Brand identification
A buffer for the impact of negative
product consumption experience
Thesis submitted for the degree of Master of Science
Frans Muthert (10326952)
University:
University of Amsterdam
Faculty:
Economics and Business
Study programme:
Executive Programme Management Studies
Study track:
Marketing
Academic year:
2013-2014
Thesis supervisor:
Drs. Frank Slisser
Version:
Final
Acknowledgements
It was my deep-seated interest in the elements that link psychology, sociology, and marketing together that led to choosing brand identification as the topic of the research described in this thesis. Not only this interest, but also the guidance of my thesis supervisor – Frank Slisser – energised me throughout the research and thesis writing process. His guidance was encouraging, constructive, and incisive. Therefore and foremost, I would like to express my sincere gratitude and thanks to Frank Slisser.
Additionally, I would like to thank Pieter van Casteren and Brian Droop of the faculty of Economics and Business. They reviewed my data analysis approach and provided useful suggestions for
significance testing. I am also thankful to all the people who participated in the research. Without their participation, the research and thesis writing process would have come to a premature and sudden end.
Abstract
This research investigates the moderating or buffering role of brand identification in the relationship between negative product consumption experience and product evaluation plus brand evaluation. As such, it provides new insights into protecting customer equity; protecting customer equity is key as restoring deteriorated customer equity is costly and sometimes ineffective. In a questionnaire, participants needed to evaluate the focal product (iMac desktop computer) plus focal brand (Apple) before and after the focal negative product consumption experience (hard disk failure). Because participants belonged to the group weak identification with focal brand or to the group strong identification with focal brand, this research design allowed to compare within-subject change in product evaluation plus brand evaluation between the two groups. The results indicate that negative product consumption experience may have a negative impact on product evaluation plus (to a lesser extent) brand evaluation and that this impact may not be buffered by brand identification. The results also reveal that brand identification may (to a limited extent) decrease the likelihood of (re)purchasing the product after a negative product consumption experience. Thereby, brand identification may prevent encountering a negative consumption experience with the same product again. So instead of directly protecting customer equity from negative product consumption experience, brand
identification prevents further deterioration of customer equity. Future research should further examine under which conditions the brand identification buffer does occur.
Table of contents
1. Introduction
1
2. Literature review
4
2.1 – Brand identification: social identity theory 4
2.1.1 – Social categories versus non-social categories 4 2.1.2 – Brand identification as a higher-order construct 5 2.2 – Brand identification as a buffer: motivated reasoning theory driven by balance theory 5 2.2.1 – Negative brand information as a cause of inconsistency in evaluation 6
2.2.2 – Role of source characteristics 7
2.2.3 – Negative product cons. experience as a cause of inconsistency in evaluation 7 2.3 – Product evaluation versus brand evaluation: associative network theory 8
2.3.1 – Associative network linkages 9
2.3.2 – Occurrence/creation of associative network linkages 9
3. Method
11
3.1 – Overall research design 11
3.1.1 – Research strategy 13 3.1.2 – Research tactics 14 3.2 – Research participants 17 3.3 – Research variables 19 3.3.1 – Independent variables 19 3.3.2 – Dependent variables 19 3.3.3 – Moderating variables 20 3.3.4 – Control variables 21 3.4 – Data analysis 22 3.4.1 – Within-subject change 22
3.4.2 – Between-group comparison of within-subject change 22 3.4.3 – Within-subject comparison of within-subject change 24
4. Results
25
4.1 – Within-subject change 25
4.2 – Between-group comparison of within-subject change 26
4.2.1 – Initial test 26
4.2.2 – Validation test 27
4.3 – Within-subject comparison of within-subject change 28
5. Discussion
29
5.1 – Discussion 29
5.1.1 – Between-group comparison of within-subject change 29 5.1.2 – Within-subject comparison of within-subject change 31
5.1.3 – Answer to research question 32
5.2 – Limitations 32
5.3 – Recommendations for future research 34
6. Conclusions
36
References
38
Appendices
48
Appendix A – Literature review table 48
Appendix B – Conceptual model 49
Appendix C – Scale items 50
Appendix D – Questionnaire 52
Appendix E – Result tables 60
List of figures
Figure 1 – Conceptual model 49
List of tables
Table 1 – Brand identification buffer literature 48
Table 2 – Total sample: means, stand. dev., Cronbach’s alpha coeff., and Pearson corr. coeff. 60 Table 3 – Group weak brand identification: means, stand. dev., and Pearson corr. coeff. 61 Table 4 – Group strong brand identification: means, stand. dev., and Pearson corr. coeff. 62 Table 5 – Within-subject change: results of paired samples t-tests 63 Table 6 – Between-group comp. of within-subject change: results of indep. samples t-tests 63 Table 7 – Between-group comp. of within-subject change: results of hier. mult. reg. analysis for product attitude change variable and product purchase intention change variable 64 Table 8 – Between-group comp. of within-subject change: results of hier. mult. reg. analysis for brand attitude change variable and brand purchase intention change variable 64 Table 9 – Within-subject comp. of within-subject change: results of paired samples t-tests 65
1.
Introduction
People who post messages on the internet like the one above, clearly have a very close relationship with the brand they post about. This relationship can become so close, that the self-concept or the inner self of these people intertwines with the brand identity: the brand has extended their inner self and has become a part of who they are. This phenomenon is called self-brand connection (Escalas & Bettman, 2003) or brand identification (Lisjak, Lee, & Gardner, 2012) and has important implications. Previous research found that brand identification can build value equity, brand equity, relationship equity, and thereby customer equity (Lemon, Rust, & Zeithaml, 2001; Rust, Zeithaml, & Lemon, 2001), which may increase future business value in the form of future revenue and future profits (Rust, Lemon, & Zeithaml, 2004). This is because brand identification increases consumer satisfaction (Kuenzel & Halliday, 2008; Tuškej, Golob, & Podnar, 2013), brand loyalty (Bhattacharya, Rao, & Glynn, 1995; Bhattacharya & Sen, 2003), positive word of mouth (Ahearne, Bhattacharya, & Gruen, 2005; Bhattacharya & Sen, 2003; Kuenzel & Halliday, 2008; Tuškej et al., 2013), and willingness to pay (Del Rio, Vazquez, & Iglesias, 2001). Previous research also found that brand identification may protect customer equity: the impact of negative brand information on brand evaluation is “buffered” by brand identification (Ahluwalia, Burnkrant, & Unnava, 2000; Einwiller, Fedorikhin, Johnson, & Kamins, 2006; Kamins, Folkes, & Perner, 1997; Lisjak et al., 2012; Swaminathan, Page, &
Gürhan‐Canli, 2007). So when the person who posted the message above on the internet encounters negative information about Alfa Romeo, it may not affect this person’s evaluation of Alfa Romeo, meaning that customer equity stays intact.
But negative brand information is not the only type of “threat to the brand” that may deteriorate customer equity. What will happen when the person who posted the message is driving a new Alfa
My passion for Alfa Romeo is strange . . . Yes, I’ve had many brands, from Ford to VW Golf GTi, to Honda to Peugot . . . Not bad cars but nothing makes my heart beat like my Alfas. I don’t claim to know everything about Alfa Romeo but I’m learning slowly everyday about the history, the legend and what it means to drive and own an Alfa. Yes, the Busso was probably the last true Alfa engine . . . But it doesnt [sic] stop me beaming with pride when I see another Alfa on the road, even a G or a Mito. . . . Owners of G’s [sic] and Mitos probably don’t know the history of our special brand, but perhaps they will fall in love with their new Alfa and pass on the message . . . I’ve rambles a bit here, sorry, but I have to defend my passion for Alfa.
consumption experience on brand evaluation also be buffered by brand identification (like the impact of negative brand information)? And will the impact on product evaluation be buffered as well? Previous research did not answer these questions, thereby forming a huge research gap that consists of two parts; the moderating or buffering role of brand identification in the relationship between:
1. Negative product consumption experience and; 2. Product level evaluation plus brand level evaluation.
Filling up this research gap is crucial for several reasons. First of all, filling up the research gap provides organisations/product managers/brand managers with new insights into preventing negative product evaluation plus brand evaluation (deterioration of customer equity) when a product-harm crisis that affects product consumption experience occurs. Usually, organisations implement expensive and sometimes ineffective reactive strategies to restore deteriorated customer equity after product-harm crises (Siomkos & Kurzbard, 1994). For instance, a voluntary product recall strategy costs most often more than $9 million to implement per case (Ernst & Young, 2011) and will most likely not restore deteriorated customer equity completely (Dawar & Pillutla, 2000; Ernst & Young, 2011; Laufer & Coombs, 2006). When there is evidence that brand identification buffers the impact of negative product consumption experience, it positions brand identification as a meaningful proactive “impact-aversion” alternative for the reactive “restoration” strategies.
A second reason is the increased risk of customer equity deterioration caused by the increased probability that consumers encounter negative product consumption experiences. This increased probability is a consequence of products becoming more complex/error-prone and consumers
becoming more demanding (Birch, 1994; Dawar & Pillutla, 2000; Patterson, 1993). The increased risk of deterioration of customer equity needs to be proactively mitigated; more insights into proactive impact-aversion strategies are crucial for the long-term survival of brands and organisations. Filling up the research gap also provides new insights into the relationship between brand architecture and deterioration of customer equity. Once there is evidence that negative consumption experiences at the product level may spill over to the brand level, organisations/product managers/brand managers can create a distinct identity for complex/error-prone products to reduce the impact of negative product consumption experience on brand evaluation. This proactively mitigates the increased risk of
deterioration of customer equity at the brand level and therefore contributes to the long-term survival of both the brand and the organisation. Finally (and more from a theoretical perspective), filling up the research gap provides new insights into why and how brand identification buffers the impact of threats to a brand. With these new insights, future research does not have to test whether brand identification buffers the impact of specific types of threats. The insights may also validate and shed new light on the findings of previous research.
Given the importance and urgency of filling up the research gap, this research aims to fill up the gap by answering the following question:
To which extent is the impact of negative product consumption experience on product evaluation plus brand evaluation moderated by brand identification?
Based on an extensive review of the literature, expectations are that brand identification does act as a moderator or buffer in the relationship between negative product consumption experience and product evaluation plus brand evaluation, but that the impact of negative product consumption experience on product evaluation is bigger than on brand evaluation. These expectations or hypotheses were tested quantitatively using an overall research design that allowed to compare change in product evaluation plus brand evaluation (change that occurred when participants encountered negative product
consumption experience) between participants that weakly or strongly identified themselves with the brand. The data were collected using a survey and specifically a delivery-and-collection hard copy questionnaire.
In the next chapter, the literature is reviewed extensively, leading to a conceptual model and
hypotheses. Thereafter, the overall research design, research participants, research variables, and data analysis are described. In the following chapter, the results of testing the hypotheses are presented. Subsequently, the results are discussed, the research question is answered, limitations of the research are described, and recommendations for future research are given. The final chapter provides
2. Literature review
This chapter reviews the literature extensively and splits up the research question into two testable hypotheses. In the first part, the brand identification construct and its context are reviewed using social identity theory. Thereafter, the findings of previous “brand identification buffer” research are reviewed (Table 1 in Appendix A provides an overview) and combined with balance theory to form H1. In the final part of this chapter, H1 is anchored even further in existing theory and H2 is formulated using associative network theory. Both the literature review and H1 are captured in a conceptual model that is depicted in Figure 1 in Appendix B.
2.1 – Brand identification: social identity theory
Previous research suggests that people have at least three key self-definitional needs (Ashforth & Mael, 1989; Bhattacharya & Sen, 2003; Dutton, Dukerich, & Harquail, 1994; Stokburger-Sauer, Ratneshwar, & Sen, 2012): self-continuity (people need to know themselves), self-distinctiveness (people need a feeling of uniqueness), and self-enhancement (people need to feel good about themselves). These key self-definitional needs revolve around the articulation of the self-concept or inner self and typically come down to the central identification question of “who am I?” (Ashforth & Mael, 1989; Bhattacharya & Sen, 2003). Social identity theory suggests that people answer this question by not only looking at their personal identity, but also by developing a social identity (Ashforth & Mael, 1989; Bhattacharya & Sen, 2003; Brewer, 1991; Tajfel & Turner, 1985). A
personal identity is all about “I” and idiosyncratic characteristics that differ from person to person, like psychological traits and interests (Ashforth & Mael, 1989). When a person develops a social identity, this person transcends his/her own personal identity by classifying him-/herself into social or non-social categories (based on category characteristics): “I” becomes “we” (Ashforth & Mael, 1989; Brewer, 1991; James, 1890). This means that the inner self is extended with the social or non-social categories. These categories thereby become a part of the inner self: the person defines him-/herself in terms of the categories and successes/setbacks of the categories are personally experienced.
2.1.1 – Social categories versus non-social categories
So people extend their inner self with social or non-social categories to fulfil key self-definitional needs. A social category consists of a social referent or a social relationship, such as a tennis hero or a doctor-patient relationship (Ashforth & Mael, 1989). Once the inner self of a person is extended with a social category consisting of, for instance, a social referent, this person defines him-/herself in terms of the social referent and may even try to become the social referent as much as possible (Kelman, 1961).
A non-social category consists of a non-social referent, such as a brand (Lisjak et al., 2012). When the inner self of a person is extended with a non-social category that consists of a brand, this person defines /herself in terms of the brand (just as if the brand were a social referent), identifies him-/herself with the brand, and probably wants that significant others identify him/her with the brand (Baum, 2009; Greenwald & Breckler, 1985). The extension of the inner self with a brand is called self-brand connection (Escalas & Bettman, 2003) or self-brand identification (Lisjak et al., 2012).
2.1.2 – Brand identification as a higher-order construct
Brand identification must not be confused with brand commitment, brand support, and brand loyalty: brand identification is a higher-order construct that overarches these constructs. Brand commitment, brand support, and brand loyalty can therefore be seen as lower-order outcome constructs of brand identification (Bergami & Bagozzi, 2000; Einwiller & Kamins, 2008).
2.2 – Brand identification as a buffer: motivated
reasoning theory driven by balance theory
Attitude formation research found that consumers give more weight on negative brand information than on positive brand information: the negativity effect (Ahluwalia et al., 2000; Fiske, 1980; Herr, Kardes, & Kim, 1991; Klein, 1996). Because of the negativity effect, negative brand information usually has a relatively big negative impact on brand evaluation and can therefore be seen as a serious threat to a brand (Einwiller et al., 2006). For consumers who strongly identify themselves with a brand (i.e. have strong brand identification), this threat to the brand is felt as a threat to the inner self.
According to previous research, these consumers will therefore try to avert the threat by defending the brand and the inner self: the impact of the negative brand information on brand evaluation is
minimised (Ahluwalia et al., 2000; Einwiller et al., 2006; Kamins et al., 1997; Lisjak et al., 2012; Swaminathan et al., 2007). So brand identification buffers the impact of negative brand information because a defence mechanism is triggered. This defence mechanism is more easily triggered for people who are more defensive by nature (Lisjak et al., 2012). Implicit self-esteem and explicit self-esteem regulate the natural defensiveness of people (Lisjak et al., 2012). Implicit self-esteem entails the implicit or unconscious/automatic evaluation of the inner self (Greenwald & Banaji, 1995; Lisjak et al., 2012), while explicit self-esteem refers to the explicit or conscious/reasoned evaluation of the inner self (Lisjak et al., 2012; Rosenberg, 1965). Previous research consistently found that people with low implicit self-esteem are more defensive by nature (e.g. Greenwald and Farnham [2000] and McGregor and Jordan [2007]). Previous research is less consistent regarding explicit self-esteem (Lisjak et al.,
(e.g. Harmon-Jones et al. [1997]), while other research found the opposite (e.g. McFarlin and Blascovich [1981]) or found that people that have the combination of low implicit self-esteem and high explicit self-esteem are more defensive by nature (e.g. Jordan, Spencer, Zanna, Hoshino-Browne, and Correll [2003]).
So previous research provides evidence that triggering the defence mechanism is key for the brand identification buffer to occur. But previous research only focussed on negative brand information as a trigger for the defence mechanism: negative product consumption experience – defined as every negative experience during the consumption or use of a product (Brakus, Schmitt, & Zarantonello, 2009) – received no attention as another potential trigger. See the column “type of threat” in Table 1 in Appendix A. So will the impact of negative product consumption experience on product level
evaluation plus brand level evaluation also be buffered by brand identification? A hypothetical answer to this question is provided below by zooming in on how the defence mechanism and the brand identification buffer actually work for negative brand information.
2.2.1 – Negative brand information as a cause of inconsistency
in evaluation
According to balance theory, consumers strive for consistency in evaluation; consumers cannot hold an overall positive evaluation resulting from positive perceptions and an overall negative evaluation resulting from negative perceptions at the same time (Crespi, 2013; Einwiller et al., 2006; Einwiller & Kamins, 2008; Hayes, 2000; Heider, 1946, 1958). When consumers experience an overall positive evaluation and an overall negative evaluation at the same time, they experience an inconsistency (resulting in unpleasant tension [Hayes, 2000]) that needs to be resolved. There are two mutually exclusive solutions for this inconsistency:
1. Deeming the negative perceptions as incorrect (thereby putting them aside)/as more positive than they really are negative evaluation is eliminated;
2. Deeming the positive perceptions as incorrect (thereby putting them aside)/as more negative than they really are positive evaluation is eliminated.
Previous research that found evidence for the brand identification buffer when consumers encounter negative brand information, explained the buffer with help of the balance theory described above. Specifically, previous research – although not explicitly mentioned in the literature – used motivated reasoning theory driven by balance theory in the explanations. When consumers encounter negative information about a brand they identify themselves with, they experience an inconsistency: new negative brand evaluation (resulting from the negative brand information) versus current positive brand evaluation (consumers with strong brand identification will most likely evaluate the brand
positively [Einwiller et al., 2006]). Previous research suggests that these consumers follow the first solution in solving the inconsistency, because they are highly motivated to reach a specific desired conclusion: a conclusion that protects the brand and themselves (because the brand that is threatened is part of their inner self). So these consumers are not motivated to reach an accurate conclusion: the accurate conclusion may be a threat to the brand and themselves (Chaiken, Giner-Sorolla, & Chen, 1996; Einwiller et al., 2006; Kunda, 1990).
2.2.2 – Role of source characteristics
So according to the previous research that found evidence for the brand identification buffer when consumers encounter negative brand information, an inconsistency in evaluation experienced by consumers with strong brand identification is key for the defence mechanism to be triggered and the brand identification buffer to occur (because then these consumers will reason towards brand
protection and self-protection). But this research did not take the credibility and likability of the source of the negative brand information into account (except Einwiller & Kamins [2008]): in all the
experiments, participants encountered negative brand information from a credible and likable source. Will an inconsistency (among consumers with strong brand identification) caused by negative brand information from a non-credible and non-likable source be resolved, resulting in triggering the defence mechanism and the occurrence of the brand identification buffer? According to the extension of balance theory by Newcomb (1968), the answer is no: consumers will only resolve an inconsistency when the information causing the inconsistency is from a credible and likable source (Eiser, 1986; Hayes, 2000). This is because inconsistency caused by information from a credible and non-likable source does not result in unpleasant tension (Hayes, 2000). So the more credible and non-likable the source, the more likely an inconsistency will be resolved.
2.2.3 – Negative product consumption experience as a cause of
inconsistency in evaluation
To sum up the above, there are two requirements that must be met for the defence mechanism to be triggered and the brand identification buffer to occur:
1. Consumers with strong brand identification experience an inconsistency in their evaluation of the brand;
2. The source of the information causing the inconsistency, is credible and likable.
Does a negative product consumption experience meet both requirements? So will a negative product consumption experience trigger the defence mechanism of consumers with strong brand identification so that the impact of the negative product consumption experience on product evaluation plus brand
that both requirements are met. Heider – the “founder” of balance theory – provides evidence that the first requirement is met: he describes an experiment in which participants were given a description of a social situation in which a person likes reading a specific poem, but dislikes the poet (Heider, 1958). Heider and the participants perceived “to like reading the poem and to dislike the poet” as an
inconsistency in evaluation. Because of this and the fact that reading a poem – although not explicitly mentioned by Heider – is actually a product consumption experience, it can be suggested that negative product consumption experience can cause an inconsistency among consumers with strong brand identification. Advertising research provides evidence that the second requirement is met: experiences are overall more credible and likable than information. This is because experiences are more personal (in general consumers believe themselves), actively sought, and direct than information (Kempf & Smith, 1998; Singh, Balasubramanian, & Chakraborty, 2000).
Based on the discussion above, the following hypothesis can be formulated:
H1: the impact of negative product consumption experience on product evaluation plus brand evaluation is smaller for consumers with strong brand identification than for consumers with weak brand identification.
2.3 – Product evaluation versus brand evaluation:
associative network theory
H1 is based on two important assumptions:
1. A negative consumption experience at the product level has an impact on both product level evaluation and brand level evaluation;
2. The impact of a negative consumption experience at the product level on product level evaluation and the impact on brand level evaluation are both buffered by brand identification.
Previous research focussed only on the brand level by conducting experiments that involved negative brand information and measurement of brand evaluation (Ahluwalia et al., 2000; Einwiller et al., 2006; Kamins et al., 1997; Lisjak et al., 2012; Swaminathan et al., 2007). No research included negative product information and measurement of product evaluation. See the column “level of consumer evaluation” in Table 1 in Appendix A. So the question remains: are the assumptions that form the base of H1 correct? And if so, will the impact of negative product consumption experience on product evaluation be bigger than on brand evaluation? An answer to the first question and a
hypothetical answer to the second question is provided below by using associative network theory to zoom in on the relationships between a product and a brand.
2.3.1 – Associative network linkages
According to brand portfolio research, a negative incident at the product level (e.g. a negative incident with extension products or sub brands) may spill over to the brand level, meaning that both product evaluation and brand evaluation are more negative than before the negative incident (John, Loken, & Joiner, 1998; Lei, Dawar, & Lemmink, 2008). This spill over effect can be explained using associative network theory (Collins & Loftus, 1975). Associative network theory suggests that products and brands are nodes in the memory of consumers to which evaluations, expectations, etc. belong. An associative memory network is formed by the linkages that may exist between the nodes. These associative network linkages may cause that both the product node and the brand node are “activated” when a consumer encounters a negative incident at the product level (Keller, 1993; Lei et al., 2008). This in turn may cause that not only the evaluation belonging to the product node is negatively “updated”, but also the evaluation belonging to the brand node: spill over occurs (Lei et al., 2008).
Whether spill over occurs, depends on the directionality of the associative network linkages:
associative network linkages can be asymmetric, meaning that activation of the brand node may cause activation of the product node, while activation of the product node may not cause activation of the brand node and vice versa (Farquhar & Herr, 1993; Lei et al., 2008). So when a consumer encounters a negative product consumption experience and the associative network linkage is mainly directed from the brand node towards the product node, the negative product consumption experience will most likely only have consequences for product evaluation (not for brand evaluation). The amount of spill over is determined by the amount of brand relatedness and thereby the strength of the associative network linkage (John et al., 1998; Lei et al., 2008; Roehm & Tybout, 2006): the more a consumer perceives the product and the brand as related, the stronger the associative network linkage between the product node and the brand node is in the memory of this consumer. Stronger associative network linkages result in a greater amount of spill over. This means that when a strong associative network linkage exists that is mainly directed from the product node towards the brand node, a negative product consumption experience can heavily affect both product evaluation and brand evaluation.
2.3.2– Occurrence/creation of associative network linkages
To sum up the above, the two assumptions on which H1 is based will only hold when two requirements are met:1. There is a strong associative network linkage between the product node and the brand node in the memory of consumers (i.e. consumers perceive the product and the brand as related); 2. This strong associative network linkage is (at least) directed from the product node towards
But how are these requirements met? Most likely, these requirements are met when
organisations/product managers/brand managers follow a branding strategy in which the brand level drives the product level and the product level inherits the identity at the brand level (Aaker & Joachimsthaler, 2000). With such a strategy, consumers will also encounter the brand when they encounter the product (e.g. the brand is highly salient in product advertisements and is clearly visible on the product itself) and consumers do not perceive a major difference between the product identity and the brand identity (e.g. the product and the brand are both prestigious).
So when the branding strategy described above is followed by organisations/product managers/brand managers, the two requirements are met and the two assumptions on which H1 is based will hold: the defence mechanism may be triggered and the brand identification buffer may occur when a consumer with strong brand identification encounters a negative product consumption experience. This is rather obvious: when a product has its own identity and is not perceived as related to the brand, consumers with strong brand identification do not feel that their inner self is threatened when the distinct identity of the product is threatened. For example, when a consumer who strongly identifies him-/herself with the Seiko brand encounters a negative consumption experience with a Pulsar watch (Pulsar is a sub brand of Seiko), he/she will most probably not feel that the Seiko brand and the inner self are threatened. This is because it is highly likely that, in the memory of the consumer, no strong
associative network linkage exists between the Pulsar product node and the Seiko brand node (Pulsar has a different identity than Seiko and is completely decoupled from Seiko in product advertisements). But when a consumer who strongly identifies him-/herself with the Alfa Romeo brand encounters a negative consumption experience with a Mito car, he/she will most probably feel that the Alfa Romeo brand and the inner self are under attack and will start defending both.
The discussion above leads to one very important precondition for testing H1: participants in the research must perceive that the focal product is driven by the focal brand and must not perceive a major difference between the product identity and the brand identity. Once this precondition is fulfilled, it is more likely that the defence mechanism is triggered and the brand identification buffer occurs. But even when the defence mechanism is triggered and the brand identification buffer occurs, the impact of a negative incident at the product level on product evaluation will be relatively big compared with the impact on brand evaluation: previous research suggests that a spill over is < 100% (Lei et al., 2008). So on this base, the following hypothesis can be formulated:
H2: the impact of negative product consumption experience on brand evaluation is smaller than the impact on product evaluation, regardless of brand identification strength.
3. Method
To answer the research question of this study, the hypotheses have been quantitatively tested. The overall research design that was used to do this, is discussed together with the research strategy and research tactics in the first part of this chapter. Thereafter, the research population and research participants are described. In the following part, the research variables from a conceptual perspective are discussed (the scale items of the measures that were used to measure the variables and the questionnaire that was used to collect the needed data can be found in Appendix C and Appendix D respectively). The chapter is ended with a description of the data analysis, which includes a description of the variables from a statistical perspective.
3.1 – Overall research design
The research question and the hypotheses are about measuring the impact of negative product consumption experience on product evaluation plus brand evaluation. In general, impact is measured by measuring change. Because change always implies that there is a before condition and an after condition, both conditions need to be measured and compared with each other to validly measure change (Bono & McNamara, 2011). In the current research, it was important that product evaluation plus brand evaluation in the before condition were actually measured before and not after the focal negative product consumption experience. This is because consumers may be unable to evaluate the product plus brand as if they not encountered a negative product consumption experience (McQueen & Knussen, 2002). Therefore, an overall research design was used in which each participant
encountered the focal negative product consumption experience during the research and needed to evaluate the focal product plus focal brand before and after this experience. So – with the repeated measurement of product evaluation plus brand evaluation – within-subject change was measured. Therefore, the overall research design can be labelled as a within-subject design. But because participants belonged to the group weak identification with focal brand (in short: the group weak brand identification) or to the group strong identification with focal brand (in short: the group strong brand identification) and the subject change in the first group was compared with the within-subject change in the second group to actually test the hypotheses and answer the research question, the overall research design can also be labelled as a between-group design. This makes the overall research design – which is depicted in Figure 2 – a split-plot design or a mixed factorial design (Cotton, 2013; Edmonds & Kennedy, 2012).
Figure 2. Overall research design.
The main disadvantage of this mixed factorial overall research design, is the increased probability of participant bias: because of the repeated measurement of product evaluation plus brand evaluation, participants may have been aware that change was measured and may have tried to be “good
participants” by answering in favour of assumed hypotheses (Dane, 2010; Mitchell, Jolley, & O’Shea, 2009). So why was this design chosen instead of an “ordinary” between-group overall research design? With a between-group overall research design, before condition groups (weak brand identification and strong brand identification) and after condition groups (weak brand identification and strong brand identification) would have been compared to check whether change in product evaluation plus brand evaluation occurred. So between-group change would have been measured instead of within-subject change. Because there most probably would have been large group-to-group variance of product evaluation plus brand evaluation (e.g. weak brand identification in the before condition does not imply low product evaluation plus brand evaluation scores), the statistical power provided by a between-group overall research design would have been low compared with the statistical power provided by a mixed factorial overall research design (Seltman, 2013). That is why a mixed factorial overall research design was chosen in the current research.
To reduce the probability of the before mentioned participant bias, there should have been a few weeks of time between measuring product evaluation plus brand evaluation in the before condition and in the after condition. Although the few weeks of filler time would have increased the risk that uncontrolled extraneous factors influenced the results of the research, participants should have been less aware that change was measured. But given the time box in which the current research had to be completed, this approach was not feasible. Therefore, participants were given instructions to prevent participant bias: participants were clearly informed that honest responses were the only correct responses and that they should not try to be “good participants”.
The overall research design provides high-level insights into what the research looked like. But how were the data to actually test the hypotheses and answer the research question obtained? This question is answered below by providing insights into the research strategy and research tactics.
3.1.1 – Research strategy
There were no accessible secondary data sources that contained the data needed to test the hypotheses and answer the research question. So new primary data needed to be collected. There are multiple research strategies to collect data: survey, experiment, case study, ethnography, action research, and grounded theory (Saunders, Lewis, & Thornhill, 2009). The research strategy should match the nature of the research question and hypotheses at hand and should fit the overall research design (Saunders et al., 2009; Yin, 2009). In the current study, the research question and hypotheses are about “what is...” (instead of “how does...” or “why is...”) and with the overall research design, participants (above all) needed to encounter the focal negative product consumption experience during the research. This rendered the case study strategy and the ethnography strategy less suitable, because these strategies focus on the context of the phenomenon being studied and are therefore more suitable for “how does...” or “why is...” research questions and hypotheses (Morris & Wood, 1991; Saunders et al., 2009; Yin, 2009). The action research strategy and the grounded theory strategy seemed also
unsuitable, because the current research studies change (change is not an end-product as it is in action research strategy [Costello, 2003]) and the hypotheses are firmly anchored in existing theory
(induction is not relevant, while it is relevant in grounded theory strategy [Charmaz, 2006; Saunders et al., 2009]). The survey strategy and the experiment strategy did seem to be appropriate for the current research: the survey strategy is one of the few research strategies that is highly suitable for “what is...” research questions and hypotheses (Saunders et al., 2009; Yin, 2009) and although the experiment strategy is more suitable for “how does...” or “why is...” research questions and hypotheses (Saunders et al., 2009; Yin, 2009), it seamlessly fitted with the overall research design (simulation of the focal negative product consumption experience).
So at first glance, both the survey strategy and the experiment strategy seemed suitable to collect the needed data. Then the question to be answered was: which one is the most suitable? For the defence mechanism to be triggered and the brand identification buffer to occur, it is essential that consumers explain the focal negative product consumption experience in terms of the focal brand: consumers need to attribute the focal negative product consumption experience to the focal brand. All other possible explanations need to be ruled out. That is why in the previous chapter one very important precondition for testing H1 was mentioned. But even when this precondition is met, consumers may still reach an alternative explanation for the negative product consumption experience: consumers may think that the product is not handled/prepared correctly (e.g. food is not served in the correct manner) or is not consumed correctly (e.g. not conform the user manual). To rule out all such alternative explanations, a research strategy was needed that provided the control to steer participants to attribute the focal negative product consumption experience to the focal brand. The experiment strategy does not fully provide this control: participants with strong brand identification may be suspicious (in real-life they may have never encountered the negative product consumption experience) and may think that the product is sabotaged and is not handled/prepared correctly. The survey strategy does provide this control: in a survey, participants may be asked to imagine they have encountered a fictive case in real-life. This fictive case may contain a negative product consumption experience that is elaborated in such a way, that it is highly likely that the participants attribute the experience to the brand. Because the aim of the fictive case is to potentially trigger a change in product evaluation plus brand evaluation (as dependent variables), it gives the survey a strong experimental edge (Saunders et al., 2009).
Based on the above, a survey – in which a fictive case was incorporated – was conducted. Because of the fictive case, it may seem that the current research is similar to previous research that used negative brand information to trigger the defence mechanism of participants. But the current research is
completely different for two key reasons:
1. The current research asked participants to imagine they have encountered a fictive case that contained a private negative product consumption experience, while previous research asked participants to read public negative brand information in the form of, for instance, a
newspaper article;
2. The current research focused on the product level (negative product consumption experience and product evaluation plus brand evaluation), while previous research mainly focused on the brand level (negative brand information and brand evaluation).
3.1.2 – Research tactics
The research tactics consisted of the following components: research technique, research procedure, focal product plus focal brand, and focal negative product consumption experience (fictive case).
Research technique
Several research techniques belong to the survey strategy (Saunders et al., 2009): structured
observation, structured interviews (or interviewer-administered questionnaires), and questionnaires. Structured observation is all about quantifying behaviour (Saunders et al., 2009) and was therefore not suitable for the current research (evaluation needed to be examined, not behaviour). Although
structured interviews and questionnaires are quite similar, structured interviews are more open for researcher bias (e.g. the tone of voice of the interviewer may affect how the interviewee respond) and participant bias (e.g. the interviewee may not be comfortable to answer personal questions face-to-face and may therefore give censored answers [Saunders et al., 2009]). Because of this and the fact that questionnaires are more time-efficient (Saunders et al., 2009), a questionnaire was used. Specifically, a delivery-and-collection hard-copy questionnaire was used so that the researcher was able to directly approach and ask people who were consuming a product of the focal brand or a product of a
competing brand to fill out the questionnaire immediately. This way, the probability that enough people that weakly or strongly identified themselves with the focal brand were reached, was increased (later on in this chapter, the sampling logic is discussed in more detail).
The questionnaire was in Dutch (because Dutch was the native language of most of the participants), meaning that the measures and the fictive case needed to be translated. In the translation process, the utmost care was given to lexical meaning, idiomatic meaning, and experiential meaning (Saunders et al., 2009). A pilot test of the questionnaire (N = 10) revealed that participants did not have any trouble interpreting the information in the questionnaire and completing the measures.
Research procedure
The people who were approached by the researcher have been asked to participate in “a scientific research about product evaluation plus brand evaluation”. To persuade these people to participate in the research, they were told that participation would only take 10 minutes and would only entail filling out a two-part questionnaire. The people who agreed to participate, received the first part of the questionnaire. Two-thirds through the data collection process however, this procedure was changed by asking people who agreed to participate to first complete the measure of the brand identification variable (depending on the brand identification score, they then received the first part of the questionnaire). This change in procedure is explained further during the discussion of the research participants later on in this chapter. The first part of the questionnaire first provided additional information about the total questionnaire. Information such as instructions on completing the
questionnaire and information about confidentiality/anonymity. It then asked participants to carefully read neutral information about the focal product to ensure that all participants were familiar with this product before they completed any measure. Thereafter, the participants were asked to complete a
procedure) and measures of the product evaluation variables and brand evaluation variables (the before condition). After completing the first part, participants were given the second part of the questionnaire. This second part asked the participants to carefully imagine they have encountered the fictive case and thereafter asked them to complete measures of (again) the product evaluation variables and brand evaluation variables (the after condition). Finally, it asked participants to complete measures of the control variables. After participants completed the second part of the questionnaire, the researcher stapled the two parts of each participant together. Then the participants were thanked.
To prevent any researcher or participant bias (e.g. participants may not have been comfortable answering questions while the researcher waits in front of them), the researcher left when participants started to fill out the first part and second part of the questionnaire.
Focal product plus focal brand The focal product needed to be:
1. From a real-life brand that some consumers use to extend their inner self;
2. Perceived by consumers as being driven by the brand and as having a similar identity as the brand (the precondition [for testing H1] mentioned in the previous chapter);
3. From a brand that consumers distinguish from the product (not the Maggi brand for instance).
The iMac desktop computer from the Apple brand may meet all three requirements. The first requirement may be met because the Apple brand is in the top five consumer electronic brands that consumers use to extend their inner self (Swisher, 2014). On top of this, there is a lot of merchandise that signals the Apple brand in some way or another (like t-shirts, caps, and clocks [CafePress, 2014]). With this merchandise, consumers are encouraged to extend their inner self with the Apple brand and can let significant others know that they want to be identified with the brand. The second requirement may be met because it is highly likely that consumers perceive the iMac computer as being driven by the Apple brand and do not perceive a major difference between the identity of the computer and the brand. This is because the Apple brand is highly visible on the iMac computer (not only on the hardware, but also while running the pre-installed operating system [Apple, 2014]). The third
requirement may be met because the product and the brand do not share the same name and because of the fact that it is well known that different products belong to the Apple brand (e.g. iMac computer, iPhone smart phone, and iPad tablet).
Based on the above, the iMac computer was used as the focal product and Apple was used as the focal brand.
Focal negative product consumption experience (fictive case)
The following fictive case was incorporated in the second part of the questionnaire:
Imagine that you have just bought an iMac desktop computer in the official Apple Store on the internet. Because the official Apple Store ships directly from the official Apple warehouse to the home address of customers, delivery is scheduled in only two days. Exactly two days later, the door bell rings. You open the door and a delivery guy carefully holds the package that contains your new iMac. You check the package for transportation damage; the package looks like the iMac was transported with the utmost care and the “shock sticker” (which tells whether the package has taken a beat during transportation) confirms this. So you sign the delivery note and take the package inside. You carefully unpack the iMac. You place the computer on a sturdy desk (so nothing will happen to the computer) and install it exactly as is described in the user manual. After you have pressed the power button, the computer boots up and a welcome message appears. You perform a test of the iMac and come to the conclusion that it works exactly as it should work!
A week later, you turn on the iMac again and you notice a strange sound inside the computer. Also, the computer is too slow to work on. Then an error message pops up that says: “Critical error in Apple hard drive (error code: 5938.0XE). Please contact AppleCare!”. After a few seconds, the iMac computer automatically shuts down and cannot be turned on again. You follow the instruction and call AppleCare. You provide the error code to the Apple employee to whom you speak. The Apple employee tells you that the hard drive is produced by Apple itself and that the error is caused by a fault in the production process.
3.2 – Research participants
Because participants needed to encounter product consumption experience and the focal product belonged to the product category of computers, it may be suggested that the research population was “consumers who consume/use computers”. But since the majority of people consume/use computers (especially in western societies [OECD, 2014]), these consumers may be representative of all consumers. Also, people that not consume/use computers may nevertheless identify themselves with the focal brand and evaluate the focal product plus focal brand (these people may well be capable to imagine they have encountered the fictive case). Therefore, the research population was defined as broad as “consumers in general”. To test the hypotheses and answer the research question, this population was divided into two strata: the stratum weak identification with focal brand (in short: the stratum weak brand identification) and the stratum strong identification with focal brand (in short: the
people in each of the strata were asked to fill out the questionnaire. But because this was not feasible, a sample of each of the strata was drawn and was asked to fill out the questionnaire. These samples of the strata filled the two groups of the overall research design: the group weak brand identification and the group strong brand identification.
But how exactly was a sample of each of the strata drawn? In defining the sampling logic, it was important to note that:
• The proportion of each of the strata in the population did not have to be reflected in the research sample: the research required that approximately 50% of the participants within the entire research sample were from the stratum weak brand identification and that the other 50% were from the stratum strong brand identification;
• Social consumer psychology had to be examined (brand identification is a social psychological process): the research did not set any requirements regarding the demographics of participants, meaning that there were no objections to only include university students that belong to one of the two strata in the research sample (Bono & McNamara, 2011).
Based on the above, the researcher randomly approached students of the University of Amsterdam who likely belonged to the stratum weak brand identification until approximately 75 people that actually belonged to this stratum filled out the questionnaire. The researcher did the same for the stratum strong brand identification. The likeliness that a person belonged to a stratum was determined by looking at the product this person used. For instance, when a person used a product from the focal brand, it was more likely that this person belonged to the stratum strong brand identification than to the stratum weak brand identification. Whether this person actually belonged to the stratum strong brand identification was determined after the person agreed to participate in the research and
completed the questionnaire by looking at the brand identification score. As is discussed later on, the brand identification score is an average of the seven scale items (rated on 7-point Likert scales) of the measure of the brand identification variable. Participants with brand identification scores < 4 belonged to the stratum weak brand identification, participants with brand identification scores ≥ 4 belonged to the stratum strong brand identification.
But by looking at the brand identification score after participants completed the questionnaire, too many questionnaires were filled out by persons that belonged to the stratum weak brand identification. Therefore it was (two-thirds through the data collection process) decided to detach the measure of the brand identification variable from the questionnaire itself and use it as a selection tool: it allowed the researcher to only ask people that actually belonged to the stratum strong brand identification to fill out the questionnaire.
In the end, 152 participants completely filled out the questionnaire (because only 12 participants did not fill out the questionnaire completely, these participants were excluded from the research). Of these 152 participants, 78 participants (45% male, 15% owned the focal product, 82% [also] owned other products from the focal brand) belonged to the stratum weak brand identification and thus filled the group weak brand identification. The remaining 74 participants (45% male, 28% owned the focal product, 100% [also] owned other products from the focal brand) belonged to the stratum strong brand identification and thus filled the group strong brand identification.
3.3 – Research variables
From a conceptual perspective, the current research involved four types of variables: independent variables, dependent variables, moderating variables, and control variables. This is captured in the conceptual model (that can be found in Appendix B). Each variable type, the variables that belonged to these types, and the measures that were used to measure the variables are discussed below. The scale items of the measures are not provided below, but can be found in Appendix C. These scale items were included as questions in the questionnaire. The total questionnaire can be found in Appendix D.
It is important to note that these variable types and variables differ from the ones that were used in the data analysis to actually test the hypotheses. The variable types and variables from a statistical
perspective are described later on in this chapter (in the description of the data analysis).
3.3.1 – Independent variables
The only independent variable was the focal negative product consumption experience. The focal negative product consumption experience was not measured: it was the research stimulus (contained in the fictive case) that gave the survey a strong experimental edge.
3.3.2 – Dependent variables
The current research consisted of four dependent variables: two product evaluation variables and two brand evaluation variables.
Product evaluation
Product attitude. The overall attitude towards the focal product was measured using the attitude measure (α = .88) developed by Mitchell and Olson (1981). Although this measure – that consists of
measure brand attitude, product attitude research used the same scale items: some product attitude research used the scale items goodness and pleasantness (e.g. Smith and Swinyard [1988]), while other product attitude research used the scale items pleasantness and liking (e.g. Leclerc, Schmitt, and Dubé [1994]). In the current research, two adjustments were made to the original measure of Mitchell and Olson (1981). First, the scale item favourability was added so that the bigger part of the scale items measured overall attitudes and not only the hedonic or utilitarian component of attitudes (Batra & Ahtola, 1991). Second, all five scale items were rated on 7-point Likert scales instead of on 5-point Likert scales. This gave assurance that even the slightest attitude change was noticed. The scale items were averaged to form one attitude score per participant.
Product purchase intention. The intention to purchase the focal product was measured using a measure that consists of one scale item. This scale item entails the overall likelihood of purchase and has been used widely in previous research that included purchase intention as a variable (e.g. Morrison [1979] and Smith and Swinyard [1988]). Following Smith and Swinyard (1988), the scale item was rated on a 7-point Likert scale.
Brand evaluation
Brand attitude. The overall attitude towards the focal brand was measured with the same measure that was used in the current research to measure attitude towards the focal product. This is because the measure is an adjusted version of the original one developed by Mitchell and Olson (1981) to measure brand attitude. All five scale items were rated on 7-point Likert scales and were averaged to form one attitude score per participant.
Brand purchase intention. The intention to purchase other products from the focal brand than the focal product was measured with the same measure that was used in the current research to measure
intention to purchase the focal product. The only scale item of this measure was rated on a 7-point Likert scale.
3.3.3 – Moderating variables
Brand identification was the only moderating variable in this research. Identification with the focal brand was measured with the brand identification measure (α = .90) developed by Escalas (2004). This measure has been used widely in previous research (e.g. Lisjak et al. [2012] and Sprott, Czellar, and Spangenberg [2009]) and consists of seven scale items. All seven scale items were rated on 7-point Likert scales and were averaged to form one brand identification score per participant.
3.3.4 – Control variables
As discussed in the previous chapter, the defence mechanism is more easily triggered (and the brand identification buffer more likely to occur) for people who are more defensive by nature. Because implicit self-esteem and explicit self-esteem regulate the natural defensiveness of people, they are important contextual factors and were therefore included as control variables. Ownership of the focal product (product ownership) and ownership of other products from the focal brand (brand ownership) may also be important contextual factors: people who own the focal product and/or other products from the focal brand may have encountered positive product consumption experiences, which may affect the impact of the focal negative product consumption experience. But because the researcher used brand ownership to determine whether to ask a person to participate in the research, only product ownership was included as a control variable. Gender and age were the demographic control variables in the current research.
Implicit self-esteem. Following Lisjak et al. (2012), the measure developed by Gebauer, Riketta, Broemer, and Maio (2008) was used to measure implicit self-esteem. This measure consists of one scale item: full-name name-liking. This scale item was rated on a 9-point Likert scale.
Explicit self-esteem. Instead of following Lisjak et al. (2012) by using the Rosenberg self-esteem measure (RSE, developed by Rosenberg [1965]), the single-item self-esteem measure (SISE, developed by Robins, Hendin, and Trzesniewski [2001]) was used to measure explicit self-esteem. This is because the SISE has been proven reliable and valid and is therefore a practical alternative to the RSE that consists of 10 scale items (Robins et al., 2001). The scale item of the SISE was rated on a 7-point Likert scale.
Product ownership. Ownership of the focal product was measured using a measure that consists of one scale item. This scale item was rated on a dichotomous scale.
Gender. Gender was measured using a measure that consists of one scale item. This scale item was rated on a dichotomous scale.
Age. Age was measured using a measure that consists of one scale item. This scale item was rated on a ratio scale.
3.4 – Data analysis
The data analysis consisted of two parts. The purpose of the first part was to get a grasp of the data. This was done by carrying out analyses that included descriptive analysis (e.g. calculation of the mean and standard deviation for all the variables) and correlation analysis (e.g. testing for significant correlations between each of the variables using Pearson correlation coefficient). Both the descriptive analysis and the correlation analysis were conducted for the total sample, the group weak brand identification, and the group strong brand identification. To carry out these and all other analyses in this research for each of the groups, a new categorical brand identification variable was computed (as discussed before, brand identification scores < 4 indicated weak brand identification, while brand identification scores ≥ 4 indicated strong brand identification). The purpose of the second part of the data analysis was to test whether significant within-subject change, significant between-group difference in within-subject change (H1), and significant within-subject difference in within-subject change (H2) occurred. The analyses that were carried out to do this, are described below together with the variable types and variables from a statistical perspective. The variable types and variables from a statistical perspective are not the same as the ones from a conceptual perspective. The latter have been discussed before in this chapter.
3.4.1 – Within-subject change
To check whether within-subject change in product evaluation plus brand evaluation occurred, new within-subject change variables were computed by subtracting the before condition scores from the after condition scores for each of the within-subject variables (product attitude, product purchase intention, brand attitude, and brand purchase intention). So this computation resulted in the following change variables: product attitude change, product purchase intention change, brand attitude change, and brand purchase intention change. The means of these four new change variables indicated whether within-subject change occurred. To check whether this within-subject change was significant, a one-tailed paired samples t-test was carried out for each of the within-subject variables (not the change variables) for the total sample, the group weak brand identification, and the group strong brand identification. Paired samples t-tests were suitable because they allowed to pair the repeated measurement of each of the within-subject variables (i.e. the before condition scores and the after condition scores of the same participants) and to test whether the scores significantly differed. So the within-subject variables acted as dependent variables in these tests.
3.4.2 – Between-group comparison of within-subject change
To check whether between-group difference in within-subject change in product evaluation plus brand evaluation occurred, the means of the four change variables in the group weak brand identification
were compared with the means of the four change variables in the group strong brand identification. Testing H1 involved testing whether this between-group difference was significant. Normally, with a mixed factorial overall research design as was used in the current research, mixed design ANOVAs are carried out to do this (Edmonds & Kennedy, 2012; Field, 2009). But mixed design ANOVAs are only more helpful than independent samples t-tests or one-way ANOVAs (these analyses allow to test for a significant difference in the scores of two not related groups [Field, 2009]) when the
measurement of the within-subject variables is repeated two times or more. The measurement of the within-subject variables was only repeated once in the current research, which allowed for the computation of the previously discussed change variables. Because of this and the need for one-tailed testing (H1 predicts directionality in results), a one-tailed independent samples t-test was carried out for each of the four change variables to test H1 (ANOVAs are not suitable for one-tailed testing [Cohen, 2013]). In these tests, the categorical brand identification variable acted as independent variable, while the change variables acted as dependent variables.
By using the categorical brand identification variable as the independent variable in the independent samples t-tests, information was lost (e.g. the difference between a brand identification score of 4 and a brand identification score of 7 was not taken into account). Because of this, a relationship between the independent variable and the dependent variables may have been overlooked (i.e. low statistical power [Cohen, 1990; Mitchell & Jolley, 2012]). Also, independent samples t-tests do not allow testing with control variables. So to validate the results of the independent samples t-tests and thereby reveal whether the control variables in the current research caused the between-group difference in within-subject change, a hierarchical multiple regression analysis was carried out for each of the four change variables: the uncategorised brand identification variable and the control variables acted as
independent (predictor) variables, while the change variables acted as dependent variables. The four analyses were carried out three times (three runs). The purpose of the first run was to determine which of the control variables (implicit self-esteem, explicit self-esteem, product ownership, gender, and age) significantly contributed to explaining the variance of one or more change variables and therefore needed to be included in the analyses in the second run and third run. The purpose of the second run was to determine whether there were any outlying and influential participants that threatened the accuracy of the analyses and therefore needed to be excluded from the analyses in the third run. This was determined using the following diagnostic statistics: Cook’s distance coefficient, Mahalanobis distance coefficient, DFBeta coefficient, centred leverage value, and covariance ratio. The analyses in the third run were the main analyses.
3.4.3 – Within-subject comparison of within-subject change
To check whether the within-subject change in product evaluation differed from the within-subject change in brand evaluation, new within-subject difference variables were computed by subtracting the scores of the product attitude change variable and the product purchase intention change variable from the scores of the brand attitude change variable and the brand purchase intention change variable respectively. So this computation resulted in the following difference variables: difference product brand attitude change and difference product brand purchase intention change. The means of these two new difference variables indicated whether within-subject difference in within-subject change
occurred. Testing H2 involved testing whether this within-subject difference was significant. This was done by carrying out one-tailed paired samples t-tests for the total sample, the group weak brand identification, and the group strong brand identification. Paired samples t-tests were suitable, because they allowed to pair the scores of the brand attitude change variable and brand purchase intention change variable with the scores of the product attitude change variable and product purchase intention change variable respectively (and thereafter to test whether these scores significantly differed). So the change variables (not the difference variables) acted as dependent variables in the paired samples t-tests. Pairing was possible, because brand attitude was measured with the same measure that was used to measure product attitude and brand purchase intention was measured with the same measure that was used to measure product purchase intention.
