University of Florida Thesis Or Dissertation Formatting

IMPLICIT ATTITUDES AS PRECURSORS OF SPORT CONSUMER BEHAVIOR: PREDICTIVE AND INCREMENTAL VALIDITY OF DUAL ATTITUDE IN ATHLETE ENDORSEMENT EVALUATION

YONGHWAN CHANG

A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2016

To Il Rang, Chloe & Claire

ACKNOWLEDGMENTS

Numerous people have provided invaluable support for the completion of my dissertation. First of all, I would like to express my deepest appreciation to the committee chairperson, Dr. Yong Jae Ko. I am truly indebted for his advice, encouragement, tremendous support, and opportunities he provided for last years at the University of Florida. I cannot find words to express my wholehearted appreciation for him. I would also thank other committee members, Drs. Michael Sagas, Brian Mills, and Chris Janiszewski, who have been really encouraging and supportive. I am also very grateful that I have worked with the best colleagues at the University of Florida and the College of Health and Human Performance. Most importantly, I deeply thank my family and friends.

TABLE OF CONTENTS

page

ACKNOWLEDGMENTS ...... 4

LIST OF TABLES ...... 9

LIST OF FIGURES ...... 10

LIST OF TERMS ...... 12

ABSTRACT ...... 14

CHAPTER

1 INTRODUCTION ...... 16

The Significance of Athlete Endorsement in the Sport Industry ...... 16 Understanding Endorsement Effectiveness ...... 17 Theoretical Approaches for Evaluating Endorsement Effectiveness ...... 18 Dual Processing Accounts of Attitude Construction in Athlete Endorsement ...... 20 Statement of the Problem ...... 22 Purpose of the Study ...... 23 Significance of the Study ...... 24 Structure of the Study ...... 24 Research Questions ...... 26 Hypotheses ...... 26

2 THEORETICAL BACKGROUND...... 29

Athlete Endorsement as a Co-Branding Practice ...... 29 Existing Theoretical Approaches in Endorsement Literature ...... 30 Source Models ...... 30 Schema Congruity Theory and Matchup Hypothesis ...... 32 Associative Network Memory Model ...... 33 Association Strength and Endorsement Effectiveness ...... 35 The Concept of Association Strength ...... 35 Association Strength and the Two-Step Model for Judgment and Decision Making .....36 Selective Attention and Endorsement Effectiveness ...... 38 The Concept of Selective Attention ...... 38 The Process of Selective Attention ...... 39 Biased competition model ...... 39 Adaptive learning theory ...... 40 Level of Fit and Selective Attention ...... 41 Dual Process of Evaluative Judgment Construction in Athlete Endorsement ...... 43 Summary ...... 51

3 EXPERIMENT ...... 52

Experimental Design ...... 52 Pretests ...... 52 Instrumentation ...... 54 Explicit Cognitive Attitude toward Athlete and Product ...... 54 Explicit Affective Attitude toward Athlete and Product ...... 55 Other Explicit Measures ...... 56 Procedures of Main Experiment ...... 56 Fit Manipulation ...... 56 Single Target-Implicit Association Test (ST-IAT) ...... 57 Data Analysis ...... 60 Data Screening: Missing Data, Outliers, Leverage Points, and Normality ...... 60 Descriptive Statistics ...... 61 Measurement Model Test ...... 61 Data Preparation of Latencies in the ST-IAT ...... 62 Hypotheses Testing ...... 63

4 RESULTS ...... 72

Demographics ...... 72 Measurement Model Test ...... 72 Cognitive explicit attitude toward athlete (expertise, trustworthiness, attractiveness) ...... 72 Affective explicit attitude toward athlete (pleasure, arousal, pride) ...... 72 Cognitive explicit attitude toward product (quality, trustworthiness, attractiveness) ...... 73 Affective explicit attitude toward product (pleasure, arousal, pride) ...... 73 A Summary of Measurement Model Tests ...... 73 Moderation Effects of Athlete Involvement on Implicit Attitude toward Athlete ...... 74 Moderation Effects of Athlete Involvement on Cognitive Explicit Attitude toward Athlete ...... 74 Moderation Effects of Athlete Involvement on Athlete Expertise ...... 75 Moderation Effects of Athlete Involvement on Athlete Trustworthiness ...... 76 Moderation Effects of Athlete Involvement on Athlete Attractiveness ...... 76 Moderation Effects of Athlete Involvement on Explicit Affective Attitude toward Athlete ...... 77 Moderation Effects of Athlete Involvement on Athlete Pleasure ...... 78 Moderation Effects of Athlete Involvement on Athlete Arousal ...... 79 Moderation Effects of Athlete Involvement on Athlete Pride ...... 79 A Summary of the Results of Moderation Effects of Athlete Involvement ...... 80 Moderation Effects of Product Involvement on Implicit Attitude toward Product ...... 81 Moderation Effects of Product Involvement on Explicit Cognitive Attitude toward Product ...... 81 Moderation Effects of Product Involvement on Product Quality ...... 82 Moderation Effects of Product Involvement on Product Trustworthiness ...... 83 Moderation Effects of Product Involvement on Product Attractiveness ...... 83 Moderation Effects of Product Involvement on Explicit Affective Attitude toward Product ...... 84

Moderation Effects of Product Involvement on Product Pleasure ...... 85 Moderation Effects of Product Involvement on Product Arousal ...... 86 Moderation Effects of Product Involvement on Product Pride ...... 86 A Summary of the Results of Moderation Effects of Product Involvement ...... 87 Path Analysis: Moderated Mediation Effects of Athlete Involvement on Implicit Attitude, Explicit Cognitive Attitude, and Behavioral Intention...... 88 Path Analysis: Moderated Mediation Effects of Athlete Involvement on Implicit Attitude, Explicit Affective Attitude, and Behavioral Intention ...... 89 Path Analysis: Moderated Mediation Effects of Product Involvement on Implicit Attitude, Explicit Cognitive Attitude, and Behavioral Intention...... 91 Path Analysis: Moderated Mediation Effects of Product Involvement on Implicit Attitude, Explicit Affective Attitude, and Behavioral Intention ...... 92

5 DISCUSSION ...... 139

Summary of the Study ...... 139 Theoretical Implications ...... 140 Fit and Attitudes Constructions ...... 140 Involvement and Attitudes Constructions ...... 143 The Causal Effects of Fit on Dual Attitudes (Implicit and Explicit) ...... 145 The Causal Relationships: Associative-Propositional-Intention (API) Model ...... 146 Practical Implications ...... 148 Limitations and Future Suggestions ...... 149 Conclusion ...... 150

APPENDIX

A LETTER OF CONSENT FORM ...... 151

B DEMOGRAPHIC INFORMATION QUESTIONNAIRE ...... 152

C MANIPULATION OF FIT – EXAMPLE STIMULI ...... 153

D INSTRUCTIONS FOR IMPLICIT ASSOCIATION TEST ...... 154

E TWO SINGLE TARGET IMPLICIT ASSOCIATION TESTS – SAMPLE PROCEDURE...... 155

F EXPLICIT MEASURES – ATHLETE EVALUATION ...... 156

G EXPLICIT MEASURES – PRODUCT EVALUATION...... 157

H EXPLICIT MEASURES – SNS ...... 158

I EXPLICIT MEASURES – INVOLVEMENT AND AFFECT...... 159

J EXPLICIT MEASURES – CONNECTION ...... 160

K EXPLICIT MEASURES - FIT ...... 161

LIST OF REFERENCES ...... 162

BIOGRAPHICAL SKETCH ...... 171

LIST OF TABLES

Table page

3-1 Examples of Product Brands List used for Pretest 1...... 66

3-2 Celebrity Athletes List used for Pretest 1 ...... 67

3-3 Preliminarily Selected Athlete and Product Brands and Their Matches ...... 69

3-4 Selected Athlete and Product Brands and Their Matches for Main Experiment ...... 70

3-5 Category Assignment and Stimulus Proportions across Five ST-IAT Blocks in Main Experiment ...... 71

4-1 Measurement Model Test: Cognitive Explicit Attitude toward Athlete ...... 94

4-2 Measurement Model Test: Affective Explicit Attitude toward Athlete ...... 95

4-3 Measurement Model Test: Cognitive Explicit Attitude toward Product ...... 96

4-4 Measurement Model Test: Affective Explicit Attitude toward Product ...... 97

4-5 A Summary of the Results of Measurement Model Tests ...... 98

4-6 Factor Correlations of Explicit Cognitive Attitude toward Athlete ...... 99

4-7 Factor Correlations of Explicit Affective Attitude toward Athlete ...... 100

4-8 Factor Correlations of Explicit Cognitive Attitude toward Product ...... 101

4-9 Factor Correlations of Explicit Affective Attitude toward Product ...... 102

4-10 A Summary of the Results of the Moderated Mediation Analysis: Implicit Attitude, Explicit “Cognitive” Evaluation toward “Athlete”, and Behavioral Intention ...... 103

4-11 A Summary of the Results of the Moderated Mediation Analysis: Implicit Attitude, Explicit “Affective” Evaluation toward “Athlete”, and Behavioral Intention ...... 104

4-12 A Summary of the Results of the Moderated Mediation Analysis: Implicit Attitude, Explicit “Cognitive” Evaluation toward “Product”, and Behavioral Intention ...... 105

4-13 A Summary of the Results of the Moderated Mediation Analysis: Implicit Attitude, Explicit “Affective” Evaluation toward “Product”, and Behavioral Intention ...... 106

LIST OF FIGURES

Figure page

4-1 Stimuli for main experiment ...... 107

4-2 A summary of a comparision of predictions between traditional and selective attention/elaboration approaches ...... 112

4-3 A summary of a comparison between the theories of reasoned action and planned behavior and the hypothesized associative-propositional-intention (API model) ...... 113

4-4 A hypothetical process of consumers’ evaluative judgment construction through elaboration (selective attention mechanism) ...... 114

4-5 Moderation effects of athlete involvement on implicit attitude toward athlete (Z- scored) ...... 115

4-6 Moderation effects of athlete involvement on cognitive explicit attitude toward athlete (Z-scored) ...... 116

4-7 Moderation effects of athlete involvement on athlete expertise (Z-scored) ...... 117

4-8 Moderation effects of athlete involvement on athlete trustworthiness (Z-scored) ...... 118

4-9 Moderation effects of athlete involvement on athlete attractiveness (Z-scored) ...... 119

4-10 Moderation effects of athlete involvement on affective explicit attitude toward athlete (Z-scored) ...... 120

4-11 Moderation effects of athlete involvement on athlete pleasure (Z-scored) ...... 121

4-12 Moderation effects of athlete involvement on athlete arousal (Z-scored) ...... 122

4-13 Moderation effects of athlete involvement on athlete pride (Z-scored) ...... 123

4-14 Moderation effects of product involvement on implicit attitude toward product (Z- scored) ...... 124

4-15 Moderation effects of product involvement on cognitive explicit attitude toward product (Z-scored) ...... 125

4-16 Moderation effects of product involvement on product quality (Z-scored) ...... 126

4-17 Moderation effects of product involvement on product trustworthiness (Z-scored) ...... 127

4-18 Moderation effects of product involvement on product attractiveness (Z-scored) ...... 128

4-19 Moderation effects of product involvement on affective explicit attitude toward product (Z-scored) ...... 129

4-20 Moderation effects of product involvement on product pleasure (Z-scored) ...... 130

4-21 Moderation effects of product involvement on product arousal (Z-scored) ...... 131

4-22 Moderation effects of product involvement on product pride (Z-scored) ...... 132

4-23 A summary of results of the moderation effects of “product” involvement ...... 133

4-24 A summary of results of the moderation effects of “athlete” involvement ...... 134

4-25 A summary of the results of moderated mediation analysis (explicit “cognitive” attitudes) ...... 135

4-26 A summary of the results of moderated mediation analysis (explicit “affective” attitudes) ...... 136

4-27 A summary of the results of moderated mediation analysis (explicit “cognitive” attitudes) ...... 137

4-28 A summary of the results of moderated mediation analysis (explicit “affective” attitudes) ...... 138

LIST OF TERMS

ASSOCIATION The strength or intensity of the connection between brands STRENGTH

ASSOCIATION Consumers’ perceived intensity of the connection between athlete and STRENGTH IN product brands caused by their existing associative memory network

ATHLETE of the two brands; association strength between brands influences the ENDORSEMENT extent to which consumers selectively attend to endorsements

A hypothetical evaluation process dealing with unconscious and ASSOCIATIVE automatic levels of evaluative judgment, which can be observed EVALUATION through implicit measure

As objects in the visual input inevitably compete for cell responses, such as mental representation, analysis or control, to interpret the BIASED inputted information due to limited human capacity for information

COMPETITION processing, attention causes a representational bias toward the competitive interactions of multiple stimuli, and therefore, attended stimuli often take a priority over unattended stimuli

Attentional control that is driven by factors that are external to the BOTTOM UP BIAS observer such as stimulus salience

Also known as brand association or alliance; equity-creation process, CO-BRANDING where both associations of celebrity endorsers and the endorsed product brands are equity drivers

EXPLICIT Evaluative judgments about the encountered stimulus, which are

ATTITUDE processed through propositional reasoning

EXPLICIT A self-reported test, given that explicit measure involves deliberative

MEASURE evaluation

One of the most popular methods for assessing implicit attitude; it was IMPLICIT designed to measure implicit attitudes by assessing their underlying ASSOCIATION automatic evaluation, affect or attitude based on the assumption that TEST (IAT) reaction time toward a stimulus of strongly associated pairs tends to be faster than that of weakly associated pairs

Unconscious and automatic levels of evaluative judgments when a IMPLICIT person encounters a relevant stimulus, which is referred to as ATTITUDE associative evaluation

Measurement outcomes that are causally produced by the to-be- IMPLICIT measured attribute in the absence of certain goals, awareness, MEASURE substantial cognitive resources, or substantial time

Selective allocation of cognitive effort to a specific stimulus due to the SELECTIVE three components of attention, including alertness, capacity, and ATTENTION selectivity

Consumers’ selective allocation of information processing resources to SELECTIVE a particular type of endorsement and information in the endorsement ATTENTION IN due to increased arousal and limited cognitive capacity; this attention ATHLETE may result in vigilant/sensitive consumer responses toward the ENDORSEMENT attended endorsement

SINGLE TARGET An application of the original implicit association test (IAT); ST-IAT IMPLICIT is especially useful in the case of measuring multiple single-targets ASSOCIATION without the need to simultaneously evaluate a counter category TEST (ST-IAT)

Attentional control that is driven by factors that are internal to the TOP DOWN BIAS observer such as relevancy of a stimulus

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

IMPLICIT ATTITUDES AS PRECURSORS OF SPORT CONSUMER BEHAVIOR: PREDICTIVE AND INCREMENTAL VALIDITY OF DUAL ATTITUDE IN ATHLETE ENDORSEMENT EVALUATION

Yonghwan Chang

August 2016

Chair: Yong Jae Ko Major: Health and Human Performance

Over the past two decades of endorsement and sponsorship scholarship, one of the general conclusions that has emerged is that consumers’ attitudes form and change as a consequence of strategically paired endorsements because these pairs induce image transfers in consumers’ minds from athlete to product and vice versa. Accordingly, it has been known that fit between endorsers/sponsors and endorsed/sponsored properties leads to greater image and meaning transfers, which ultimately results in favorable endorsement/sponsorship evaluations.

However, through their attentional and elaboration processes, consumers may also develop positive endorsement/sponsorship evaluations toward highly incongruent pairings between endorsers/sponsors and the properties.

In addition, numerous scholars have suggested that, along with explicit memory, implicit memory may also play a major role in the processing of endorsement and sponsorship information. Yet, there is a lack of research examining the extent to which endorsers and brands are associated in consumers’ mind, and how this association systematically influences their attentional/elaboration process, as well as their implicit and explicit evaluative judgments.

Accordingly, this study is designed to fill any explanatory gaps that have been traditionally less

explored when considering the interplay of implicit/automatic/unconscious and explicit/deliberate/conscious attitudes in predicting sport consumer behavior in the context of athlete endorsement. Four separate studies were conducted; the main experiment employed

Single Target Implicit Association Test (ST-IAT) as a measure of implicit attitudes. In contrast to the predictions suggested from traditional dual processing accounts, the results support the existence as well as the incremental validity of dual attitudes. Implicit attitudes shaped through associative evaluation toward both cognitive dissonant and consonant stimuli were dynamically magnified through propositional reasoning. The results also support the proposed hierarchical model of associative propositional intention (API), including stimuli exposure, implicit attitudes, explicit cognitive and affective attitudes, and behavioral intention. This study seeks to go beyond existing theoretical approaches in the given context by providing novel theoretical insights and empirical support.

CHAPTER 1 INTRODUCTION

The Significance of Athlete Endorsement in the Sport Industry

The general belief among business managers is that sport sponsorship, including celebrity

athlete endorsements, can generate intangible (e.g., enhancement of image and prestige

perceptions) and tangible benefits (e.g., financial returns) given its key role as one of the most

promising marketing tools. Indeed, researchers have found that sport sponsorship, including

celebrity athlete endorsements, results in favorable advertisement ratings and product

evaluations, which ultimately have a substantial positive impact on financial returns for the

companies (Swerdlow & Swerdlow, 2003). According to an IEG Sponsorship Report, sport

sponsorship spending in North America was approximately $13.79 billion in 2013 – an increase

of 6% over 2012; additionally, of all the sponsorships reported during 2013, sports sponsorships

accounted for 69%, while entertainments and arts accounted for only 10% and 5%, respectively.

Particularly, sport celebrities have been considered to be attractive endorsers because

they represent a healthy, strong, vigorous, enthusiastic, and energetic image that the agencies

pursue (Bush, Martin, & Bush, 2004). A celebrity athlete endorser refers to a famous individual,

usually a well-known athlete or coach, who promotes a brand through their name recognition

(Stafford, Spears, & Hsu, 2003). Sport celebrities are one of the most profitable assets in today’s

sports industry because they significantly increase fans’ interest in sports, events, and leagues

(Turner, 2004). They receive a substantial level of media attention, which creates numerous business opportunities through such channels as athlete endorsements (Charbonneau & Garland,

2005).

In reality, most top professional athletes make considerably higher level of financial success when compared to other athletes in their sport; one of the main sources of their incomes

has become endorsement contracts. For example, Danica Patrick (professional athlete in motor

sport) is the top product endorser among the National Association for Stock Car Auto Racing

(i.e., NASCAR) players, and annually earns over $10 millions from endorsement contracts with

Go Daddy (Internet domain registrar and web hosting company, Scottsdale, AZ), Chevrolet

(Chevrolet Division of General Motors LLC, Detroit, MI), Coca-Cola (Atlanta, GA), Nationwide

Insurance (insurance and financial services company, Columbus, Ohio), and Tissot (Swiss

manufacturer of luxury watch, the Swatch Group Ltd., Le Locle, Switzerland). Similarly, Sports

Illustrated (sports magazine, Time Inc., NY) recently released its annual "Fortune 50" list of the

highest-paid athletes in 2013, documenting that the top 10 list is filled with athletes (e.g., Lebron

James) who make substantially more financial earnings from endorsements than they do from

participating in their chosen sport.

Understanding Endorsement Effectiveness

Given that marketers worldwide consider athlete endorsement an effective promotional

tool, several scholars (e.g., Ohanian, 1991) have identified certain conditions that maximize

endorsement effectiveness. In particular, these scholars have identified the level of perceived fit

between endorsers and endorsed products as one of these key conditions. From a brand

marketing perspective, understanding athlete-company/product brand fit involves identifying the

network of “brand associations” that connects the consumers’ memories to both the endorsers

and the endorsed products. The associations, in the form of retrieval cues, play an important role

in both consumers’ product evaluation and choices (Cunha, Janiszewski, & Laran, 2008).

Researchers suggest that this condition significantly influences consumers’ response (i.e., beliefs,

attitude, and behavior) to the endorsement activation (Till & Busler, 2000). Thus, understanding

these associations becomes more important in the case of brand extension or co-branding, where

two or more brands are strategically paired (Thomson, 2006).

In the context of athlete endorsements, consumers have previously established knowledge structures corresponding to both product brands and celebrity endorsers; when they encounter a particular endorsement contract, the two knowledge structures are associated within a single network. In this re-organizational process, it can be expected that the “association strength” between athlete endorsers and endorsed product brands stored in consumers’ memory network may simultaneously change through image transfer from a celebrity to a product or vice versa.

This phenomenon occurs because the strength and nature of the individual brand image varies across endorsers and product brands (Seno & Lukas, 2007).

Theoretical Approaches for Evaluating Endorsement Effectiveness

Scholars have suggested various approaches to identify the complex network of associations present in consumers’ memory; furthermore, they seek to explain how the strength of multiple associations changes over time. In the context of celebrity endorsement, these approaches include: source models (Ohanian, 1991; Trampe, Stapel, Siero, & Mulder, 2010), matchup hypothesis (Till & Busler, 2000), schema congruity theory (Meyers-Levy & Tybout,

1989), and associative network memory models (Till & Shimp, 1998).

The source credibility and attractiveness (Ohanian, 1991; Trampe et al., 2010) models have been dominant in celebrity endorsement studies. However, this approach focuses only on uni-directional image influence from the athlete endorsers to product brands. In the fields of business and marketing, scholars applied the schema congruity theory (SCT: Meyers-Levy &

Tybout, 1989) to sponsorship contexts as a means to explain the connections between sponsors and sponsored products. In the context of athlete endorsement, a schema is an active organization structure of the associations of a brand. Consumers’ evaluations of an endorser rely on the level of congruity between the characteristics of the athlete and those of the schema associated with the endorsed product (Till & Busler, 2000).

The matchup hypothesis (Kamins, 1990; Till & Busler, 2000) was also developed based

on the SCT. The matchup hypothesis provides a macroscopic explanation (i.e., schema level),

instead of accounting for the microscopic aspect (i.e., node level) of image matching between

celebrity endorsers and product brands, and meaning transfer from an endorser to a brand

(Madden et al., 2000). In this process, consumers’ newly formed memory structures become

linked with the celebrity and the endorsed brand in their pre-existing memory (Chen, Lin, &

Hsiao, 2012). However, these two approaches (i.e., SCT and matchup hypothesis) would be limited in their ability to explain which specific attributes of an endorser and a product brand are associated on a node level.

As an alternative approach, the associative network memory model (ANMM) was proposed by Till and Shimp (1998). The ANMM explains the consumers’ detailed memory process weighting on individual nodes in the consumer’s memory structure. According to Till and Shimp (1998), human memory can be described as “a network consisting of various nodes connected by associative links” (p. 68). These nodes are pieces of information that become connected through a variety of associative links (Spry, Pappu, & Cornwell, 2011). Each node is a potential source of activation for all associated nodes (Biswas, Biswas, & Das, 2006). In the marketing field, the ANMM has been frequently used to explain the consumers’ memory structure (Till & Nowak, 2000), brand associations (Keller, 1993; Ozsomer & Altaras, 2008),

and endorsement effectiveness (Spry et al., 2011). Specific attributions of a celebrity endorser

(e.g., expertise) and a product brand (e.g., quality) are believed to represent individual nodes

associated into an association set (Spry et al., 2011).

Against this background, recently, Chang, Ko, Tasci, Arai, and Kim (2014) argued that it is necessary to fully explain the extent to which different types and levels of brand associations

exist and the ways in which brand associations bi-directionally influence each other, and ultimately affect association strength between the brands. In addition, these existing approaches

also lack an explanation of consumers’ varied endorsement evaluations due to their selective

attention toward given endorsement information.

To clearly explain these phenomena, Chang et al. (2014) examined multiple cues of

athlete endorsement by using the adaptive learning theory (Cunha et al., 2008). The authors

found that American consumers placed a higher weight on endorsers’ attractiveness while, Asian

consumers focused on athletic performance and ranking of athlete endorsers. This result

indicates that consumers often selectively attend to specific information due to their limited

capacity of processing available information. The researchers supported the basic assumption of

adaptive learning theory asserting that the extents to which brands are associated in consumers’

mind systematically influences the consumers’ attention and learning process as well as their

product brand evaluation.

Dual Processing Accounts of Attitude Construction in Athlete Endorsement

In the fields of (sport) marketing and consumer behavior, a substantial amount of existing

studies predominantly assume that consumers’ evaluative judgment is a unitary construct

weighing on deliberate and effortful processes. For example, traditional dual processing accounts

(e.g., elaboration likelihood model; Petty, Cacioppo, & Schumann, 1983 or heuristic systematic

model; Eagly & Chaiken, 1984) suggest that as soon as an attitude is newly constructed through

peripheral or deliberate processing, only the newly shaped attitude can remain tarnishing the

original attitude (Dempsey & Mitchell, 2010). However, contemporary attitude research (e.g.,

Gawronski & Bodenhausen, 2006; Petty, 2006; Petty, Brinol, & DeMarree, 2007) suggests that

the processes of consumers’ judgment and decision-making could inherently originate from

automatic and unconscious processes that occur extemporaneously and outside of conscious

awareness or control. In addition, evaluative judgments of positive and negative valences toward an object may exist together in consumers’ minds, and the strengths of the judgments also can vary depending on the associative network in their memory (Petty et al., 2007).

The associative-propositional evaluation model (APE; Gawronski & Bodenhausen, 2006) suggests that consumers’ evaluative tendencies can be rooted in two types of mental processes: associative evaluation (implicit attitudes) and propositional reasoning (explicit attitudes).

Following the APE, associative evaluation processes represent unconscious and automatic levels of evaluation processing, while propositional reasoning can be understood as syllogistic inferences derived from a reflective evaluation system. Therefore, the processes of propositional reasoning are generally relevant whether evaluations, beliefs, or attitudes are true or false, while associative processes are characterized by the mere activation of existing associations relevant to an attitude object independent of subjective truth or falsity.

This emerging theoretical perspective of dual attitude could provide meaningful contributions especially to the context of athlete endorsement for various reasons. First, athlete endorsement can be understood as a co-branding process in which two complex knowledge structures are simultaneously integrated within a single network (Thomson, 2006); therefore, identifications of consumers’ associative processes are especially critical to accurately explain and predict endorsement effectiveness. Second, the perspectives of dual attitude provide innovative and radically different predictions of consumer evaluation of endorsement effectiveness. For example, theories of selective attention (Berger, Henik, & Rafal, 2005;

Theeuwes, 2010; Yantis, 2008) suggest that highly incongruent pairings between an endorser and a property could induce involuntary consumer attention and elaboration of automatic/associative processes (bottom-up or exogenous control of attention), which in turn, could lead to processing

fluency (Lee & Labroo, 2004) as well as even favorable evaluations in a subsequent judgment

(Janiszewski, Kuo, & Tavassoli, 2013); this prediction contradicts the results derived from the

most of existing endorsement studies developed based on the source models (Ohanian, 1991), the

matchup hypothesis (Till & Busler, 2000), the schema congruity theory (Meyers-Levy & Tybout,

1989), and the associative network memory models (Till & Shimp, 1998). Third, if associative

evaluation indeed reflects highly stable evaluative representations that are rooted in long-term socialization experiences, the outcomes of associative evaluation should generally be more robust than the results of propositional reasoning in predicting actual consumption behavior

(Gawronski & Strack, 2004), which has seldom been explored in the fields of sport marketing

and management.

Statement of the Problem

Over the past two decades of endorsement and sponsorship scholarship, various research

models and theoretical approaches have been suggested to account for endorsement

effectiveness. While these academic efforts have allowed scholars to draw meaningful

conclusions and find fruitful insights, there are several areas to be further explored in the field of

endorsement and sponsorship study. First, existing endorsement and sponsorship literature

supports that the fit between endorsers/sponsors and sponsored/endorsed properties leads to

greater image and meaning transfers and ultimately results in favorable endorsement/sponsorship

evaluation. However, little research has been done to provide a comprehensive and systematic

understanding of how consumers’ attentional/elaborational processes influence their evaluative judgment in the context of athlete endorsement. Through their attentional and elaboration processes, consumers may also develop positive endorsement/sponsorship evaluations toward highly incongruent pairings between endorsers/sponsors and the properties.

Second, from a brand marketing perspective, it is necessary to identify the “complex

network” of brand associations that connects consumers’ memories of athlete endorsers and product brands. Particularly, the concept of association strength has been regarded as one of the most fruitful constructs in explaining such complex memory network as brand association; thus, parting from Keller’s (1993) seminal work (consumer-based brand equity paradigm), this concept has been extensively researched in general marketing and branding research. However, few attempts have been made to apply this concept in the context of athlete endorsement.

Third, numerous scholars have increasingly suggested that research on evaluative judgments as precursors of behavior should be further extended through the joint application of both conscious and unconscious levels of judgment, because these two often display incremental validity in explaining actual behavior (Greenwald et al., 2009). Following this trend, several scholars (e.g., Cornwell, Weeks, & Roy, 2005) have suggested that implicit memory may play a major role in the processing of sponsorship information. Yet, there is little research examining how endorsement information systematically flows into the consumers’ associative and propositional processes; in fact, to date, only a handful of scholars have empirically investigated consumers’ implicit memory, with most studies mainly focusing solely on explicit attitudes.

Purpose of the Study

Accordingly, this study is designed to illuminate the gap by accounting for any emergence in respect to the interactions between explicit and implicit attitudes in the context of athlete endorsement. The specific objectives of this study are to: (1) examine the causal effects of fit between an endorser and an endorsed brand on both consumers’ implicit and explicit attitudes, and (2) examine the moderating effects of consumer involvement (both athlete and product) in the relationship between fit and both implicit and explicit attitudes. In addition, to provide a comprehensive theoretical understanding of the sequential processes of consumer evaluation, the researcher attempts to (3) develop and empirically test the associative propositional intention

model (API), including four hierarchically ordered evaluation processes (i.e., stimuli exposure,

associative evaluation, propositional reasoning, and behavioral intention).

Significance of the Study

The researcher argues that existing endorsement/sponsorship studies developed based on

the aforementioned research approaches—source models, schema congruity theory, matchup

hypothesis, and associative network memory model—may need to be further explored.

Particulalry, the researcher shows contradicting predictions of endorsement/sponsorship effectiveness by systematically incorporating existing theories in the fields of psychology and marketing, including theories of selective attention (e.g., biased competition model).

From a methodological and measurement point of view, the current study first introduces the implicit association test (IAT) in the fields of sport management and marketing; in fact, the same type of measurement for attitude (i.e., single target-implicit association test; ST-IAT) has not been applied to date. In the (sport) consumer behavior area, most studies have solely employed explicit attitude, which could not explain variances beyond the conscious level of consumer evaluation. Therefore, through an examination of consumers’ both implicit and explicit attitude, the current study may help optimize an accurate prediction of (sport) consumer behavior through the theoretical integrations of alternative approaches.

Structure of the Study

Chapter 1 presented the main background of this study, including significance of athlete endorsement in sport industries, understanding endorsement effectiveness from a co-branding perspective; it also offered a brief summary of existing endorsement studies and their theoretical and practical limitations in explaining endorsement effectiveness. Along with the limitations/problems statements, the researcher attempts to justify why selective attention, elaboration, and related theoretical approaches are critical in both theoretical and practical

examinations of endorsement effectiveness. Then, the researcher presented the objectives of the

current study, specific research questions and hypotheses, in addition to discussing the

contributions of the current study to the fields of sport marketing and management.

Chapter 2 presents a theoretical background of the study. The researcher approaches

athlete endorsement from a co-branding perspective. The chapter discusses existing theoretical

applications of associative network memory models to athlete endorsement and their limitations.

At the same time, the researcher discusses a need for adopting association strength and selective

attention constructs to fully address endorsement effectiveness. From a theoretical standpoint,

the researcher incorporates existing (neuro) psychological theory – biased competition model– in

addition to existing marketing theories, including adaptive learning, availability-valence

hypothesis, and the two-step model of judgment and evaluation. All of these theoretical

adoptions and applications may help provide a comprehensive understanding of endorsement

effectiveness. Lastly, this research adopts and applies the associative-propositional evaluation

model toward developing a way to explain perceived endorsement effectiveness. Particularly, the researcher argues that fit manipulation systematically influences consumers’ implicit and explicit attitude constructions in the context of athlete endorsement. Through this adoption and application, the current study may contribute to optimizing an accurate prediction of (sport) consumer behavior and identify interactions that enhance desired consumption behaviors.

Chapter 3 describes the experiment. The researcher briefly reintroduces the objectives of the experiment and re-establishes the research hypotheses; this then leads to a specific description of the documented experimental design, the stimuli development through a series of pretests, and the instrumentations. Moreover, the chapter offers experimental procedures related to association strength manipulation and two single target-implicit association tests (ST-IATs).

The researcher also includes a specific procedure of data analysis, including descriptive statistics,

procedures involved with data screening (e.g., missing, outliers, leverage point, and linearity),

measurement model testing, and data preparations of latencies in the ST-IAT. The particular

procedures conducted for analysis of variance (ANOVA) and moderated mediation analysis for hypothesis testing are also presented. Chapter 4 provides results of empirical examinations of the

hypotheses as well as of further inquiries about causal relationships among implicit, explicit, and

intention. Chapter 5 discusses theoretical and practical implications, as well as limitations and

suggestions for future scholarship.

Research Questions

The following research questions were developed for the purposes of this study:

1. How does fit or misfit impact on consumers’ dual attitude?

2. What mechanisms underlie in regards to the concepts of fit, association strength, selective

attention, and elaboration in explaining endorsement effectiveness?

3. How do implicit and explicit attitudes theoretically interplay in consumers’ evaluation of

athlete endorsement? What theoretical implications would such a difference is able to

generate?

Hypotheses

To answer the research questions and fill the theoretical gaps existing in endorsement studies, following hypotheses were formulated:

H1-1: Fit between athlete endorsers and endorsed brands significantly influences consumers’ level of elaboration and evaluative judgment. Specifically, the relationship between fit and evaluative judgment is U-shaped, where the highest and lowest levels of fit lead to greater elaboration as well as more favorable evaluative judgment.

H1-2: Involvement level in athlete endorsers and product brands moderates the relationship

between fit and consumers’ evaluative judgment. Specifically, for consumers with low involvement, this relationship is U-shaped, where the highest and lowest levels of fit lead to greater elaboration, as well as more favorable evaluative judgment. On the other hand, high involvement consumers, regardless of fit, will show greater elaboration as well as favorable evaluative judgment.

To further explain perceived endorsement effectiveness, the author extended the hypotheses 1-1 and 1-2 by incorporating the central tenet of the associative-propositional evaluation model (Gawronski & Bodenhausen, 2006):

H2-1: Fit between athlete endorsers and endorsed brands positively influences consumers’ processing fluency, which in turn, leads to more attention and elaboration, which in turn, leads to initially favorable implicit attitudes. Therefore, high fit rather than moderate fit will show more favorable implicit attitudes.

The initially developed favorable implicit attitudes are likely to be magnified through the propositional reasoning processes. Therefore:

H2-2: High fit, rather than moderate fit, will increase favorable explicit attitudes to a greater extent when compared to implicit attitudes.

Low fit between athlete endorsers and endorsed brands is likely to increase consumers’ cognitive dissonance, which in turn, leads to unfavorable explicit attitudes. However:

H3-1: Low fit, rather than moderate fit, leads to more attention and elaboration, which in turn, leads to enhanced association strength of an attitude object in their memory, which in turn, leads to processing fluency as well as favorable evaluation in a subsequent evaluation. Therefore, low fit rather than moderate fit will show more favorable implicit attitudes.

The initially developed favorable implicit attitudes may be memorized in different

storage from the unfavorable explicit attitude formed through the propositional reasoning processes. Accordingly:

H3-2: Low fit, rather than moderate fit, will increase favorable implicit attitudes to a greater extent when compared to explicit attitudes.

In consideration of the top-down control of selective attention (Theeuwes, 2010; Yantis,

2008) as well as the central-route processing model (Cornwell, Weeks, & Roy, 2005),

H4: Compared to consumers with low involvement, highly involved consumers will show consistently favorable evaluative judgments implicitly and explicitly regardless of fit manipulations.

In addition, integrating the associative-propositional evaluation model (Gawronski &

Bodenhausen, 2006), and the theories of reasoned action and planned behavior (Ajzen, 1991), the researcher developed the associative propositional intention model (API). The API includes four hierarchically ordered evaluation processes: (1) observation and recognition of a stimulus – stimuli exposure, (2) associative evaluation – implicit attitudes, (3) propositional reasoning – explicit cognitive and affective attitudes, and (4) behavioral intention. Specific hypothesis was formulated as follows:

H5: Stimuli exposure automatically leads to associative evaluation, which in turn, influences propositional reasoning, which in turn, leads to behavioral intention.

CHAPTER 2 THEORETICAL BACKGROUND

Athlete Endorsement as a Co-Branding Practice

The effective use of celebrity endorsers enables brands to capture the consumers’ attention, thus, helping the brand stand out from its competitors (Ilicic & Webster, 2012).

Consumers recognize celebrities as human brands, given that celebrities now actively employ brand management techniques, such as managing, trademarking and licensing their names, launching their own product lines, and making product endorsement deals; such branding

practices enhance consumer’s perceived value toward the celebrity, and endorsed brands and

their brand equity (Thomson, 2006). Therefore, branding researchers consider celebrities as co-

branding partners, where athletes and products brands are strategically paired with one another

(Ilicic & Webster, 2012).

Co-branding—also known as brand association or alliance—refers to an equity-creation process, where both associations of celebrity endorsers and the endorsed product brands are equity drivers. A celebrity endorser has the potential to affect the brand equity of the endorsed product by amplifying the product’s brand image associations; at the same time, a product brand also has the potential to affect the equity of the endorsing celebrity by augmenting the celebrity’s image associations (Seno & Lukas, 2007). Hence, when consumers hold strong and/or positive associations of celebrities, those associations transfer to the endorsed product brand. A set of associations can also be transferred from the products to endorsers.

In the field of sport marketing, several researchers have started to consider athletes as brands (Arai, Ko, & Ross, 2014; Chang et al., 2014). As a result, from a co-branding perspective, research on proactive branding efforts using athlete endorsement is still in its infancy. In particular, there has not been systematic research on the strategic match of athlete endorsement

from both product and the athlete brand marketing standpoints.

Existing Theoretical Approaches in Endorsement Literature

Given that marketers worldwide consider athlete endorsement an effective promotional tool, several scholars have identified certain key conditions that maximize endorsement effectiveness (Ohanian, 1991). In particular, the level of perceived fit between endorsers and endorsed products has been identified as one of these critical conditions. For example, Till and

Busler (2000) found that this condition significantly influences consumers’ response (i.e., beliefs, attitude, and behavioral intention) to the endorsement activation. From a brand marketing perspective, understanding the athlete-brand fit involves identifying the network of “brand associations” that connects the consumers’ memories to endorsers and the endorsed product. The associations, in the form of retrieval cues, play an important role in both consumers’ product evaluation and choices (Cunha, Janiszewski, & Laran, 2008). Thus, understanding these associations becomes more important in the brand extension or co-branding contexts, where two or more brands are strategically paired (Thomson, 2006).

Scholars have suggested various approaches to identify the complex network of associations in consumers’ memory; furthermore, they seek to explain how the strength of multiple associations changes over time. In the context of celebrity endorsement, these approaches include: source models (Ohanian, 1991; Trampe, Stapel, Siero, & Mulder, 2010), matchup hypothesis (Till & Busler, 2000), schema congruity theory (Meyers-Levy & Tybout,

1989), and associative network memory models (Till & Shimp, 1998).

Source Models

The source credibility and attractiveness models (Ohanian, 1991; Trampe et al., 2010) have been dominant in celebrity endorsement studies. Source credibility refers to the influence the perceived level of expertise and trustworthiness of a celebrity endorser has on the

effectiveness of a message delivery (Ohanian, 1991). The source attractiveness model assumes

that the effectiveness of a message delivery depends on the perceived level of the physical

attractiveness of a celebrity endorser (Till & Busler, 2000). Therefore, the two source models are

fundamentally based on the normative theory; it is a matter of determining whether the message

a celebrity endorser delivers would be effectively persuasive as he/she holds positive image

associations. In other words, these models assume a linear relationship between consumer

responses toward endorsement and the meaning and image a celebrity endorser tries to deliver,

and the extent to which a celebrity endorser is perceived as having expertise, trustworthiness, and

attractiveness.

Over the past two decades of endorsement literature, source models have provided a

fruitful insight in understanding endorsement effectiveness. Recently, however, several scholars

criticized these models for its limited explanatory power. First, this approach does not take into

account the multidimensional aspect of a celebrity endorser as a persuasive source in the

endorsement. In other words, merely focusing on several attributes of celebrity endorsers is

inappropriate or incomplete as consumers may be able to retrieve numerous image associations

through spreading activation when they come across a specific endorsement (Fleck, Korchia, &

Roy, 2012).

Second, this approach focuses only on uni-directional image influence from the athlete

endorsers to product brands. The emergence of the concept of co-branding in recent marketing research and practices has highlighted the bi-directional process of equity-creation in celebrity endorsement, where both associations of celebrity endorsers and the endorsed product brands are joint equity drivers. In reality, not only does a celebrity endorser have the potential to affect the brand equity of the endorsed product by amplifying the product’s brand image associations, but

also a product brand has the potential to affect the equity of the endorsing celebrity by augmenting the celebrity’s image associations (Seno & Lukas, 2007). Hence, beyond the source models, it would be appropriate to further consider bi-directional image influence to provide a comprehensive understanding of endorsement effectiveness.

Schema Congruity Theory and Matchup Hypothesis

In the business and marketing literature, scholars applied the schema congruity theory

(SCT: Meyers-Levy & Tybout, 1989) to the context of sponsorships to explain the connections in consumers’ mind between sponsors and sponsored products. In the context of athlete endorsement, a schema has been understood as an active organization structure of associations a brand or endorser holds; consumers’ evaluations of an endorser rely on the level of congruity between the characteristics of the athlete and those of the schema associated with the endorsed product (Till & Busler, 2000).

Later, scholars (Kamins, 1990; Till & Busler, 2000) developed the matchup hypothesis based on the SCT. This hypothesis assumes that the effectiveness of the message delivery using celebrity endorsers depends on the perceived congruity of schematic information between celebrity endorser and endorsed product brands. This approach assumes that regardless of source credibility and attractiveness, the message is persuasive when the message a celebrity endorser delivers (1) can contribute to a clear identification of an endorsed brand or (2) fits into consumers’ existing schematic memory structure toward an endorsed brand. As an alternative model overcoming the limitations of source models discussed above, scholars and business managers have adopted this approach as an efficient way to understand endorsement effectiveness and have provided useful information for strategic matching efforts.

Nonetheless, the matchup hypothesis and the schema congruity theory also have limitations in fully explaining which specific attributes of an endorser and a product brand are

associated on a node level. Given that this approach mainly focuses on a macroscopic

explanation (i.e., schema level), it does not account for the microscopic aspects (i.e., node level) of image matching or meaning transfer between celebrity endorsers and product brands (Madden et al., 2000). For example, based on the matchup hypothesis, it would be improbable to image the more complex and detailed levels of relationship as consumers’ newly formed memory structures become actively linked in their pre-existing memory with the celebrity and the endorsed brand (Chen, Lin, & Hsiao, 2012). Along with this complex memory activations and network, a certain level of misfit may also have a positive impact on message persuasiveness and consumer responses in contrast to the basic assumption derived from the matchup hypothesis; consequently, Till and Shimp (1998) introduced the associative network memory model

(ANMM) to overcome such limitations.

Associative Network Memory Model

The ANMM was proposed by Till and Shimp (1998) to explain the consumers’ detailed memory process weighting on individual nodes in the consumer’s memory structure. According to Till and Shimp (1998), this approach is particularly significant since human memory can be described as “a network consisting of various nodes connected by associative links” (p. 68).

These nodes are pieces of information that become connected through a variety of associative links, and each node is a potential source of activation for all associated nodes (Collins & Loftus,

1975). Thus, in the context of athlete endorsement, by spreading activation, the product brand name may not only stimulate other associations the product brand possesses, such as price and quality, but also endorser associations, such as expertise and attractiveness. Going back to Till and Shimp’s (1998) conceptualization of spreading activation, it is clear that specific attributions of a celebrity endorser and a product brand represent individual nodes and they are associated into an association set.

Recently, extending the ANMM, Chang et al. (2014) argued that the extent to which

different types and levels of complex brand associations exist in consumers’ memory. Going

further, the authors suggested that those associations simultaneously and bi-directionally influence each other; the bi-directional networks induce different levels of association strength between brands, and ultimately affect the consumers’ attentional process and responses toward an endorsement. In practice, consumers may not be able to fully comprehend the endorsements when they encounter endorsement contracts because of the complex relationships of multiple associations of brands. Instead, they may focus on a particular type of endorsement contract due to limited cognitive resources, existing levels of consumers’ arousal and involvement, or time constraints. Chang et al. (2014) showed systematic differences in consumers’ endorsement evaluation beyond the debates of sampling bias caused by unadjusted covariates distribution.

Another important aspect that remains to be further explored is the role of fit in determining endorsement effectiveness. Specifically, one of the commonly agreed conclusions

that has emerged from existing endorsement scholarship is that fit between endorsers/sponsors and sponsored/endorsed properties, rather than misfit, leads to greater image and meaning transfers, which ultimately results in favorable endorsement/sponsorship evaluation. However, several studies (e.g., Fleck, Korchia, & Roy, 2012) suggest that highly misfit between endorsers/sponsors and properties can also lead to greater cognition and attention among consumers, as well as their positive endorsement/sponsorship evaluations.

To face these challenges and overcome the drawbacks inherent in the existing endorsement literature, it would be necessary to seek out more comprehensive and robust theoretical paradigms. Considering this growing need of the alternative approach, the researcher proposes that, in the constructions of evaluative judgment, consumers would selectively pay

attention to particular aspects of endorsements, which may induce varied consumers’ attentional/elaborational processes and responses toward the endorsements. In a complex learning environment like athlete endorsements, consumers may selectively attend to and adaptively learn about the endorsements due to varied levels of endorsements and consumer characteristics, such as association strength of endorsement and product involvement. To fully address these assumptions, the researcher adopts and applies emerging theoretical approaches

(i.e., the association strength, selective attention, adaptive learning, and dual process of evaluative judgment) to the context of athlete endorsement. They are discussed in details below.

Association Strength and Endorsement Effectiveness

The Concept of Association Strength

Association strength refers to the strength or intensity of the connection between brands

(Keller, 1993). Hutchison (2003) conceptualized association strength as the percentage of people responding to a target brand when given a cue brand. Conceptually, the stronger an association, the more likely it will be recalled from memory via the spreading activation process that underpins mental maps (French & Smith, 2013; Keller, 1993). According to Cunha et al. (2008), the association strength between cues (e.g., brand name) and outcomes (e.g., quality perception) updates and evolves as cues interact with those outcomes. In addition, cues often compete against each other to predict outcomes; that is, the strength of the association between a cue and an outcome depends on how uniquely the cue can predict the outcome.

In the context of athlete endorsement, Chang et al. (2014) suggested that the association strength between athlete and product brands plays a central role in consumers’ selective attention toward endorsement and endorsement evaluation. It is predicted because by spreading activation, the product brand name may not only stimulate other associations related to the product brand, such as price and quality, but also endorser associations, such as expertise and attractiveness.

Therefore, the spreading activation is caused by the fact that specific attributions of a celebrity

endorser and a product brand represent individual nodes, and are associated into an association

set. As such, understanding the concept of association strength between brands further gains

significance, given that the co-branding perspective suggests that athlete endorsement may

involve complex relationships of multiple brand associations. Despite the central role association

strength plays in the consumers’ decision-making process, to date, athlete endorsement scholars

have made few attempts to examine the role of association strength between the brands when

explaining and predicting endorsement evaluation. Based on this theoretical background,

association strength in athlete endorsement can be defined as consumers’ perceived intensity of

the connection between athlete and product brands caused by their existing associative memory

network of the two brands. The researcher also assumes that association strength between brands

influences the extent to which consumers selectively attend to endorsements.

Association Strength and the Two-Step Model for Judgment and Decision Making

Among numerous antecedents influencing association strength between athlete and

product brands in associative memory networks suggested in the existing literature, scholars identify several key elements in an effort to quantitatively measure this latent construct. For example, early psychologist Mackintosh (1975) suggested four components of association strength in the relationship between a single cue and a single outcome: (1) cue accessibility, which refers to the relevance of the cue to the outcome; (2) learning rate parameter or quality of processing, which refers to how fluently a person can process the relationship between a cue and an outcome; (3) experienced outcome level; and (4) expected outcome level by the cue.

Recently, in consumer behavior literature, French and Smith (2013) suggested that association strength of a single brand can be measured by quantifying both (a) the number of association a brand possesses and (b) the number of maximum links a brand could have. As such, it seems that

there is little agreement on the determinants of association strength due to variability in approach

and its dimensions. Nonetheless, one general agreement is that, in people’s minds, the more

commonalities exist between objects, the greater the association strength can be found between

them.

The researcher assumes that there are two manageable elements which function as key

determinants of association strength: (i) product category congruency and (ii) brand-specific association alignability. Category congruency could be baseline similarity influencing association strength because it may be the easiest way to diagnose a brand; additionally, it may represent one of the most accessible points of information about the brand. While category congruency deals with a superficial level of similarities, a brand-specific association involves analytical, reason-oriented, logical, and detailed levels of similarities. A brand-specific association is defined as an attribute that differentiates a brand from competing brands

(Broniarczyk & Alba, 1994). Brand-specific association alignability as a semantic level of similarity may positively influence association strength.

The two-step model for judgment and decision-making (Smith, Shoben & Rips, 1974) further explains the two types of similarities as key elements of association strength between brands. In the first step, a consumer tries to quickly match the characters of the category to the characters of the object. When there is a clear fit or a clear misfit, the process is complete. When some features match and some do not match, a second stage of processing is needed. During the second stage, a more careful judgment of the defining characters of the category (e.g., brand- specific associations) is made to determine if the object is a member of the category (Smith et al.,

1974). Therefore, on one hand, endorsement evaluation involving product type is decided in the first stage, which is a rapid and global process. On the other hand, how much endorsement

evaluation and brand-specific associations are involved is determined in the second stage, which is slower and more deliberate. Hence, consumers may evaluate the two similarities in a sequential process; in other words, the two may not be comparable, but they interact with each other on association strength.

Selective Attention and Endorsement Effectiveness

The Concept of Selective Attention

Consumer attention can be understood according to the seminal work of early psychologists Posner and Boies’s (1971). The authors suggested three components of attention: alertness, capacity, and selectivity. Following their conceptualization, alertness presumably involves human ability to perform long and boring tasks, like vigilance; alertness also refers to the subject’s sensitivity to external stimulation (e.g., response time). Capacity mechanism suggests that any two operations requiring attention will interfere with each other, which can be compared to muscle function. Selectivity refers to the ability to select information from one source or of one kind rather than another. In the field of marketing, it seems that attention is interchangeably used with selectivity, one of the three dimensions. In the present study, the researcher assumes alertness highlights the somewhat methodological/measurement aspect of attention; capacity may point out causes of attention, while selectivity may be closer to the results aspect of attention.

Based on this understanding, the researcher assumes that, in the context of athlete endorsement, attention refers to consumers’ selective allocation of information processing resources to a particular type of endorsement and information in the endorsement due to increased arousal and limited cognitive capacity; this attention may result in vigilant/sensitive consumer responses toward the attended endorsement.

The Process of Selective Attention

Biased competition model

It seems that recent studies in the consumer behavior area are shifting their weight from the traditional “account of preference-from-information” to the “account of preference-from- process” (Janiszewski, Kuo, & Tavassoli, 2013). For example, the concepts of feature matching

(Tversky, 1972) from psychology and focus-shift models from marketing (Dhar, Nowlis,

Sherman, 1999) explain how initial consumers’ preference to specific features/attributes through selective attention and inattention influences their subsequent judgment. Theories of selective attention provide a fundamental understanding of early preference formation and its influence on the subsequent judgmental process as one of main themes in the preference-from-process area.

For example, in a situation where two stimuli are presented at the same time requiring separate responses, these two stimuli independently activate their neuronal representations, and then they compete with each other for the separate responses.

The rationale behind the competition is that, following the biased competition model in the field of neuropsychology (Desimone & Duncan, 1995), objects in the visual input inevitably compete for cell responses, such as mental representation, analysis or control, to interpret the inputted information due to limited human capacity for information processing. The biased competition model also suggests that attention causes a representational bias toward the competitive interactions of multiple stimuli, and therefore, attended stimuli often take a priority over unattended stimuli (Theeuwes, 2010). Two major causes of the attentional effects on resolving this competition include: (1) bottom-up neural mechanisms for object selection (or exogenous control) and (2) top-down control of selection (or endogenous control). Bottom-up bias is derived from the learned biases of the perceptual systems toward certain types of stimuli and represents largely automatic processes. Top-down bias stems from pre-existing knowledge

or situational relevance toward the stimuli (e.g., relevancy of a stimulus). For example, novelty

or unconventional endorsement contract induce involuntary attention and bottom-up bias, while

directing attention to relevant stimuli are understood as voluntary and goal-driven selections

causing top-down bias.

Adaptive learning theory

Along with this neuropsychological understanding of the process of selective attention, in the field of marketing, a number of studies have attempted to explain the mechanisms that underlie consumers’ attentional processes as well as the ways in which attention influences consumers’ preference formation. Recently, Janiszewski, Kuo, and Tavassoli (2013) suggested a paintbrush metaphor to describe the visual attention process. The authors assumed that when

visual environments are simple, easy to interpret, and relevant to each other, there may be

enough paintbrushes to paint everything in detail on a canvas, so the entire scene can be

elegantly and quickly painted. Therefore, the level of attention can be expressed as: (1) how

much the detail associated with certain objects/ locations is painted on a canvas, and (2) how

quickly and smoothly the act of painting is completed.

Adaptive learning theory (Cunha et al., 2008) further explains consumers’ attentional process in the situation of multiple brand associations. As briefly documented above, adaptive learning is a relatively emerging model in consumer behavior literature, and it is a recently introduced theoretical application in the domain of brand extension research, including celebrity endorsement. The fundamental assumption of adaptive learning is that consumers selectively attend to brand names or other attributes as retrieval cues to predict product performance when the attributes efficiently function as cues for consumers to gather diagnostic information about the product (Cunha et al. 2008). Hence, in terms of athlete endorsement, consumers may selectively attend to specific types/characters of endorsements intentionally, or sometimes

unintentionally. For example, in the case of an endorsement contract between a well-known athlete endorser and a newly launched brand, people may initially and exclusively attend to the athlete, rather than the product brand itself, as a way to predict the equity of the product brand.

Level of Fit and Selective Attention

In endorsement literature, various models and theoretical applications have been suggested to account for the constructions of consumers’ evaluative judgment and attitude toward endorsements; some of these suggestions were derived from source models (Ohanian,

1991), the matchup hypothesis (Kamins & Gupta, 2006), and the schema congruity theory

(McDaniel, 1999). Over the past two decades of endorsement and sponsorship scholarship, one of the general conclusions that has emerged is that consumers’ attitude forms and changes as a consequence of strategically paired endorsements because this pair induces image transfers in consumers’ mind from athlete to product and vice versa. From this traditional point of view,

Keller (1993) suggested that image transfers could take place when two or more brands are linked because the linkage enables consumers to connect the preexisting image attributes of the brands held in their memories. Finally, fit between endorsers/sponsors and sponsored/endorsed

properties, rather than misfit, leads to greater image and meaning transfers, which ultimately

results in favorable endorsement/sponsorship evaluation. Indeed, research has extensively

investigated how the extent to which two properties are congruent is a main source to the success

of the brand alliance marketing, including endorsement and sponsorship (Crimmins & Horn,

1996; McDonald, 1991; Olson & Thjomoe, 2011).

However, several studies suggest that highly incongruent pairings between

endorser/sponsor and property can lead to greater cognition, attention, and elaboration as well as

positive endorsement/sponsorship evaluations among consumers. In other words, the greater the

incongruity, the more effort is required from the individual, which results in increased recall and

recognition. Thus, depending on the objectives of brand alliances, even extremely incongruent endorsement/sponsorship would be favorably evaluated. For example, most recently, Chang et al. (2014) suggested that creating cognitive dissonance induced by a misfit situation could also be a desirable strategy in some marketing contexts. Following their arguments, both of the product- and athlete-merited deals (i.e., AMD and PMD) have a greater probability of changing consumers’ preexisting attitudes as well as being evaluated favorably over athlete-product- merited deals (i.e., APMD) because these cases would indicate: (1) that the existing image of the brand is not well-defined or not desirable and needs to be stirred up, and formed through effortful processing; and (2) a shock effect, which is a useful strategy to stand out in consumers’ mind before grabbing their attention for another purpose regarding the brand.

Based on this understanding, in contrast to the traditional accounts of endorsement/sponsorship effectiveness, misfit between endorser/sponsor and the sponsored/endorsed property also may induce favorable consumer evaluations due to biased competition. Specifically, the bottom-up neural mechanisms for object selection in the biased competition model suggest that a unique target in an array of homogeneous non-targets or a sudden appearance of a new object (e.g., larger, brighter, faster) in a visual display attract greater selectivity and cognitive attention (Yantis, 2008). This selective attention and enhanced cognitive efforts then lead to increased likelihood that the object is chosen in a subsequent choice task (Janiszewski et al., 2013). In addition, following the top-down control of selection (i.e., things that are relevant are more likely to be selected to attend), it is predictable that consumers with high athlete and product involvement may also show greater attention. The increased attention and cognitive efforts then may result in a favorable evaluative judgment. According to the availability-valence hypothesis of memory-based accounts for judgmental process (Kisielius

& Sternthal, 1986), this occurs because increased attention to a target enhances the ease with which consumers can process the target in subsequent encounters, and in turn, this fluency leads to more favorable attitudes toward the target. Therefore, the researcher hypothesizes that:

H1-2: Involvement level in athlete endorsers and product brands moderates the relationship between fit and consumers’ evaluative judgment. Specifically, for consumers with low involvement, this relationship is U-shaped, where the highest and lowest levels of fit lead to greater elaboration, as well as more favorable evaluative judgment. On the other hand, high involvement consumers, regardless of fit, will show greater elaboration as well as favorable evaluative judgment.

Dual Process of Evaluative Judgment Construction in Athlete Endorsement

The associative-propositional (APE) model: Understanding attitudes as information

Attitudes have been defined as psychological tendencies to evaluate a given entity with some degree of favor or disfavor (Eagly & Chaiken, 1993; Zanna & Rempel, 1988). When considering attitudes as mental representations in an individual’s memory, attitudes can be characterized as the knowledge structures and associative networks of interconnected evaluations

(Olson & Zanna, 1993). This approach assumes attitudes are one of many retrievable information sources; elicitation of one attitude makes closely related attitudes accessible through the process of spreading activation. Therefore, a change in an individual’s evaluative judgment toward an object can cause the individual to change his or her attitudes toward other objects closely associated with the target object.

One of the most compelling approaches conceiving of attitudes as retrievable information is the associative-propositional evaluation (APE; Gawronski & Bodenhausen, 2006) model. The

APE model goes beyond existing unitary attitudes conception by suggesting that such evaluative tendencies can be rooted from two types of evaluative processes, producing implicit attitudes through associative evaluation and explicit attitudes through propositional reasoning. Rydell and

McConnell (2006) suggest that associative evaluations represent automatic reactions that are activated unconsciously in memory when a consumer encounters a particular stimulus. Thus, such activations often require less effortful mental processes or cognitive capacity. Associative networks of memory are often activated effortlessly by experiencing feature similarity and spatiotemporal proximity between the stimulus and existing memory (Bassili & Brown, 2005).

Therefore, implicit attitudes represent unconscious and automatic levels of responses (De

Houwer, Teige-Mocigemba, Spruyt, & Moors, 2009), and provide relatively consistent predictions across variant consumption environments (Gawronski & Bodenhausen, 2006).

In contrast, propositional reasoning is characterized as syllogistic inferences derived from a reflective evaluation system (Strack & Deutsch, 2004). Hence, the processes of propositional reasoning are generally of concern whether evaluations, beliefs, or attitudes are true or false.

Explicit attitudes tend to be shaped in a manner consistent with fast-changing processes, and are vulnerable to new information. Thus, contemporary psychologists suggest that propositional reasoning can represent a superordinate level compared to an associative evaluation (Bohner &

Dickel, 2011; Gawronski & Bodenhausen, 2006; Strack & Deutsch, 2004).

Formations and changes of implicit and explicit attitudes

Implicit attitudes toward a particular object can be formed or changed through cumulative experiences with the object such as in the cases of evaluative conditioning (Sweldens, Van

Osselaer, & Janiszewski, 2010) or sequential priming (Cameron, Brown-Iannuzzi, & Payne,

2012). For example, repeated pairings of a news article about a doping scandal (i.e., conditioned

stimuli; CS) and a celebrity athlete representing neutral valence (i.e., unconditioned stimuli; US)

can result in negative implicit attitudes toward the athlete. As such, when consumers learn about

a new evaluation toward the US, repetitive exposures of the same or similar patterns of

information about the pairs of CS and US can shape implicit attitudes toward the US. Going

further, even in a situation where consumers hold a positive valence toward the athlete (US), the

preexisting associative evaluation can also be changed by being exposed to repetitive parings of

the athlete with negatively conditioned stimuli.

Another way to form or change implicit attitudes is a temporal change in the activation of

preexisting associative patterns (Hughes, Barnes-Holmes, & Houwer, 2011). In other words,

differential activation of existing associative evaluations can rebuild implicit attitudes. Therefore,

underlying assumptions of this type of associative evaluation constructions are that consumers

already have a certain amount of information about the attitude object, and CS activates different

associative patterns. For example, Tiger Woods may be evaluated favorably when existing

associative networks are activated through positive CS such as athletic expertise or high winning

records. At the same time, sexual scandal related CS might activate his recent poor performance, which would result in negative associative evaluation. As such, associative patterns and evaluations that are activated by a given consumer can differ as a function of the context surrounding the consumer (Bohner & Dickel, 2011).

Explicit attitudes formed through propositional reasoning are often changed by experiencing cognitive dissonance because the processes of cognitive reduction would be largely handled through propositional reasoning (Gawronski & Strack, 2004). People often experience

cognitive dissonance when several propositions are contradicting each other, such as favorable

and unfavorable. With respect to the processes of dissonance reduction, consumers often resolve the dissonance either by explicitly rejecting one of the inconsistent propositions as false or by finding additional propositions that resolve the inconsistency (Hofmann, Gschwendner, Nosek,

& Schmitt, 2005).

The role of fit in the formations and changes of implicit and explicit attitudes

Consumers often encounter two consumption situations in which the information about an attitude object is either “consistent or inconsistent” with their existing memory. The associative-propositional evaluation model (APE) suggests that people translate responses from the associative evaluation into propositional structure (Gawronski & Bodenhausen, 2006). In other words, an associative evaluation about the object is likely to add new propositions to the set of propositions derived from the propositional reasoning processes. For example, consumers often automatically and favorably react to a brand they have previously liked, which is transformed into the proposition of “I like this brand since I have a good experience with this brand.” Therefore, their evaluation if fluently processed in a situation where their existing knowledge structure in memory relevant to the object is “consistent” to the information about the object (Olson & Thjomoe, 2011; Winkielman, Huber, Kavanagh, & Schwarz, 2012). This fluency then leads to more attention, elaboration, and large amount of comprehension efforts since people often seek out evidence or further information to confirm their existing beliefs or expectations given that what is likely to be closed to their existing knowledge structure is likely also to be desirable (Nickerson, 1998). The increased attention, elaboration, and comprehension efforts then lead to more favorable attitudes toward the object because people also are likely to show favorable attitudes toward what is perceptually available (Janiszewski et al., 2013; Lee &

Labroo, 2004). Accordingly, consumers would automatically and initially evaluate the

preexisting knowledge structure consistent information favorably, and then the favorable implicit

attitudes would be further confirmed explicitly through the propositional reasoning processes.

H2-1: Fit between athlete endorsers and endorsed brands positively influences consumers’

processing fluency, which in turn, leads to more attention and elaboration, which in turn, leads to

initially favorable implicit attitudes. Therefore, high fit rather than moderate fit will show more

favorable implicit attitudes.

The initially developed favorable implicit attitudes are likely to be magnified through the

propositional reasoning processes. Therefore:

H2-2: High fit, rather than moderate fit, will increase favorable explicit attitudes to a greater

extent when compared to implicit attitudes.

At the same time, preexisting knowledge structure “inconsistent” information also may

trigger intensive associative processes since this type of information tends to induce such

cognitive biases as bottom-up control of selective attention, which often results in enhanced

involuntary attention and elaboration (Theeuwes, 2010; Yantis, 2008). Through the biased attention and elaboration, the strength of associations relevant to the object in consumers’ memory becomes reinforced. Then, in a subsequent judgment, the reinforced association strength enhances processing fluency. However, in the propositional reasoning stage, in spite of the enhanced processing fluency, consumers may experience cognitive dissonance since the information is still explicitly inconsistent with their existing knowledge structure. Therefore, such probabilities may be still large enough as they are likely to reject the inconsistent propositions derived from the associative processes as false to resolve the cognitive dissonance,

and to confirm their existing evaluative judgments. This process occurs because consumers

generally evaluate the validation of proposition in the propositional processes, so that it becomes

independent of associative evaluations when the propositions derived from automatic reaction is

evaluated as invalid (Gawronski & Bodenhausen, 2006). Thus, multiple attitudes could be stored

separately in a manner consistent to the “past attitudes are still there” model (PSAT; Petty,

2006).

The traditional dual processing accounts (e.g., elaboration likelihood model; Petty,

Cacioppo, & Schumann, 1983 or heuristic systematic model; Eagly & Chaiken, 1984) suggest

that as soon as an attitude is newly constructed through peripheral or deliberate processing, only

the newly shaped attitude can remain tarnishing the original attitude (Dempsey & Mitchell,

2010). However, contemporary attitude models (e.g., the metacognitive model; Petty, Brinol, &

DeMarree, 2007 or the associative-propositional evaluation model; Gawronski & Bodenhausen,

2006) suggest that evaluative judgments of positive and negative valences toward an object may exist together in consumers’ mind, and the strengths of the judgments also can vary depending on their associative network in memory (Petty et al., 2007). This phenomenon occurs because the two types of attitudes, including explicit and implicit judgments, are managed through different underlying mental processes (Bohner & Dickel, 2011).

H3-1: Low fit, rather than moderate fit, leads to more attention and elaboration, which in turn,

leads to enhanced association strength of an attitude object in their memory, which in turn, leads

to processing fluency as well as favorable evaluation in a subsequent evaluation. Therefore, low

fit rather than moderate fit will show more favorable implicit attitudes.

Low fit between athlete endorsers and endorsed brands is likely to increase consumers’ cognitive dissonance, which in turn, leads to unfavorable explicit attitudes. However, the unfavorable explicit attitude formed through the propositional reasoning processes may be

memorized in different storage from the initially developed favorable implicit attitudes.

Accordingly:

H3-2: Low fit, rather than moderate fit, will increase favorable implicit attitudes to a greater

extent when compared to explicit attitudes.

Moderation effects of involvement

It is commonly agreed that consumers who experience greater involvement in an information-processing situation tend to show such cognitive biases as greater attention and

elaboration (i.e., top-down control of selective attention; Theeuwes, 2010; Yantis, 2008) as well

as central-route processing requiring high levels of comprehension efforts (Cornwell, Weeks, &

Roy, 2005). This phenomenon occurs because such consumers as highly involved groups often

have more complex and strengthened network of associations relevant to the object in their

memory. Therefore, observing a specific cue can activate more associations connected to the

object rapidly and broadly. Therefore, it is predicted that consumers with high involvement

process their associative evaluation more fluently regardless of whether the information about an

attitude object is either consistent or inconsistent with their existing memory network, which in turn, leads to favorable implicit attitudes.

In the propositional reasoning processes, consumers with high involvement also may experience cognitive dissonance due to the preexisting knowledge structure “inconsistent” information in a low fit condition. However, they will be relatively insusceptible to the inconsistency presumably because, due to their relatively abundant associative network, they often possess a larger probability to be activated on various associative patterns stored in their memory by observing the same object (Gawronski & Bodenhausen, 2006). This large probability may equate to different types of reactions and evaluations compared to the reactions and

evaluations observed from the consumers with low involvement. For example, the associative pattern activated by the set of stimulus Phil Mickelson and Trojan, a condom brand, may be divided into two ways: (1) negative - scandal and cheating issues in the golf industry; and (2) positive - golf, champion, productivity, family, and credibility. Hence, even though the concept

“Phil Mickelson” could be associated with both positive and negative words in the consumers’ memory, which of the two categories becomes differently activated depends on consumers’ preexisting subjective memory network or the particular context in which the stimulus Phil

Mickelson is encountered. In other words, by observing a novel endorsement, highly involved consumers may relatively have a higher probability of finding an additional proposition that resolves the inconsistency.

H4: Compared to consumers with low involvement, highly involved consumers will show consistently favorable evaluative judgments implicitly and explicitly regardless of fit manipulations.

Hierarchical relationships of evaluation processes

Integrating existing theoretical approaches, including the associative-propositional evaluation model (Gawronski & Bodenhausen, 2006), and the theories of reasoned action and planned behavior (Ajzen, 1991), we developed the associative propositional intention model

(API). The API includes four hierarchically ordered evaluation processes: (1) observation and recognition of a stimulus – stimuli exposure, (2) associative evaluation – implicit attitudes, (3) propositional reasoning – explicit attitudes, and (4) behavioral intention. That is, the API assumes that as soon as consumers are exposed to a particular stimulus, they automatically and initially employ the associative evaluation processes based on their existing memory network relevant to the stimulus. Then, as time goes by, effortful, elaborative, and propositional reasoning

processes are utilized to further evaluate the stimulus, which, in turn, influences various types of

behavioral intention.

H5: Stimuli exposure automatically leads to associative evaluation, which in turn, influences

propositional reasoning, which in turn, leads to behavioral intention.

Summary

Despite the breadth of research in the cognitive information processing and attitude

formation/change, only handful scholars have examined consumers’ attentional/elaborational

process in the constructions of implicit and explicit attitudes in the context of athlete

endorsement. The researcher argues that fit between athlete endorsers and endorsed product

brands significantly influences consumers’ attention, elaboration, and evaluative judgments,

including explicit and implicit attitude, where involvement level moderates this relationship. The

researcher attempts to show contradicting predictions of endorsement/sponsorship effectiveness

by systematically incorporating existing theories from psychology and marketing, including

theories of selective attention and biased competition model. As little research has been done

investigating automatic and unconscious levels of information processing in the contexts of endorsement/sponsorship, the current study may contribute to the theoretical understanding of such phenomenon by empirically examining the concept of implicit evaluative judgment. The results of the current study may provide (sport) brand managers with useful information for evaluating the effectiveness of athlete endorsement, as well as a guideline for diagnosing existing and potential endorsements. For example, managers may have more flexibility/variability in searching for an endorsement partner, since an endorsement contract, conventionally perceived

not well matched, also can create consumer attention as well as a positive endorsed product

brand evaluation.

CHAPTER 3 EXPERIMENT

Experimental Design

This experiment was a 3 (fit: high vs. moderate vs. low; between subjects) × 2 (athlete and product brand involvement level: high vs. low; between subjects) × 2 (replicate target: athlete endorser and product brand; within subjects) design with a between subject manipulation of the 12 hypothetical endorsements. One hundred and eighty undergraduate students were recruited from a large university for the main experiment, and they received course credits for their participation.

Pretests

Three pretests, including two qualitative studies and one field study were conducted to identify potential endorsement deals, which differ in perceived fit between endorsers and brands and involvement levels. The starting point was a set of 357 brands; the researcher initially identified 156 professional athletes and 201 global brands taken from several credible online sources (e.g., Best Global Brands by Interbrand and Powerful Brands by Forbes).

Then, the researcher recruited five faculty members and 15 graduate students who were considered to have a decent level of knowledge and interest in endorsements of professional athletes and global product brands. The researcher first presented the participants with a brief description and key concepts. The participants served as judges; they were asked to examine the entire lists of athlete and product brands (357 brands), and to identify all brands that meet the following key requirements: (1) each brand should be on a maturity stage in term of brand life cycle to possess similar levels of brand familiarity and awareness, (2) each brand should possess specific associations that are highly salient and could differentiate the brand from others, so that the researcher could track where the level of fit would stem from after matching tasks; and (3)

each brand would possess a greater potential to be consumed by university students samples to ensure external validity of the study. Based on the judges’ agreement on brand selection, a new set of athlete and product brands was attained, including 20 athletes and 60 product brands. Next, they were asked to identify possible endorsement deals between athlete and product brands by using the retained brands based on the level of fit (i.e., low, moderate, and high). In this matching task, the researcher requested them to avoid existing matched endorsement contracts to prevent potential confounding effects such as mere exposure. As a result, 36 potential endorsements were identified, 12 deals for each fit condition (see Table 4-3).

For the second pretest, we recruited another 36 undergraduate students of which 18 were male. The participants were asked to respond to the following open-ended question: “Please write descriptions in an open-ended way about why and how much the athlete endorser is fitted to the product brand or vice versa. If they are not fitted, please explain why and how much they are not fitted.” The 36 retained endorsements were randomly assigned to each participant. The researcher recruited another five graduate students serving as a judge to evaluate and diagnose the open-ended responses. They were asked to judge the responses about how much each description is theoretically and conceptually aligned with the concept of fit. The three key criteria used in the first pretest were also utilized again in evaluating the responses (i.e., maturity stage, specific associations, and appropriateness for student consumers). Then, they were asked to classify each response into the three fit conditions. Based on the results, the researcher finally identified 12 potential endorsements (i.e., four deals for each fit condition), dropping 24 endorsements showing mixed results of fit evaluation as well as showing a lack of the three key requirements.

To further examine the experimental validity of the selected stimuli, the researcher

recruited another 72 undergraduate students of which 39 were male. They were randomly assigned to each potential endorsement (six participants for each endorsement). They were asked to evaluate the potential endorsement contract based on brand affect (Broniarczyk & Alba, 1994) and familiarity (Broniarczyk & Alba, 1994) toward an athlete endorser and endorsed brand as well as fit between the two (Speed & Thompson, 2000). The format for the instrument was a seven-point Likert type scale ranged from (1) ‘Strongly Disagree’ to (7) ‘Strongly Agree.’ The researcher conducted one-way ANOVAs to compare the scores on the three conditions using the package car (Fox, 2014) in R 3.1.1 (R Development Core Team, 2014). The single-item measures of athlete affect (F(1, 70) = 1.63, p = .21; Mhigh = 4.62, Mmoderate = 4.87, and Mlow =

5.12), athlete familiarity (F(1, 70) = 2.79, p = .09; Mhigh = 5.38, Mmoderate = 5.78, and Mlow =

6.17), product affect (F(1, 70) = .99, p = .32; Mhigh = 4.47, Mmoderate = 4.22, and Mlow = 3.97), and product familiarity (F(1, 70) = .19, p = .66; Mhigh = 5.81, Mmoderate = 5.92, and Mlow = 6.02) were not significantly different among the three conditions. The five items measuring fit were averaged to form a single scale. Fit was significantly different, F(1, 70) = 73.56, p < .001, among the high (M = 4.7), moderate (M = 3.49), and low conditions (M = 2.29). Therefore, the 12 potential endorsements as depicted in Table 4 identified through a series of pretest were considered to be appropriate for hypothesis testing in the main experiment.

Instrumentation

Explicit Cognitive Attitude toward Athlete and Product

Explicit attitudes often include two separate aspects of cognitive and affective evaluative judgments, while implicit attitudes do not distinguish the dual types but consider attitudes simply as a retrievable and activated set of information. Explicit attitudes can be measured through a self-reported test, given that explicit attitude involves deliberative evaluation (Gawronski &

Bodenhausen, 2006).

Numerous scale items are available when measuring consumers’ cognitive side of

evaluative judgments toward celebrity endorser and endorsed brand. For example, the source

models including credibility and attractiveness (e.g., Ohanian, 1991) have been dominant in

celebrity endorsement studies. Most recently, applying and extending the source models, Chang

et al. (2014) suggested three-dimensional mirror measures of celebrity endorsers and brands. The

traditional source models use multiple scale items, while the recently extended three-dimensional mirror measures employ single-item measures applicable to both endorsers and brands. By

adapting both approaches, in the current study, the researcher developed multiple scale items of

three-dimensional cognitive mirror attitudes applicable to both athlete endorsers and endorsed

brands. To reduce participants’ boredom, frustration, and fatigue associated with answering

highly similar questions repeatedly, the researcher employed bipolar adjective scales.

Particularly, the researcher modified and used existing scale items, including celebrity

dimensions of expertise, trustworthiness, and attractiveness (Chang et al., 2014; Fleck, Forchia,

& Roy, 2012; Kamins & Gupta, 2006; Ohanian, 1991), and product dimensions of quality,

trustworthiness and attractiveness (Chang & Ko, 2014; Kamins, 1990; Kamins & Gupta, 2006;

Zeithaml, 1988).

Explicit Affective Attitude toward Athlete and Product

In the consumer behavior literature, one of the most compelling approaches that account

for the affective aspect of evaluative judgments are the pleasure, arousal, and dominance

paradigm (PAD; Russell & Mehrabian, 1977). Adapting and extending the PAD as well as its

applications in the consumer behavior realm (Kerr, Wilson, Nakamura, & Sudo, 2005; Richins,

1997), the researcher developed three dimensional affective mirror attitudes applicable to both

endorsers and endorsed brands. Similar to the explicit cognitive attitudes measures, the

researcher employed bipolar adjective scales again.

Other Explicit Measures

In addition to the explicit cognitive and affective attitudes, the researcher adapted and

modified several existing scale items to measure dependent variables, including recommendation

intention of endorsed brands through social network sites (SNS) and intention to establish a

friendship with athlete endorsers through SNS (Bhattacherjee, 2001; Jin & Phua, 2014). The

scale items measuring involvement levels with athlete endorsers and endorsed product brands

were adopted from the existing study (Zaichkowsky, 1985).

Procedures of Main Experiment

Fit Manipulation

Upon arrival at the experimental laboratory, the participants were told that they are

participating in a “Display distance and visual fatigue” study. They were randomly assigned to

one of 12 endorsements. They were sequentially exposed to three different images of the

randomly selected sole endorsement contract. The researcher used Qualtrics, a computerized

survey software, for the manipulation of fit. Specifically, the researcher used pictorial stimuli; for athlete endorsers, the upper half of the body was located on the left side of the screen with their name, while for product brands, the brand name/logo was located on the right side of the screen.

Before exposure of the stimuli, participants were asked to read a brief description, which reads as: “Please take your time to imagine this potential endorsement. Please tick to see the endorsement. When you complete this imagination task, please tick to turn to the next page.” At the same time, the researcher recorded the number of seconds the participants spends attending to the stimuli of the three images in milliseconds with the timing feature in Qualtrics; this record measures the time between when the page loads and when the participants’ click on the submit/next button on page. Time spent for the three images, respectably, was combined, which generates one continuous variable. The researcher assumes the time the participants spend

processing endorsement stimuli as one of indicators of level/amount of attention/elaboration

effort.

Single Target-Implicit Association Test (ST-IAT)

As soon as they complete the fit manipulation task, the participants were asked to take the single target-implicit association test (SA-IAT; Bluemke & Friese, 2008) to measure implicit

attitude toward both the endorsed brand and the athlete endorser they just looked at. ST-IAT is

an application of the original implicit association test (IAT) developed by Greenwald, McGhee,

and Schwartz (1998). The IAT is one of the most popular methods for assessing implicit attitude;

it was designed to measure implicit attitudes by assessing their underlying automatic evaluation,

affect or attitude (Greenwald et al., 1998). One fundamental assumption of the IAT is that

reaction time toward a stimulus of strongly associated pairs is faster than that of weakly

associated pairs.

The original IAT includes a series of five discrimination tasks. In the first and second

tasks, for example, a pair of target concepts (e.g., Tiger Woods vs. Tim Tebow) and a pair of

attributes (e.g., morality vs. immorality) can be introduced, respectably. In the third task, the

target concepts and attributes are combined (e.g., Woods-morality and Tebow-immorality), and then recombined in the fifth task after reversing the response assignment (e.g., Woods-

immorality and Tebow-morality). In the two combined tasks, the subject finds Woods and

immorality, and Tebow and morality to be consistently and considerably easier than the other

(i.e., Woods-morality and Tebow-immorality) in the experiment that presents the scenario right

after Tiger Woods’ sex scandal. Therefore, the reaction time difference provides numerical

estimation of implicit attitude toward the two target categories (i.e., the discrepancy between

negative and positive attitudes).

Instead of measuring two targets in the original IAT, some researchers may prefer to

employ ST-IAT in the case of assessing a sole target object without the need to simultaneously

evaluate a counter category. As Bluemke and Friese (2008) suggested, the ST-IAT is especially useful in the case of measuring multiple single-targets, like the three conditions of fit in the current study, because this relatively new measure enables researchers to remove a counter category that would introduce a nuisance present in a traditional IAT. Existing studies also showed that the ST-IAT is internally consistent, as well as it has good test-retest reliability and discriminant validity (Bluemke & Friese, 2008).

In the current study, two separate ST-IATs were employed: one for the endorsed product brand and one for the athlete endorser. The two target concepts are endorsed product brand and athlete endorser, and the attribute categories are favorable and unfavorable attitudes. Stimuli of the two ST-IATs include: (1) six stimuli describing the randomly selected target product brand by using logo, images, and emblems; (2) six stimuli describing the randomly selected target athlete by using the selected athlete’ images; (3) six positive words describing favorable associations; and (4) six negative words describing unfavorable associations.

Two separate ST-IATs include a total of five blocks of trials. In the first block of trials, participants practiced sorting attribute category stimuli (i.e., favorable-positive words or unfavorable-negative words). Attribute category labels (i.e., favorable and unfavorable) were

placed on the top corners of the screen. All stimulus words were presented in the middle of the

screen throughout the five blocks. Participants were asked to categorize the presented stimulus

words by pressing keys (“e” or “i”) corresponding to the category labels. After the practice block

of 24 trials with only positive and negative words, the combined ST-IAT blocks were presented

(2 and 3 blocks). The target category label (e.g., one randomly selected endorsed brand name

he/she just viewed in the Qualtrics task) was placed either below the ‘favorable’ label or below

the ‘unfavorable’ label. Participants were asked to categorize the presented stimulus by pressing keys (“e” or “i”) corresponding to the category labels on screen. In these two blocks, 24 were attribute trials (i.e., two replicates of six positive and six negative words) and 12 were target trials (i.e., four replicates of six words describing the target brand). In one block, target brand words were combined with positive words, while target brand words were combined with negative words in another block. The order of these two blocks was randomly counterbalanced

over participants. A large X was shortly appeared if participants make a mistake.

As described above, each participant was asked to complete two ST-IATs: (1) endorsed

product brand, and positive and negative words (2 & 3 blocks); and (2) athlete endorser, and

positive and negative words (4 & 5 blocks). The order of the two ST-IATs was counterbalanced

over participants, and participants were asked to start with the same block in both ST-IATs; if a

participant started with the “positive”–target “product” brand block, she/he also was asked to

complete the “negative”–target “product” brand block. Table 4-5 displays an overview of the

category assignment and stimulus proportions across two ST-IATs.

As soon as the participants completed the two ST-IATs, they were asked to respond to

explicit measures, including explicit cognitive (i.e., quality, trustworthiness, and attractiveness

for product brand and expertise, trustworthiness, and attractiveness for athlete endorser) and

explicit affective attitudes (i.e., pleasure, arousal, and pride) toward both athlete and product

brand, behavioral intentions (i.e., game watching, game watching recommendation, SNS

friendship establishment for athlete endorser, and recommendation for product brand), and

involvement levels, psychological connections, awareness, and affect toward both athlete and

product brand. Lastly, participants were asked to respond to fit measures between athlete

endorser and endorsed brand. After completing the experiment, the participants were fully debriefed and thanked for their cooperation.

Data Analysis

Data analysis was performed in five stages. First, data from the experiment was carefully screened to adjust missing data, outliers, and leverage points as well as to test normality before entering main analyses. Second, descriptive statistics for the variables were performed to describe the basic characteristics of the data, including gender, age, education, and ethnic background. Third, measurement models were examined to establish reliability and validity of scale items, including explicit cognitive and affective attitudes and involvement levels. Fourth, following existing algorithm of ST-IAT scoring, several steps of data preparation in regard to the obtained latencies were preceded for main analyses. Lastly, the researcher performed between- groups analysis of variance (ANOVA) and moderated mediation path analysis. Data was entered into and analyzed utilizing R statistical packages. Specific descriptions for each step of analyses are documented below.

Data Screening: Missing Data, Outliers, Leverage Points, and Normality

Prior to the main analyses, all the variables was examined employing the psych package

(Revelle, 2014) in R (R Development Core Team, 2014) for accuracy of data entry, outliers, leverage points, and fit between the characteristics of the data. First of all, the dataset was evaluated for accuracy of data entry and missing data. After missing data is addressed, the

Little’s missing completely at random test (MCRA: Little, 1988) was performed to examine whether missing occurs depending on the variables in the dataset. The missing is completely at random when the Little’s MCRA test has significance value greater than .05, the null hypothesis fails to reject. In the case of missing completely at random, the missing value was replaced with a predicted value by the Expectation Maximization (EM) algorithm (Dempster, Laird, & Rubin,

1977), which is one of the most popular methods in the implementation of maximum-likelihood estimation (Cappe & Moulines, 2009).

Each variable was evaluated for univariate outliers using the boxplot outlier labeling rule

(Hoaglin et al., 1986), which is one of the most common graphical and statistical procedures that provides much more accurate evidence than using a visual inspection of checking scatterplot

(Frigge, Hoaglin, & Iglewicz, 1989). Elimination of case or variable, transformation, and score alteration were considered to reduce the influence of outliers and leverage points based on the nature of the outlier and leverage point. Normality of the observed variables was also assessed through examination of histogram and a summary of descriptive statistics.

Descriptive Statistics

Descriptive analyses were performed on participants’ socioeconomic-demographic characteristics (i.e., gender, age, ethnic background, and education) to describe the basic characteristics of the data in this study. Various descriptive statistics of the variables used in this study, such as measures of central tendency (e. g., mean, mode, and median) and measures of variability (e. g., range, variance, and standard deviation), were obtained by using the psych package (Revelle, 2014) in R (R Development Core Team, 2014).

Measurement Model Test

The data was subjected to further scale purification using confirmatory factor analysis

(CFA) on the designed latent constructs, including explicit cognitive and affective attitudes and involvement levels. Psychometric properties, theoretical relevance of the items, and scale parsimony were assessed. The CFA was conducted using the package lavaan (Rosseel, 2014) in

R 2.15.2 (R Development Core Team, 2014). The results of CFA were examined with the overall fit index scores, including comparative fit index (CFI), standard root mean squared residual

(SRMR), and root-mean-square error of approximation (RMSEA). A CFI value close to .95 or

higher and an SRMR value less than .08 imply good-fitting models (Kline, 2010). An RMSEA

value less than .06 suggests a good fit, whereas a value between .06 and .08 indicates an

acceptable fit, and a value higher than .10 implies an unacceptable fit (Hu & Bentler, 1999).

Internal consistency values (Cronbach’s alpha coefficients) were also employed to

examine how well the subscale items are correlated with each other. When the values are greater

than .70, the reliability is indicated as acceptable (Kline, 2010). Additionally, Average Variance

Explained (AVE) values were utilized to evaluate how well the items measuring a specific

subscale collectively explain the underlying construct’s variance. When the AVE value is greater

than .50, the composite reliability of the construct is indicated as acceptable (Kline, 2010). To

establish convergent and discriminant validity, as Kline (2010) suggested, the researcher

examined item loadings and factor correlations. The researcher also asserted established

convergent validity when item loading is equal to or greater than .50 (Hair, Black, Babin, &

Anderson, 2009). Moreover, as Kline (2010) suggested, the researcher also considered discriminant validity is established when correlations among constructs are less than .85 and a squared correlation between constructs are lower than the AVE value for each construct.

Data Preparation of Latencies in the ST-IAT

The researcher followed existing algorithm of ST-IAT scoring suggested by Bluemke and

Friese (2008). The suggested scoring algorithm includes: (1) skipping error trials by recoding

latencies below and above 300 milliseconds, (2) z-transforming participants’ latencies to control latency variability between participants (subtracting an individual’s mean latency from a given reaction time, which is divided by the individuals’ standard deviation), (3) dropping the first trial of each block to control latency variability, and (4) estimating ST-IAT effects by subtracting the mean latency in the product-favorable (athlete-favorable) block from the product-unfavorable

(athlete-unfavorable) block; a positive score was interpreted as a favorable reaction to the

respective target stimuli.

Hypotheses Testing

The researcher employed two-way between-group analysis of variance (ANOVA) to test the main hypotheses. Instead of multiple one-way ANOVAs, the researcher employed a factorial

ANOVA to avoid any increased risk in regard to a Type I error and to examine both main and interaction effects. Since the levels of independent variable (i.e., fit manipulation and involvement) were deliberately selected by the researcher, and remained constant from one replication to another, the research model to be tested through ANOVA in the current study was a fixed-model. Key assumptions underlying the use of ANOVA suggested by Howell (2009) were confirmed through the package car (Fox, 2014) in R 2.15.2 (R Development Core Team,

2014). These assumptions include: (1) independent samples from defined populations, (2) normal distribution, and (3) whether population variances in all cells of the factorial design are equal.

The model equation of the two-way ANOVA hypothesized in the current study is:

where the are the observations, is the baseline, are the main effects for the levels of fit, are the main effects for the levels of involvement, are the interactions for the combinations of fit and involvement, and are the errors that satisfy the conditions of mean equal to 0, equal variances, normality, and independence.

Effect size statistics were computed via the following formula:

Partial = Sum of Squares Factor / (Sum of squares Factor + Sum of Squares Error)

The primary interest of the partial is the extent to which both explicit and implicit attitudes

are associated with the factor. The partial ranges in value from 0 to 1. The partial for

the factor is interpreted as the proportion of variance of the dependent variables related to the

factor, holding constant the covariate. It is unclear what are small, medium, and large values for

partial ; however, following conventional cutoffs (Howell, 2009), it is interpreted as .01

(small), .06 (medium), and .14 (large).

The emerged inquiry about the causal relationships among implicit, explicit (cognitive and affective), and behavioral intention were examined through moderated mediation path analysis using the package lavaan (Rosseel, 2014) in R 3.1.1 (R Development Core Team, 2014).

Before implementing moderated mediation path analysis, the researcher checked several required

assumptions as follows (Edwards & Lambert, 2007): (1) relationships among variables are linear,

(2) interval level data for all variables except for the manipulation variable; it was specified as an independent variable by creating three dummy codes, therefore, the path coefficients from the exogenous manipulation variable to other variables represent the validity of experimental manipulation (Mackenzie, 2001), (3) residual variables are uncorrelated with any of the variables specified in the proposed models, (4) residual errors are uncorrelated with any of the endogenous variables specified in the proposed models, (5) low multicollinearity, (6) there are very few structural equations to resolve for the unknowns – underidentification, (7) all path coefficients flow uni-directionally so that there is no feedback looping – recursivity, and (8) residual errors for the endogenous variables also are uncorrelated confirming the recursivity.

The results of moderated mediation path analysis were examined with the likelihood ratio chi-square test. However, given that the likelihood ratio chi-square test cannot be relied upon alone in the result evaluation process because significance may occur in the test even under the condition in which there are very small differences between the model-implied and observed

covariance matrices. Therefore, the researcher also with the comparative fit index (CFI) and standard root mean squared residual (SRMR) as well as and root-mean-square error of

approximation (RMSEA).

Table 3-1. Examples of Product Brands List used for Pretest 1 Brand name Chase TGI Friday’s TrojanTM Polo Ralph Lauren Colgate IKEA® XBOX JOHNNIE WALKER OfficeMax® Calvin Klein Wendy’s Walmart Mini Jack Daniel’s Abercrombie & Fitch TAGHeuer STARBUCKSTM UGG® Timex® Nestlé PHILIP MORRIS INTERNATIONAL MasterCard® Nescafé® Forever 21 GNC® Taco Bell Dove Yoplait® Heinz GEICO® FedEx® Centrum® Louis Vuitton® Red Bull® Sprite® Kleenex®

Table 3-2. Celebrity Athletes List used for Pretest 1 No. Athlete name No. Athlete name 1 TIGER WOODS 43 ZACK GRIENKE 2 ROGER FEDERER 44 GILBERT ARENAS 3 KOBE BRYANT 45 PAU GASOL 4 LEBRON JAMES 46 DIRK NOWITZKI 5 DREW BREES 47 ADRIAN GONZALEZ 6 AARON RODGERS 48 VERNON WELLS 7 PHIL MICKELSON 49 WAYNE ROONEY 8 DAVID BECKHAM 50 RYAN HOWARD 9 TOM BRADY 51 SERGIO AGUERO 10 DERRICK ROSE 52 DIDIER DROGBA 11 JOE FLACCO 53 JOE JOHNSON 12 KEVIN DURANT 54 JUSTIN VERLANDER 13 ALEX RODRIGUEZ 55 YAYA TOURE 14 FERNANDO ALONSO 56 ROY HALLADAY 15 LIONEL MESSI 57 NEYMAR 16 PEYTON MANNING 58 TIM LINCECUM 17 RORY MCILROY 59 CARL CRAWFORD 18 MARIA SHARAPOVA 60 FERNANDO TORRES 19 DWYANE WADE 61 BARRY ZITO 20 TONY ROMO 62 CHRIS BOSH 21 LEWIS HAMILTON 63 ICHIRO SUZUKI 22 NOVAK DJOKOVIC 64 ERNIE ELS 23 DWIGHT HOWARD 65 KAKA 24 CALVIN JOHNSON 66 DERON WILLIAMS 25 RAY RICE 67 PAUL PIERCE 26 JOE MAUER 68 TONY STEWART 27 JOHAN SANTANA 69 CARL NICKS 28 DEREK JETER 70 JEFF GORDON 29 DWAYNE BOWE 71 LI NA 30 JOSH HAMILTON 72 CARLOS TEVEZ 31 RAFAEL NADAL 73 MATT CAIN 32 USAIN BOLT 74 MATT HOLLIDAY 33 JIMMIE JOHNSON 75 MICHAEL YOUNG 34 PRINCE FIELDER 76 RUDY GAY 35 CC SABATHIA 77 STEVEN GERRARD 36 VINCENT JACKSON 78 ZACH RONDOLPH 37 CLIFF LEE 79 ANDREW BYNUM 38 MARK TEIXEIRA 80 A.J. BURNETT 39 COLE HAMELS 81 BRANDON ROY 40 MATT SCHAUB 82 SAMUEL ETO’O 41 VALENTINO ROSSI 83 LI NA 42 SCHIN TENDULKAR 84 ALEX MORGAN

Table 3-2. Continued No. Athlete name No. Athlete name 85 CHRIS PAUL 121 CRISTIANO RONALDO 86 ALLYSON FELIX 122 FLOYD MAYWEATHER 87 DANICA PATRICK 123 GEORGES ST-PIERRE 88 ANA IVANOVIC 124 MAHENDRA SINGH DHONI 89 VENUS WILLIAMS 125 CARMELO ANTHONY 90 LINDSEY VONN 126 AMAR’E STOUDEMIRE 91 LERYN FRANCO 127 DALE EARNHARDT, JR. 92 HOPE SOLO 128 DEMETRIOUS JOHNSON 93 LOLO JONES 129 WLADIMIR KLITSCHKO 94 CANDACE PARKER 130 ZLATAN IBRAHIMOVIC 95 MICHAEL JORDAN 131 RICKY STENHOUSE, JR. 96 MIKE TYSON 132 VICTORIA AZARENKA 97 GABBY DOUGLAS 133 BRANDT SNEDEKER 98 MISSY FRANKLIN 134 RESSELL WESTBROOK 99 BRITTNEY GRINER 135 ALFONSO SORIANO 100 MICHAEL PHELPS 136 SEBASTIAN VETTEL 101 TIM TEBOW 137 ANNA KOURNIKOVA 102 KEENAN ALLEN 138 VICTORIA AZARENKA 103 EDDIE LACY 139 AGNIESZKA RADWANSKA 104 GIOVANI BERNARD 140 CAROLINE WOZNIZCKI 105 MIKE GLENNON 141 JESSICA ENNIS-HILL 106 ZAC STACY 142 MARIA SHARAPOVA 107 TYRANN MATHIEU 143 MICHAEL CARTER-WILLIAMS 108 KIKO ALONSO 144 CHAQUILLE O’NEAL 109 WILL MYERS 145 DOMINICK CRUZ 110 JOSE FERNANDEZ 146 JOSE ALDO 111 VICTOR OLADIPO 147 ANTHONY PETTIS 112 JORHAN SPIETH 148 CHRIS WELDMAN 113 RUSSELL HENLEY 149 JON JONES 114 CAROLINE MASSON 150 CAIN VELASQUEZ 115 SARA ERRANI 151 RONDA ROUSEY 116 JON JONES 152 CRO COP 117 RENAN BARAO 153 MANNY PACQUIAO 118 DOS SANTOS 154 FELIX HERNANDEX 119 CAT ZINGANO 155 BILLIE JEAN KING 120 YUNA KIM 156 CLAY MATTHEWS

Table 3-3. Preliminarily Selected Athlete and Product Brands and Their Matches Potential endorsements Athletes Product brands Name & B.S.A Fit Name Category B.S.A category Phil Expertise, H Chase Financial service Trust, sincere Mickelson sincere, M TGI Friday’s Food service Family (Golf) family L TrojanTM Personal care/sex Sexual Sexually H Polo Ralph Lauren Apparel/accessory Luxury Rafael Nadal attractive, M Colgate Personal care Cure, clean (Tennis) young L IKEA® Home furnishing Cheap, popular Serena H XBOX Video game Tech, fun Muscular, Williams M JOHNNIE WALKER Alcohol beverage Lavish expertise (Tennis) L OfficeMax® Office Calm Danica Unique, H Calvin Klein Apparel/accessory Tough, sexy Patrick innovative, M Wendy’s Food service Cheap, popular (Motor) attractive L Walmart Discount store The cheapest H Mini Automobile Small, fast Lionel Messi Smart, fast M Jack Daniel’s Alcohol Lavish (Soccer) L Abercrombie & Fitch Apparel/accessory Thin, tall, sexy Michael Record H TAGHeuer Technology Luxury watch Phelps breaker, M STARBUCKSTM Coffee Popular (Swimming) achievement L UGG® Shoes Warm, fur ® Usain Bolt Humble H Timex Electronics Cheap, time (Track& beginning, M Nestlé Coffee Fragrance PHILIP MORRIS Field) confidence L INTERNATIONAL Tobacco Unhealthy H MasterCard® Financial service Trust, sincere Derek Jeter Hero, M Nescafé® Food Coffee (Baseball) Conservative L Forever 21 Apparel/accessory Cheap, young LeBron H GNC® Health/nutrition Exciting, health Exciting, James M Taco Bell Food service Popular, cheap tough (Basketball) L Dove Personal care Soft Gabby H Yoplait® Food Tender Rising star, Douglas Heinz young, thin M Food Ketchup (Gymnastics) L GEICO® Insurance Safe, secure Peyton Average H FedEx® Delivery service Quick, accurate Manning person, M Centrum® Personal care Care (Football) honest L Louis Vuitton® Apparel/accessory Luxury Mirko H Red Bull® Beverage Rugged “Cro Cop” Rugged M Sprite® Beverage Sparkling (MMA) L Kleenex® Personal care Clean Note. B.S.A = brand specific association; Fit; H = high; M = Moderate; L = low

Table 3-4. Selected Athlete and Product Brands and Their Matches for Main Experiment Potential endorsements

Athletes Product brands

Name & B.S.A Fit Name Category B.S.A category

Phil Expertise, H Chase Financial service Trust, sincere

Mickelson sincere, M TGI Friday’s Food service Family

(Golf) family L TrojanTM Personal care/sex Sexual

Sexually H Polo Ralph Lauren Apparel/accessory Luxury Rafael Nadal attractive, M Colgate Personal care Cure, clean (Tennis) young L IKEA® Home furnishing Cheap, popular

Michael Record H TAGHeuer Technology Luxury watch

Phelps breaker, M STARBUCKSTM Coffee Popular

(Swimming) achievement L UGG® Shoes Warm, fur

LeBron H GNC® Health/nutrition Exciting, health Exciting, James M Taco Bell Food service Popular, cheap tough (Basketball) L Dove Personal care Soft

Note. B.S.A = brand specific association; Fit; H = high; M = Moderate; L = low

Table 3-5. Category Assignment and Stimulus Proportions across Five ST-IAT Blocks in Main Experiment Number of stimuli Left key concepts Right key concepts Block Task description Target Target “e” “i” Positive Negative product Athlete

1 Evaluative training Positive Negative 6×2 6×2 - -

2 Combined-task Positive–Target product brand Negative 6×2 6×2 6×2 -

3 Combined-task Positive Negative–Target product brand 6×2 6×2 6×2 -

4 Combined-task Positive–Target athlete Negative 6×2 6×2 - 6×2

5 Combined-task Positive Negative–Target athlete 6×2 6×2 - 6×2

Note. Order of the second-third and fourth-fifth blocks as well as the combinations of the two blocks (2&3 and 4&5) was randomly counterbalanced over participants. Six positive words stimuli include good, pleasant, nice, great, enjoyable, awesome; Six negative words stimuli include bad, awful, terrible, failure, poor, crappy; Six stimuli for each target were images; all of the images was 160 ×

220 pixels.

CHAPTER 4 RESULTS

Demographics

In terms of sample characteristics, 50.6% of the participants were male (n = 91), and the average age of the participants was 21 years old (M = 21.2, SD = 3.1). The majority of the participants were undergraduate senior or junior 62.8% (n = 113), followed by sophomore (n =

31, 17.2%), freshman (n = 18, 10%), and graduate (n = 18, 10%). The majority of the participants were Caucasian (n = 116, 64.4%), followed by Hispanic (n = 31, 17.2%), Asian (n =

16, 8.9%), African American (n = 14, 7.8%), and others (n = 3, 1.7%).

Measurement Model Test

A series of confirmatory factor analyses was conducted to test the four proposed measurement models, including the first-order measurement models of explicit cognitive and affective attitudes toward both athlete and product brand.

Cognitive explicit attitude toward athlete (expertise, trustworthiness, attractiveness)

The measurement model showed an acceptable fit to the data (χ2 = 51.47, df = 24,

RMSEA = .08, CFI = .97, SRMR = .06, p = .001). Factor loadings ranged from .63

(Attractiveness 3) to .95 (Attractiveness 2). Acceptable CFA model fit and high factor loadings functioned as empirical evidence of convergent validity of the measures. All average variance extracted (AVE) scores were acceptable in consideration of the .50 standard (Hair, Black, Babin,

& Anderson, 2009). Means, standard errors, factor loadings, and factor correlations are displayed in Table 6 and 11.

Affective explicit attitude toward athlete (pleasure, arousal, pride)

The measurement model showed an acceptable fit to the data (χ2 = 116.26, df = 51,

RMSEA = .08, CFI = .96, SRMR = .05, p < .001). Factor loadings ranged from .57 (Arousal 4)

to .93 (Arousal 2). Acceptable CFA model fit and high factor loadings functioned as empirical

evidence of convergent validity of the measures. All AVE scores were acceptable in

consideration of the .50 standard (Hair et al., 2009). Means, standard errors, factor loadings, and factor correlations are displayed in Table 7 and 12.

Cognitive explicit attitude toward product (quality, trustworthiness, attractiveness)

The measurement model showed an acceptable fit to the data (χ2 = 42.98, df = 23,

RMSEA = .07, CFI = .98, SRMR = .04, p = .007). Factor loadings ranged from .55

(Attractiveness 3) to .96 (Attractiveness 2). Acceptable CFA model fit and high factor loadings functioned as empirical evidence of convergent validity of the measures. All AVE scores were

acceptable in consideration of the .50 standard (Hair et al., 2009). Means, standard errors, factor

loadings, and factor correlations are displayed in Table 8 and 13.

Affective explicit attitude toward product (pleasure, arousal, pride)

The measurement model showed an acceptable fit to the data (χ2 = 106.32, df = 47,

RMSEA = .08, CFI = .97, SRMR = .05, p < .001). Factor loadings ranged from .66 (Arousal 4)

to .91 (Pleasure 1). Acceptable CFA model fit and high factor loadings functioned as empirical

evidence of convergent validity of the measures. All AVE scores were acceptable in

consideration of the .50 standard (Hair et al., 2009). Means, standard errors, factor loadings, and factor correlations are displayed in Table 9 and 14.

A Summary of Measurement Model Tests

The results showed acceptable fits of the models to the data: (1) cognitive attitudes toward athlete (χ2 = 51.47, df = 24, RMSEA = .08, CFI = .97, SRMR = .06, p = .001), (2)

affective attitudes toward athlete (χ2 = 116.26, df = 51, RMSEA = .08, CFI = .96, SRMR = .05, p

< .001), (3) cognitive attitudes toward product (χ2 = 42.98, df = 23, RMSEA = .07, CFI = .98,

SRMR = .04, p = .007), (4) affective attitudes toward product (χ2 = 106.32, df = 47, RMSEA =

.08, CFI = .97, SRMR = .05, p < .001). In all, acceptable CFA model fits and high factor

loadings functioned as empirical evidence of the convergent validity of the measures. All AVE

scores were greater than the .50 standard (Hair et al., 2009). Means, standard errors, and factor

loadings for each measurement model are summarized in Table 10.

Moderation Effects of Athlete Involvement on Implicit Attitude toward Athlete

Responses were classified into a low (n = 88, the total scores ranged from 1 to 3.9) and a

high group (n = 92, the total scores ranged from 4 to 7) on the basis of median split of the total

score of the eight items measuring athlete involvement. The results of a ANOVA yielded a

significant fit × involvement level interaction for implicit attitude, F(2, 174) = 5.03, p = .007. In the low involvement group, implicit attitude was marginally significantly different, F(2, 85) =

2.72, p = .07, among the high (M = .09), moderate (M = 0), and low fit conditions (M = -.19).

Similarly, for consumers with high athlete involvement, implicit attitude was also marginally

significantly different, F(2, 89) = 2.45, p = .09, among the high (M = .06), moderate (M = .21),

and low fit conditions (M = .33).

An ANOVA was employed again to test the main effects of athlete involvement and fit

manipulation on the implicit attitude. There was a significant main effect of athlete involvement

level (F(1, 174) = 11.13, p = .001) and a non-significant main effect of fit manipulation (F(2,

174) = .084, p = .43) for implicit attitude. Consumers with high level of involvement showed

more favorable attitude (M = .19) than low level of involvement (M = -.04).

Moderation Effects of Athlete Involvement on Cognitive Explicit Attitude toward Athlete

Responses were classified into a low (n = 88, the total scores ranged from 1 to 3.9) and a

high group (n = 92, the total scores ranged from 4 to 7) on the basis of median split of the total

score of the eight items measuring athlete involvement. The results of a ANOVA yielded a non

significant fit × involvement level interaction for cognitive explicit attitude, F(2, 174) = 1.98, p =

.14. In the low involvement group, cognitive explicit attitude toward athlete was not significantly different, F(2, 85) = .94, p = .39, among the high (M = -.29), moderate (M = -.53), and low fit conditions (M = -.58). Similarly, for consumers with high athlete involvement, cognitive explicit attitude toward athlete was also not significantly different, F(2, 89) = 1.26, p = .29, among the

high (M = .22), moderate (M = .56), and low fit conditions (M = .49).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the cognitive explicit attitude. There was a significant main effect of athlete involvement level (F(1, 174) = 41.77, p < .001) and a non-significant main effect of fit manipulation (F(2, 174) = 2.14, p = .12) for cognitive explicit attitude. Consumers with high level of involvement showed more favorable attitude (M = .41) than low level of involvement (M

= -.43).

Moderation Effects of Athlete Involvement on Athlete Expertise

Responses were classified into a low (n = 88, the total scores ranged from 1 to 3.9) and a high group (n = 92, the total scores ranged from 4 to 7) on the basis of median split of the total score of the eight items measuring athlete involvement. The results of a ANOVA yielded a non significant fit × involvement level interaction for athlete expertise, F(2, 174) = .45, p = .64. In the low involvement group, athlete expertise was not significantly different, F(2, 85) = .06, p =

.94, among the high (M = -.35), moderate (M = -.36), and low fit conditions (M = -.43).

Similarly, for consumers with high athlete involvement, athlete expertise was also not

significantly different, F(2, 89) = .55, p = .58, among the high (M = .25), moderate (M = .37),

and low fit conditions (M = .49).

An ANOVA was employed again to test the main effects of athlete involvement and fit

manipulation on the athlete expertise. There was a significant main effect of athlete involvement

level (F(1, 174) = 28.25, p < .001) and a non-significant main effect of fit manipulation (F(2,

174) = .79, p = .46) for athlete expertise. Consumers with high level of involvement showed

more favorable athlete expertise evaluation (M = .35) than low level of involvement (M = -.37).

Moderation Effects of Athlete Involvement on Athlete Trustworthiness

Responses were classified into a low (n = 88, the total scores ranged from 1 to 3.9) and a

high group (n = 92, the total scores ranged from 4 to 7) on the basis of median split of the total

score of the eight items measuring athlete involvement. The results of a ANOVA yielded a non

significant fit × involvement level interaction for athlete expertise, F(2, 174) = 1.5, p = .23. In

the low involvement group, athlete trustworthiness was not significantly different, F(2, 85) =

2.24, p = .11, among the high (M = -.16), moderate (M = -.19), and low fit conditions (M = -.58).

Similarly, for consumers with high athlete involvement, athlete trustworthiness was also not significantly different, F(2, 89) = .69, p = .51, among the high (M = .17), moderate (M = .46), and low fit conditions (M = .33).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the athlete trustworthiness. There was a significant main effects of athlete involvement level (F(1, 174) = 20.08, p < .001) and fit manipulation (F(2, 174) = 3.22, p = .04) for athlete trustworthiness. Consumers with high level of involvement showed more favorable attitude (M = .29) than low level of involvement (M = -.31). Consumers evaluated more favorably toward athlete trustworthiness on moderate condition (M = .25), followed by high (M

= -.01) and low fit conditions (M = -.19).

Moderation Effects of Athlete Involvement on Athlete Attractiveness

In the low involvement group, athlete attractiveness was not significantly different, F(2, 85) =

1.51, p = .23, among the high (M = -.21), moderate (M = -.72), and low fit conditions (M = -.41).

Similarly, for consumers with high athlete involvement, athlete attractiveness was also not significantly different, F(2, 89) = 1.76, p = .18, among the high (M = .13), moderate (M = .54), and low fit conditions (M = .41).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the athlete attractiveness. There was a significant main effect of athlete involvement level (F(1, 174) = 30.59, p < .001) for athlete trustworthiness while, there was a non significant main effect of fit manipulation (F(2, 174) = .79, p = .45) for athlete trustworthiness.

Consumers with high level of involvement showed more favorable attractiveness evaluation (M

= .36) than low level of involvement (M = -.38).

Moderation Effects of Athlete Involvement on Explicit Affective Attitude toward Athlete

173) = 1.08, p = .34. In the low involvement group, explicit affective attitude toward athlete was not significantly different, F(2, 84) = .47, p = .63, among the high (M = -.35), moderate (M = -

.53), and low fit conditions (M = -.53). Similarly, for consumers with high athlete involvement, explicit affective attitude toward athlete was also not significantly different, F(2,89) = .79, p =

.46, among the high (M = .26), moderate (M = .57), and low fit conditions (M = .45).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the affective explicit attitude toward athlete. There was a significant main effect of athlete involvement level (F(1, 173) = 40.95, p < .001) for affective explicit attitude toward

athlete while, there was a marginally significant main effect of fit manipulation (F(2, 173) =

2.38, p = .09) for affective explicit attitude toward athlete. Consumers with high level of

involvement showed more favorable affective explicit attitude toward athlete (M = .41) than low level of involvement (M = -.43). Consumers showed more favorable affective explicit attitude toward athlete on moderate condition (M = .23), followed by high (M = -.07) and low fit conditions (M = -.11).

Moderation Effects of Athlete Involvement on Athlete Pleasure

Responses were classified into a low (n = 88, the total scores ranged from 1 to 3.9) and a high group (n = 92, the total scores ranged from 4 to 7) on the basis of median split of the total score of the eight items measuring athlete involvement. The results of a ANOVA yielded a non significant fit × involvement level interaction for athlete pleasure, F(2, 174) = 1.35, p = .26. In the low involvement group, athlete pleasure was not significantly different, F(2, 85) = .49, p =

.61, among the high (M = -.36), moderate (M = -.51), and low fit conditions (M = -.56).

Similarly, for consumers with high athlete involvement, athlete pleasure was also not

significantly different, F(2, 89) = 1.25, p = .29, among the high (M = .24), moderate (M = .61),

and low fit conditions (M = .46).

An ANOVA was employed again to test the main effects of athlete involvement and fit

manipulation on the athlete pleasure. There was a significant main effect of athlete involvement

level (F(1, 174) = 44.1, p < .001) for athlete pleasure while, there was a marginally significant

main effect of fit manipulation (F(2, 174) = 3.04, p = .05) for athlete pleasure. Consumers with

high level of involvement showed more favorable athlete pleasure evaluation (M = .005) than

low level of involvement (M = -.006). Consumers showed more favorable athlete pleasure on

low fit condition (M = .14), followed by high (M = -.03) and moderate fit conditions (M = -.14).

Moderation Effects of Athlete Involvement on Athlete Arousal

Responses were classified into a low (n = 88, the total scores ranged from 1 to 3.9) and a

high group (n = 92, the total scores ranged from 4 to 7) on the basis of median split of the total

score of the eight items measuring athlete involvement. The results of a ANOVA yielded a non

significant fit × involvement level interaction for athlete arousal, F(2, 173) = .36, p = .69. In the

low involvement group, athlete arousal was not significantly different, F(2, 84) = .33, p = .72,

among the high (M = -.24), moderate (M = -.37), and low fit conditions (M = -.42). Similarly, for consumers with high athlete involvement, athlete arousal was also not significantly different,

F(2, 89) = .14, p = .87, among the high (M = .25), moderate (M = .38), and low fit conditions (M

= .33).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the athlete arousal. There was a significant main effect of athlete involvement level (F(1, 173) = 20.64, p < .001) for athlete arousal while, there was a non significant main effect of fit manipulation (F(2, 173) = .99, p = .37) for athlete arousal. Consumers with high level of involvement showed more favorable athlete arousal evaluation (M = .31) than low level of involvement (M = -.32).

Moderation Effects of Athlete Involvement on Athlete Pride

Responses were classified into a low (n = 88, the total scores ranged from 1 to 3.9) and a high group (n = 92, the total scores ranged from 4 to 7) on the basis of median split of the total score of the eight items measuring athlete involvement. The results of a ANOVA yielded a non significant fit × involvement level interaction for athlete pride, F(2, 174) = .87, p = .42. In the low involvement group, athlete pride was not significantly different, F(2, 85) = .19, p = .82, among the high (M = -.32), moderate (M = -.49), and low fit conditions (M = -.41). Similarly, for consumers with high athlete involvement, athlete pride was also not significantly different, F(2,

89) = .82, p = .45, among the high (M = .19), moderate (M = .51), and low fit conditions (M =

.38).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the athlete pride. There was a significant main effect of athlete involvement level (F(1, 174) = 27.64, p < .001) for athlete pride, while there was a non significant main effect of fit manipulation (F(2, 174) = 1.55, p = .21) for athlete pride. Consumers with high level of involvement showed more favorable athlete pride evaluation (M = .35) than low level of involvement (M = -.36).

A Summary of the Results of Moderation Effects of Athlete Involvement

Responses were classified into a low (n = 88, the total scores ranged from 1 to 3.99) and a high group (n = 92, the total scores ranged from 4 to 7) on the basis of median split of the total score of the eight items measuring athlete involvement. The ANOVA results yielded a significant fit × involvement interaction for implicit attitudes, F(2, 174) = 5.03, p = .007. There were significant main effects of athlete involvement (F(1, 174) = 11.13, p = .001) and a non- significant main effect of fit manipulation (F(2, 174) = .084, p = .43) for implicit attitudes.

In terms of explicit cognitive evaluations, the ANOVA results yielded: (1) a non- significant fit × involvement interaction for athlete expertise, F(2, 174) = .45, p = .64; (2) a non- significant interaction for athlete trustworthiness, F(2, 174) = 1.5, p = .23; and (3) a significant interaction for athlete attractiveness, F(2, 174) = 3.24, p = .04. In regard to explicit affective evaluations, the ANOVA results yielded: (i) a non-significant fit × involvement interaction for athlete pleasure, F(2, 174) = 1.35, p = .26; (ii) a non-significant interaction for athlete arousal,

F(2, 173) = .36, p = .69;and (iii) a non-significant interaction for athlete pride, F(2, 174) = .87, p

= .42. Most of the main effects of both athlete involvement and fit manipulation were significant.

For example, there was a significant main effect of athlete involvement (F(1, 174) = 44.1, p <

.001) as well as fit manipulation (F(2, 174) = 3.04, p = .05) for athlete pleasure.

Moderation Effects of Product Involvement on Implicit Attitude toward Product

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.1) and a high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total score of the eight items measuring product involvement. The results of a ANOVA yielded a marginally significant fit × involvement level interaction for implicit attitude toward product,

F(2, 174) = 2.65, p = .07. In the low product involvement group, implicit attitude was not

significantly different, F(2, 85) = 2.28, p = .11, among the high (M = .09), moderate (M = -.11),

and low fit conditions (M = -.11). Similarly, for consumers with high product involvement,

implicit attitude was not significantly different, F(2, 89) = .53, p = .59, among the high (M =

.13), moderate (M = .18), and low fit conditions (M = .21).

An ANOVA was employed again to test the main effects of athlete involvement and fit

manipulation on the implicit attitude toward product. There was a significant main effect of

athlete involvement level (F(1, 174) = 14.1, p < .001) and a non-significant main effect of fit

manipulation (F(2, 174) = .61, p = .55) for implicit attitude toward product. Consumers with

high level of involvement showed more favorable implicit attitude (M = .17) than low level of

involvement (M = -.04).

Moderation Effects of Product Involvement on Explicit Cognitive Attitude toward Product

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.1) and a

high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total

score of the eight items measuring product involvement. The results of a ANOVA yielded a

significant fit × involvement level interaction for cognitive explicit attitude toward product, F(2,

174) = 3.5, p = .03. In the low product involvement group, explicit cognitive attitude toward

product was significantly different, F(2, 85) = 5.06, p = .008, among the high (M = -.37),

moderate (M = -1.00), and low fit conditions (M = -.63). However, for consumers with high product involvement, explicit cognitive attitude toward product was not significantly different,

F(2, 89) = .17, p = .84, among the high (M = .57), moderate (M = .65), and low fit conditions (M

= .66).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the cognitive explicit attitude toward product. There was a significant main effect of athlete involvement level (F(1, 174) = 132.93, p < .001) and a marginally significant main effect of fit manipulation (F(2, 174) = 2.77, p = .06) for cognitive explicit attitude toward

product. Consumers with high level of involvement showed more favorable cognitive explicit

attitude (M = .17) than low level of involvement (M = -.03).

Moderation Effects of Product Involvement on Product Quality

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.1) and a high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total score of the eight items measuring product involvement. The results of a ANOVA yielded a significant fit × involvement level interaction for product quality evaluation, F(2, 174) = 3.47, p

= .03. In the low product involvement group, product quality was significantly different, F(2, 85)

= 4.92, p = .009, among the high (M = -.35), moderate (M = -1.00), and low fit conditions (M = -

.56). However, for consumers with high product involvement, product quality was not significantly different, F(2, 89) = .14, p = .87, among the high (M = .55), moderate (M = .63), and low fit conditions (M = .63).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the product quality evaluation. There was a significant main effect of athlete involvement level (F(1, 174) = 113.62, p < .001) and a marginally significant main effect of fit manipulation (F(2, 174) = 2.95, p = .05) for product quality evaluation. Consumers with high

level of involvement showed more favorable product quality evaluation (M = .59) than low level of involvement (M = -.63). Consumers showed more favorable product quality evaluation on low fit condition (M = .09), followed by high (M = .08) and moderate fit conditions (M = -.22).

Moderation Effects of Product Involvement on Product Trustworthiness

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.1) and a high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total score of the eight items measuring product involvement. The results of a ANOVA yielded a non significant fit × involvement level interaction for product trustworthiness evaluation, F(2, 174) =

1.57, p = .21. In the low product involvement group, product trustworthiness was not significantly different, F(2, 85) = 2.15, p = .12, among the high (M = -.45), moderate (M = -.87), and low fit conditions (M = -.45). Similarly, for consumers with high product involvement, product trustworthiness was not significantly different, F(2, 89) = .15, p = .86, among the high

(M = .49), moderate (M = .59), and low fit conditions (M = .58).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the product trustworthiness evaluation. There was a significant main effect of athlete involvement level (F(1, 174) = 85.66, p < .001) and a non significant main effect of fit manipulation (F(2, 174) = 1.71, p = .19) for product trustworthiness evaluation. Consumers with high level of involvement showed more favorable product trustworthiness evaluation (M = .55) than low level of involvement (M = -.58). Consumers showed more favorable product trustworthiness evaluation on low fit condition (M = .12), followed by high (M = .002) and moderate fit conditions (M = -.15).

Moderation Effects of Product Involvement on Product Attractiveness

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.1) and a high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total

score of the eight items measuring product involvement. The results of a ANOVA yielded a

significant fit × involvement level interaction for product attractiveness evaluation, F(2, 174) =

3.29, p = .03. In the low product involvement group, product attractiveness was significantly

different, F(2, 85) = 4.75, p = .01, among the high (M = -.21), moderate (M = -.89), and low fit

conditions (M = -.69). Similarly, for consumers with high product involvement, product

attractiveness was not significantly different, F(2, 89) = .12, p = .89, among the high (M = .50), moderate (M = .56), and low fit conditions (M = .59).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the product attractiveness evaluation. There was a significant main effect of athlete involvement level (F(1, 174) = 88.49, p < .001) and a non significant main effect of fit manipulation (F(2, 174) = 2.12, p = .12) for product attractiveness evaluation. Consumers with high level of involvement showed more favorable product attractiveness evaluation (M = .55)

than low level of involvement (M = -.57).

Moderation Effects of Product Involvement on Explicit Affective Attitude toward Product

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.1) and a

high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total

score of the eight items measuring product involvement. The results of a ANOVA yielded a

marginally significant fit × involvement level interaction for affective explicit attitude toward

product, F(2, 174) = 2.44, p = .09. In the low product involvement group, explicit affective

attitude toward product was marginally significantly different, F(2, 85) = 2.88, p = .06, among

the high (M = -.004), moderate (M = -.43), and low fit conditions (M = .19). However, for

consumers with high product involvement, explicit affective attitude toward product was not

significantly different, F(2, 89) = .44, p = .64, among the high (M = .06), moderate (M = -.13),

and low fit conditions (M = .10).

An ANOVA was employed again to test the main effects of athlete involvement and fit

manipulation on the affective explicit attitude toward product. There was a significant main

effects of athlete involvement level (F(1, 174) = 94.27, p < .001) as well as fit manipulation

(F(2, 174) = 3.48, p = .03) for affective explicit attitude toward product. Consumers with high

level of involvement showed more favorable affective explicit attitude toward product (M = .56)

than low level of involvement (M = -.59). Consumers showed more favorable affective explicit

attitude toward product on low fit condition (M = .16), followed by high (M = .03) and moderate fit conditions (M = -.22).

Moderation Effects of Product Involvement on Product Pleasure

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.1) and a

high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total

score of the eight items measuring product involvement. The results of a ANOVA yielded a

significant fit × involvement level interaction for product pleasure evaluation, F(2, 174) = 4.49, p

= .01. In the low product involvement group, product pleasure was significantly different, F(2,

85) = 4.18, p = .01, among the high (M = -.43), moderate (M = -.98), and low fit conditions (M =

-.55). However, for consumers with high product involvement, product pleasure was not significantly different, F(2, 89) = 1.48, p = .23, among the high (M = .41), moderate (M = .72), and low fit conditions (M = .69).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the product pleasure evaluation. There was a significant main effects of athlete involvement level (F(1, 174) = 118.41, p < .001), while there was a non significant main effect of fit manipulation (F(2, 174) = 2.07, p = .13) for product pleasure evaluation. Consumers with high level of involvement showed more favorable product pleasure evaluation (M = .61) than low level of involvement (M = -.63).

Moderation Effects of Product Involvement on Product Arousal

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.1) and a high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total score of the eight items measuring product involvement. The results of a ANOVA yielded a non significant fit × involvement level interaction for product arousal evaluation, F(2, 174) = 1.03, p

= .36. In the low product involvement group, product pleasure was marginally significantly different, F(2, 85) = 2.72, p = .07, among the high (M = -.38), moderate (M = -.75), and low fit conditions (M = -.30). However, for consumers with high product involvement, product pleasure was not significantly different, F(2, 89) = .63, p = .54, among the high (M = .31), moderate (M =

.42), and low fit conditions (M = .58).

An ANOVA was employed again to test the main effects of athlete involvement and fit manipulation on the product arousal evaluation. There was a significant main effects of athlete involvement level (F(1, 174) = 46.47, p < .001), while there was a marginally significant main effect of fit manipulation (F(2, 174) = 2.6, p = .07) for product arousal evaluation. Consumers with high level of involvement showed more favorable product arousal evaluation (M = .44) than low level of involvement (M = -.46). Consumers showed more favorable product arousal evaluation on low fit condition (M = .18), followed by high (M = -.04) and moderate fit conditions (M = -.18).

Moderation Effects of Product Involvement on Product Pride

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.1) and a high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total score of the eight items measuring product involvement. The results of a ANOVA yielded a non significant fit × involvement level interaction for product pride evaluation, F(2, 174) = 1.02, p =

.36. In the low product involvement group, product pride was not significantly different, F(2, 85)

= 1.28, p = .28, among the high (M = -.29), moderate (M = .07), and low fit conditions (M = -

.22). Similarly, for consumers with high product involvement, product pride was not

significantly different, F(2, 89) = .44, p = .65, among the high (M = .13), moderate (M = .31),

and low fit conditions (M = .06).

An ANOVA was employed again to test the main effects of athlete involvement and fit

manipulation on the product pride evaluation. There was a significant main effects of athlete

involvement level (F(1, 174) = 59.38, p < .001) as well as fit manipulation (F(2, 174) = 4.78, p =

.009) for product pride evaluation. Consumers with high level of involvement showed more

favorable product pride evaluation (M = .15) than low level of involvement (M = -.16).

Consumers showed more favorable product pride evaluation on moderate fit condition (M = .19),

followed by low (M = -.07) and high fit conditions (M = -.09).

A Summary of the Results of Moderation Effects of Product Involvement

Responses were classified into a low (n = 88, the total scores ranged from 1 to 4.19) and a

high group (n = 92, the total scores ranged from 4.2 to 7) on the basis of median split of the total

score of the eight items measuring product involvement. The ANOVA results yielded a

marginally significant fit × involvement interaction for implicit attitudes toward product, F(2,

174) = 2.65, p = .07. There were significant main effects of athlete involvement (F(1, 174) =

14.1, p < .001) and a non-significant main effect of fit manipulation (F(2, 174) = .61, p = .55) for implicit attitudes toward a product.

In regard to explicit cognitive evaluations, the ANOVA results yielded: (1) a significant

fit × involvement interaction for product quality, F(2, 174) = 3.47, p = .03, (2) a non-significant interaction for product trustworthiness, F(2, 174) = 1.57, p = .21; (3) a significant interaction for product attractiveness, F(2, 174) = 3.29, p = .03. In terms of explicit affective evaluations, the

ANOVA results yielded: (i) a significant fit × involvement interaction for product pleasure, F(2,

174) = 4.49, p = .01; (ii) a non-significant interaction for product arousal, F(2, 174) = 1.03, p =

.36; and (iii) a non-significant interaction for product pride, F(2, 174) = 1.02, p = .36. Most of

the main effects of both product involvement and fit manipulation were significant. For example,

there was a significant main effect of product involvement (F(1, 174) = 59.38, p < .001) as well

as fit manipulation (F(2, 174) = 4.78, p = .009) for product pride.

Path Analysis: Moderated Mediation Effects of Athlete Involvement on Implicit Attitude, Explicit Cognitive Attitude, and Behavioral Intention

To examine the causal relationships among implicit attitude, explicit cognitive attitude,

and behavioral intention toward athlete, moderated mediation analysis was conducted using the

package lavaan (Rosseel, 2014) in R 2.15.2 (R Development Core Team, 2014). The researcher

created composite variables by summing and averaging subscales included in explicit cognitive

attitude and behavioral intention respectively. In the proposed research model, the fit

manipulation was specified as an independent variable by creating dummy codes; low condition

was coded as 1, moderate as 2, and high as 3. Therefore, the path coefficients from the

exogenous manipulation variable to implicit attitude represent the validity of experimental manipulation (Mackenzie, 2001). Responses were classified into a low (n = 88, the total scores

ranged from 1 to 3.9) and a high group (n = 92, the total scores ranged from 4 to 7) on the basis

of median split of the total score of the eight items measuring athlete involvement. The path

model, estimated using maximum likelihood estimation, showed an acceptable fit to the data (χ2

= 23.11, df = 18, RMSEA = .06, CFI = .98, SRMR = .04, p = .19; see Figure 25). The predictive

value and sufficiency of the model were also achieved in consideration of large values of the

model’s coefficient of determination and effect size (e.g., = .25 and = .33 for watching

intention and = .21 and = .27 for SNS friendship establishment intention for high

involvement group; Cohen, 1992).

The results of the moderated mediation analysis showed that the direct path from fit manipulation to implicit attitude was significant and positive (low involvement: β = .24, p < .05; high involvement: β = -.22, p < .05). The causal relationships from implicit attitude to explicit cognitive evaluation including expertise (low involvement: β = .27, p < .01; high involvement: β

= -.03, p > .05), trustworthiness (low involvement: β = .39, p < .001; high involvement: β = .05, p > .05), and attractiveness (low involvement: β = .37, p < .001; high involvement: β = .01, p >

.05) were significant only for low involvement group. The causal relationships from explicit cognitive evaluation including expertise, trustworthiness, and attractiveness to three types of behavioral intention were positive and significant. For example, trustworthiness positively and significantly influenced SNS friendship establishment only for high involvement group (low: β =

.11, p > .005; high: β = .24, p < .05). However, the direct path from attractiveness to SNS

friendship establishment was significant only for low involvement group (low: β = .31, p < .01;

high: β = –.08, p > .05). A summary of the results is depicted in Table 15 and Figure 25.

Path Analysis: Moderated Mediation Effects of Athlete Involvement on Implicit Attitude, Explicit Affective Attitude, and Behavioral Intention

To examine the causal relationships among implicit attitude, explicit affective attitude,

and behavioral intention toward athlete, moderated mediation analysis was conducted again

using the package lavaan (Rosseel, 2014) in R 2.15.2 (R Development Core Team, 2014). The

researcher created composite variables by summing and averaging subscales included in explicit

affective attitude and behavioral intention respectively. In the proposed research model, the fit

manipulation was specified as an independent variable by creating dummy codes; low condition

was coded as 1, moderate as 2, and high as 3. Therefore, the path coefficients from the

exogenous manipulation variable to implicit attitude represent the validity of experimental

manipulation (Mackenzie, 2001). Responses were classified into a low (n = 88, the total scores

ranged from 1 to 3.9) and a high group (n = 92, the total scores ranged from 4 to 7) on the basis

of median split of the total score of the eight items measuring athlete involvement. The path

model, estimated using maximum likelihood estimation, showed an acceptable fit to the data (χ2

= 24.14, df = 18, RMSEA = .06, CFI = .99, SRMR = .04, p = .15; see Figure 26). The predictive

value and sufficiency of the model were also achieved in consideration of large values of the

model’s coefficient of determination and effect size (e.g., = .23 and = .29 for watching

intention and = .31 and = .45 for watching recommendation intention for high

involvement group; Cohen, 1992).

= -.08, p > .05), trustworthiness (low involvement: β = .33, p < .01; high involvement: β = .18, p

< .05), and attractiveness (low involvement: β = .29, p < .01; high involvement: β = .09, p > .05)

were significant. The causal relationships from explicit cognitive evaluation including pleasure,

arousal, and pride to three types of behavioral intention were positive and significant. For

example, pleasure positively and significantly influenced SNS friendship establishment only for

low involvement group (low: β = .38, p < .05; high: β = -.08, p > .05). However, the direct path

from pride to SNS friendship establishment was significant only for high involvement group

(low: β = -.12, p > .05; high: β = .34, p < .05). A summary of the results is depicted in Table 16

and Figure 26.

Path Analysis: Moderated Mediation Effects of Product Involvement on Implicit Attitude, Explicit Cognitive Attitude, and Behavioral Intention

To examine the causal relationships among implicit attitude, explicit cognitive attitude,

and behavioral intention toward product, moderated mediation analysis was conducted again

using the package lavaan (Rosseel, 2014) in R 2.15.2 (R Development Core Team, 2014). The

researcher created composite variables by summing and averaging subscales included in explicit

cognitive attitude and behavioral intention respectively. In the proposed research model, the fit

manipulation was specified as an independent variable by creating dummy codes; low condition

was coded as 1, moderate as 2, and high as 3. Therefore, the path coefficients from the

exogenous manipulation variable to implicit attitude represent the validity of experimental manipulation (Mackenzie, 2001). Responses were classified into a low (n = 88, the total scores

ranged from 1 to 3.9) and a high group (n = 92, the total scores ranged from 4 to 7) on the basis

of median split of the total score of the eight items measuring product involvement. The path

model, estimated using maximum likelihood estimation, showed an acceptable fit to the data (χ2

= 15.6, df = 14, RMSEA = .04, CFI = .99, SRMR = .05, p = .338; see Figure 27). The predictive

value and sufficiency of the model were also achieved in consideration of large values of the

model’s coefficient of determination and effect size (e.g., = .43 and = .75 for

recommendation intention of endorsed brands through SNS for high involvement group; Cohen,

1992).

The results of the moderated mediation analysis showed that the direct path from fit

manipulation to implicit attitude was significant and positive only for low involvement group

(low involvement: β = .20, p < .05; high involvement: β = -.10, p > .05). The causal relationships

from implicit attitude to explicit cognitive evaluation including quality (low involvement: β =

.18, p < .05; high involvement: β = -.06, p > .05), trustworthiness (low involvement: β = .16, p >

.05; high involvement: β = -.06, p > .05), and attractiveness (low involvement: β = .21, p < .05;

high involvement: β = .16, p > .05) were significant. The causal relationships from explicit

cognitive evaluation including quality, trustworthiness, and attractiveness to three types of

behavioral intention were positive and significant. For example, the direct path from quality to

recommendation of endorsed product brand through SNS was significant for both low and high

product involvement groups (low: β = .28, p < .05; high: β = .49, p < .001). However,

trustworthiness positively and significantly influenced SNS friendship establishment with athlete endorser only for high product involvement group (low: β = .17, p > .05; high: β = .26, p < .05).

A summary of the results is depicted in Table 17 and Figure 27.

Path Analysis: Moderated Mediation Effects of Product Involvement on Implicit Attitude, Explicit Affective Attitude, and Behavioral Intention