argue this ability to process such complex stimuli so efficiently might be unique to faces for evolutionary reasons. This conclusion might explain why GA occurs in faces but not in discount items. Perhaps the schemata to process stimuli such as discount items are not as evolved as the face processing skills people appear to possess, causing the participants to elaborate on the discount items more than they would on faces. If the participants rationalized their assessment of the stimuli, it is no surprise the cognitive bias constituting the GA-effect did not occur. The results support this idea by the fact that across the grouping conditions there were no differences in the estimation of the average discount. This estimate requires rationalization specifically, although it was shown the guess was systematically too low. One could only speculate this might be due to participants rounding off their guesses to, for example, 25 percent instead of 26 percent.
The small difference between the rating condition on one hand, and the group-member and individual-ratings conditions on the other hand, may also have its explanation in the complexity of the stimulus. It has often been demonstrated that processing fluency increases liking for the processed stimulus (Westerman, Lanska & Olds, 2005); the easier people (feel they) can process a stimulus, the more they like it. In the group-rating condition, the participants were asked to assess six different discount items simultaneously; in the group-member and individual-ratings conditions, they were asked to assess six discount items one by one. Hence, the latter task was probably somewhat easier, prompting a slightly higher liking of the stimuli.
The only finding that is somewhat puzzling is that those in the group-member and individual-ratings condition reported taking significantly greater effort to look for the best bargains. Then again, it complements the earlier speculation that the participants in these conditions had a slightly less demanding task, assessing the discounts one by one. This could
mean they had more resources ‘free’ to consider and compare the best discounts, leading them to report a greater effort in doing so.
It is unfortunate that the prime, as opposed to the control condition without a prime, did not appear to have any effect on the participants’ reported motivations to seek out the best bargains. It cannot be deduced from the data whether the intended goal prime was perhaps activated unconsciously nonetheless. In that case, it did not have the desired effect, and may have instead motivated people to compare the discounts more, instead of just paying attention to the highest discounts. This would explain that no differences were found between the prime conditions, as both would have not given the participant a reason to concentrate more on some items than others.
The fact that not the overall evaluation of the discount items was influenced by group condition, but one of the factors was, adds to the idea that the participants have scrutinized the discount items. The responses to whether the discount represented a fair price was insensitive to group condition. If one would elaborate thoroughly on that question, it becomes actually quite two-sided. Perhaps fairness was interpreted in terms of fair-trade, which pertains to both the buying and selling parties. A high discount, then, may feel more fair to the buyer, but be perceived at the same time as unfair to the seller, and vice versa. The question whether the participants thought the item was an excellent buy for the money, is more one sided and perhaps reflects most the type of ‘attractiveness’ this study aimed to measure. Why the participants, contrary to the expectations, responded more positively to this question in the individual-ratings condition than in the group-rating condition is more difficult to explain. Perhaps, the individual-ratings condition allowed the least for comparing the items and putting them in perspective. The group-member condition produced slightly more positive responses than the group-ratings condition, but less so as the individual-ratings did. This could be, then, explained in the same vein, in that the one-by-one (instead of
simultaneous, overall) evaluation, distracted the participants from the whole perspective, but allowed for some comparison. However, it must be kept in mind that attractiveness, ‘fair price’ and ‘excellent buy’ remain closely related judgements (as the factor analysis demonstrated), and they were altogether hardly affected by the group condition.
Estimation of the discounts’ average appeared not to be influenced by the group nor prime condition. This estimation was measured to assess whether the GA-effect would extend into memorization of the (discount) items. This was not observed, nor was the GA-effect at all, so no conclusive answer can be provided.
The slightest differences that were found favor Van Osch et al.’s (2015) interpretation of the GA-effect over Walker and Vul’s (2014). The data indicates marginal differences between the group-ratings condition on one hand, and the group-member and ratings on the other hand, whereas the differences between the group-member and individual-ratings were next to none. That means that, next to presentation, an important determinant of the GA-effect might be whether the elements in a group are judged altogether or individually – again supporting the selective attention account.
Limitations
The foremost limitation of this study is the sample size. Due to an unforeseen amount of invalid entries in Qualtrics, the data set turned out to contain 122 valid entries instead of the intended 180 (by rule-of-thumb that all six conditions should contain 30 participants). This sample’s statistical power is therefore somewhat limited. However, the full 180, with regard to statistical power, were only necessary for investigating the interaction between the prime conditions and the grouping conditions. Since the prime conditions appeared to have no main nor interaction effects at all, one can safely say that a larger sample would unlikely have demonstrated otherwise.
Another factor that may have caused noise is that the participants took the questionnaire on their own computers, without time limits. This perhaps allowed the participants to spend too much time assessing the stimuli. Also, the evaluation questions were presented simultaneously with the discount item(s) these referred to, allowing the participants to refer back and forth between stimulus and question. Although this appears not to have impaired the studies of Van Osch et al. (2015), it might be of importance to this particular type of stimulus. In a follow-up study, one could try limiting exposure time and posing evaluation questions after exposure.
Future recommendations
Future research into the GA-effect should focus on two things. Firstly, the question of what exactly determines people’s judgements of human attractiveness, to gain a deeper understanding of the GA-effect and why it arises (or not). Secondly, if more accurate determinants of the GA-effect are known, other stimuli that the GA-effect might occur with can be better identified based on these determinants. To further elaborate on the selective attention account, it would be valuable to investigate whether a primed goal to attend to the more attractive persons would accentuate the GA-effect. The goal prime did not have any effect in the current study, but also it is unclear whether the goal was truly activated and in the intended way.
For advertising the most important message is to consider the complexity of their messages. The discount items made for this study are based on the formats used in most Dutch supermarket websites, and they do contain much information: original price, discounted price and discount percentage. Only two of these pieces of information are necessary to imply the third one. In other words: discounts could be presented more efficiently. Presenting them more concisely would make processing them easier, which could
have many positive effects on people’s evaluation of the discounts (Westerman et al., 2005). From a purely perceptional perspective, one could argue that it is better to present discounts separately than altogether, to impair comparison, which in this study was perhaps the reason for a slightly more positive evaluation of the discount items.
Conclusion
This study came forth from a quite large theoretical leap; from human face attractiveness to ‘attractiveness’ of discount items. It has demonstrated that the GA-effect does not occur with discount items. Apparently, it is not applicable to just anything, and likely relies on the specific characteristics of facial perception and processing. It appears that discount items are more sensitive to comparison, scrutiny and rationalization when being evaluated than human faces.
REFERENCES
Alford, B. L., & Biswas, A. (2002). The effects of discount level, price consciousness and sale proneness on consumers’ price perception and behavioral intention. Journal of Business Research, 55, 775-783.
Anderson, N. H., Lindner, R., & Lopes, L. L. (1973). Integration theory applied to judgements of group attractiveness. Journal of Personality and Social Psychology, 26(3), 400-408.
Chartrand, T. L. (2005). The role of conscious awareness in consumer behavior. Journal of Consumer Psychology, 15(3), 203-210.
Chartrand, T. L., & Bargh, J. A. (1996). Automatic activation of impression formation and memorization goals: Nonconscious goal priming reproduces effects of explicit task instructions. Journal of Personality and Social Psychology, 71(3), 464-478.
Dijksterhuis, A., Smith, P. K., Baaren, R. B. van, & Wigboldus, D. H. J. (2005). The unconscious consumer: effects of environment on consumer behavior. Journal of Consumer Psychology, 15(3), 193-202.
Geiselman, R. E., Haight, N. A., & Kimata, L. G. (1984). Context effects on the perceived physical attractiveness of faces. Journal of Experimental Social Psychology, 20, 409-424.
Haberman, J., & Whitney, D. (2009). Seeing the mean: Ensemble coding for sets of faces. Journal of Experimental Psychology: Human Perception and Performance, 35(3), 718-734.
Higgins, E., Leinenger, M., & Rayner, K. (2014). Eye movements when viewing advertisements. In J. Simola, J. Hyönä & J. Kuisma (Eds.), Perception of visual advertising in different media: From attention to distraction, persuasion, preference and memory, (pp. 6-22). Lausanne: Frontiers Media SA.
Janiszewski, C., & Cunha, M. (2004). The influence of price discount framing on the evaluation of a product bundle. Journal of Consumer Research, 30(4), 534-546.
Janiszewski, C., Kuo, A., & Tavassoli, N. T. (2012). The influence of selective attention and inattention to products on subsequent choice. Journal of Consumer Research, 39, 1258-1274.
Mack, A., & Rock, I. (1998). Inattentional Blindness. Cambridge: MIT Press.
Nordfang, M., Dyrholm, M., & Bundesen, C. (2013). Identifying bottom-up and top-down components of attentional weight by experimental analysis and computational modeling. Journal of Experimental Psychology, 142(2), 510-535.
Rodway, P., Schepman, A., & Lambert, J. (2013). The influence of position and context on facial attractiveness. Acta Psychologica, 144, 522-529.
Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211, 453-458.
Van Osch, Y., Blanken, I., Meijs, M. H. J., & Van Wolferen. (2015). A group’s physical attractiveness is greater than the average attractiveness of its members: The group attractiveness effect. Personality and Social Psychology Bulletin, 41(4), 559-574. Walker, D., & Vul, E. (2014). Hierarchical encoding makes individuals in a group seem more
attractive. Psychological Science, 25(1), 230-235.
Westerman, D. L., Lanska, M., & Olds, J. M. (2015). The effect of processing fluency on impressions of familiarity and liking. Journal of Experimental Psychology: Learning, Memory and Cognition, 41(2), 426-438.
Wood, N. L., & Cowan, N. (1995). The cocktail party phenomenon revisited: Attention and memory in the classic selective listening procedure of Cherry (1953). Journal of Experimental Psychology, 124(3), 243-262.
APPENDIX
A. Examples of the stimulus materials. Group-rating stimulus (discount variation 1)
Individual-ratings stimulus (discount variation 2)
B. Discount variations between the two versions.
product
version 1
original price (discount %)
version 2
original price (discount %)
1 2.85 (35) 2.85 (10) 2 2.85 (10) 2.85 (35) 3 2.19 (45) 2.19 (15) 4 2.19 (15) 2.19 (45) 5 1.35 (40) 1.35 (15) 6 1.35 (15) 1.35 (40)
Note. Price pairs are 1 and 2, 3 and 4, 5 and 6.
C. Scale construction of Discount Appreciation
The scale Discount Appreciation was constructed of the three questions the participants were asked about the discount items. First, for the participants in the group-member condition and the individual ratings-condition, who answered the three questions six times, the average of these six responses was calculated for each question. Then, these means and the responses from the group-rating condition were recoded into three variables, so that each one contained
the replies of all participants to question 1, 2 and 3, respectively. The Discount Appreciation scale was constructed by computing the mean of these three scores for each participant. To investigate the unidimensionality of the constructed scale, a principal components analysis with Varimax rotation was carried out. This resulted in one component strong enough to demonstrate unidimensionality (Eigenvalue = 2.18; explaining 72.76% of variance, see Table 2) and no other components were meaningful (Eigenvalue < 0.56).
Table 2. Component matrix of the 3 constituent items of Discount Appreciation (N = 122)
Factor Factor load M (SD)
1) 0=Very unattractive; 5=Very attractive .82 4.08 (1.18) 2) 0=Strongly disagree; 5=Strongly agree .92 4.31 (1.14) 3) 0=Strongly disagree; 5=Strongly agree .82 4.39 (1.05)
These items provided a reliable scale ( = .81) that indicated Discount Appreciation, wherein min = 1 indicates the most negative evaluation and max = 6 the most positive, M = 4.26, SD = 0.96.
