ENDORSERS IN NARRATIVE ADVERTISEMENTS An online experiment
Arjen Schroth (5975964) Master’s Thesis
Graduate School of Communication Persuasive Communication
INDEX ABSTRACT……….………..4 INTRODUCTION……….………..……..5 THEORETICAL FRAMEWORK………….……….……….………8 Brand Attitude………..……….…….8 Purchase Intention………..……….10 Level of Identification……….….………11 Wishful Identification……….………..12 Similarity Identification………..……..13 METHOD.………15 Design………..…..15 Sample………..……….16 Procedure………..…………16 Material………..………..…….……18 Pilot Test………..………..18 Measurements………..………..………..19 Dependent Variables……….………19 Mediators……….……..20 Control Variables……….…….…21 Manipulation Check………..………..22
Analyses….………..………….23 RESULTS………..………24 Randomization Check……….…………..….…24 Manipulation Check……….………..…26 Hypotheses Testing………..………27 Hypothesis 1……….……….27 Hypothesis 2……….……….27 Hypothesis 3……….……….28
The mediating role of wishful identification……….………..30
Hypothesis 4……….……….33
The mediating role of similarity identification………..………..34
CONCLUSION……….……….………..36 DISCUSSION……….….…….37 Implications…………..……….………..…..37 Limitations…………..……….………..…..40 LITERATURE……….……42 APPENDICES……….……….…………47 Appendix A…………..……….………..…..47 Appendix B…………..……….………..…..48
Appendix D…………..……….………..…..57 ABSTRACT
In today's advertising business, the use of endorsers has become a beloved tool for brands to help reach, and influence consumers. In this experiment a first effort has been made to study the affect that different endorsers (celebrities vs. non celebrities) in narrative ads (ads) can have on brand attitudes, and indirectly on purchase intentions of consumers. Furthermore, the extent to which the levels of wishful identification, and similarity identification mediate this affect has also been examined. The results showed that celebrity endorsers, and non-celebrity endorsers do not influ-ence brand attitudes of consumers differently. Brand attitudes however, were found to be good predictors of the purchase intentions of consumers. The more positive the attitude of consumers, the higher is their purchase intention. The outcomes showed that wishful identification mediates the affect of endorsers on brand attitude, and similarity identification does not. Furthermore, people who are exposed to celebrity endorsers experience higher levels of wishful identification. The levels of similarity identification on the other hand, are the same for people who are exposed to celebrity endorsers, and non-celebrity endorsers. The findings of this research imply that the use of celebrity or non-celebrity endorsers have the same affect on brand attitude, which urges marketeers to start thinking about using less renowned endorsers, to preserve marketing budgets for more consumer engaging activities. Additionally, future campaigns could focus on trying to increase the levels of wishful identification, to create more positive brand attitudes among consumers.
Keywords: celebrity endorsement, non-celebrity endorsement, wishful identification, similarity identification, brand attitude, purchase intention.
INTRODUCTION
In their daily lives, consumers use narratives to understand their own lives, and the world around them. These narratives can help people to identify themselves within the society they live in (Escales, 2004). In advertising, brands also use narratives to get their messages across to
consumers. In the United States, 24.5% of television prime time ads contained narratives in 2012 (Chang, 2013). When people are exposed to narrative ads, and link the ad induced goals to their own goals, they form self-brand connections (Chang, 2013). Connections between brands, and consumers can lead to more engagement of consumers with the brand. To create these kinds of connections, brands use cues like music, and endorsers to relate to, and influence consumers.
This study aims to examine one of the cues that brands use to influence consumers, namely the influence of endorser type in narrative, audiovisual, sports ads has on consumers. Endorsers have been used to help promote products, and brands since the 1930s (Gaied & Rached, 2010). In this research the focus lies on two types of endorsers; celebrity endorsers, and non-celebrity endor-sers. A celebrity endorser can be defined as “a person who enjoys the public recognition and who uses this recognition in the name of goods while appearing with this one in advertising”
(McCracken, 1989), a person who is famous among the public (North, Bland & Ellis, 2005). Nike, for example, uses famous athletes like LeBron James (basketball athlete), as brand endorsers in ads. A non-celebrity endorser on the other hand, is a person in an ad that is not recognized by the public. When Nike uses an ordinary male student in an ad, he would not be recognized by the public, but he would still be the endorser for the Nike brand, thus he would be a non-celebrity endorser.
Earlier research has proven that celebrity endorsers can positively influence the brand attitude (Pradhan, Duraipandian & Sethi, 2014), and purchase intention (Pradhan et al., 2014; Weng, Cheng & Chu, 2013). Despite these studies focusing on the influence of celebrity endorsers on consumers, to this day very little research exists about the influence of non-celebrity endorsers on consumers. Some research has already focused on comparing these two endorser types (Buthada & Rollins, 2015; Gaied & Rached, 2010; Nataraajan & Chawla, 2008), but the results are still inconclusive. Gaied, and Rached (2010) for example, found that non-celebrities generate more positive attitudes, and were seen as more credible compared to celebrity endorsers. Other research, argued that celebrity endorsers are more credible, and that they generate more favorable levels of attitudes than non-celebrity endorsers (Buthada & Rollins, 2015). These different
findings could be explained by the fact that the researches were conducted in different market segments. It is plausible that people in different segments of different markets value different types of endorsers.
As brand attitude, and purchase intention have been found to be essential in the consumer decision process (Buthada & Rollins, 2015; Daneshvary & Schwer, 2000; Gaied & Rached, 2010; Wang et al., 2013), this research will aim to examine and compare the effects of the endorser type on both brand attitudes, and purchase intentions of consumers. The study could help brands to make more fitting, and efficient decisions while designing an ad campaign. Using non-celebrity endorsers could mean that brands do not have to spend their marketing budgets on large fee structures for celebrities (Erdogan, 1999; Nataraajan & Chawla, 1997). In 1998, athletes like Michael Jordan, and Tiger woods received fees of $45 million, and $25 million for
When people are exposed to a narrative, they often link back the information in the narrative to themselves (Chang, 2013; Escales, 2004). The level of identification of a consumer with an endorser can be seen as a process in which a person recognizes him- or herself, or wants to be, another person (Addis & Holbrook, 2010). In this case, the extent to which the receiver of a message adopts the attitude or behavior from an endorser, creates a relationship between the receiver, and endorser (Um, 2013). As a result, the feelings, and thoughts of the endorser are adopted (Addis & Holbrook, 2010). Sood (2002) made a distinction between two different forms of identification. Similarity identification arises when one feels that he or she is similar to the person seen in an ad, and wishful identification occurs when one wishes to be the person he or she sees in an ad (Sood, 2002). In this research, these two types of identification will be studied as mediators, to see how the level of identification mediates the effect of endorser type on brand attitude, and purchase intention. This led to the following research question:
RQ: To what extent does endorser type (celebrity vs. non-celebrity) in audiovisual narrative sports ads affect consumers’ brand attitude and purchase intention, and to what
extent does the level of identification with the endorser (wishful vs. similarity) mediate this affect?
This study will add to the knowledge about the affects of non-celebrity endorsement on
consumers. Very little research exists that has focused on comparing non-celebrity endorsements to celebrity endorsements (Bhutada & Rollins, 2015, Nataraajan & Chawla, 2008), and while celebrity endorsement has already proven to influence the decision-making process of
Wishful identification and similarity identification are expected to be evoked by different endorsers; a celebrity endorser will most likely evoke wishful identification, while a
non-celebrity endorser is expected to evoke more similarity identification (Sood, 2002). In daily practice, the results of this study could help marketers in designing successful campaigns.
Outcomes of this research could justify why brands pay these large fees to celebrity endorsers, or offer brands another less expensive alternative for endorsement, with non-celebrity endorsers.
THEORETICAL FRAMEWORK
In the theoretical framework, scientific theories that can explain the possible effects of endorsers on brand attitude, purchase intention, and the mediating role of level of identification are
discussed. First, the expected influence of endorser type on brand attitude will be worked out, and the connection between brand attitude, and purchase intention will be explained. Then the expected influence of brand attitude on purchase intention will be discussed, and lastly the expected influence of the level of identification with the endorser will be discussed. For an oversight of the conceptual model of this experiment see Figure 1 in Appendix Based on literature and the expectations of this study, hypotheses will be formulated, which will be answered, and verified in the results section of this research.
Brand Attitude
The brand attitude of consumers can be described as the overall value that consumers give to a brand, based upon its actions, products, image, and so forth (Faircloth, Capella & Alford, 2001).
A model that can explain how endorser type affects brand attitudes of consumers is the Extended Elaboration Likelihood Model (E-ELM) (Slater & Rouner, 2002). The E-ELM aims at explaining the effects of narrative messages, using the level of transportation into the story as the
determining predecessor of a persons’ attitude. According to the E-ELM people can be absorbed into a story. The level of absorption depends on four factors; storyline appeal, quality of
production, unobtrusiveness of persuasive text, and homophily. Of which homophily can be described as the feeling of similarity between the character in the narrative, and a consumer (Slater & Rouner, 2002). The E-ELM assumes that the higher the absorption, the more positive the attitude towards the narrative message. In accordance, Green, and Brock (2000) found that when people are highly absorbed in a narrative, they mark less false notes in those narratives. This means that highly absorbed people come up with less counterarguments against the message in the narrative, which leads to more positive attitudes of consumers (Green & Brock, 2000; Slater & Rouner, 2002).
As far as research about endorser type goes, Gaied, and Rached (2010) found that non-celebrity endorsers have more positive effects on brand attitudes than celebrity endorsers, whereas Erdogan (1999) argues that celebrity endorsers are more effective in generating desirable outcomes such as positive attitudes. In addition, a study by Metha (1994) showed no significant difference between brand attitudes after being exposed to a celebrity or a non-celebrity endorser. He did find that people who were exposed to a non-celebrity endorser focused more on the brand and its features, whereas people that were exposed to a celebrity endorser mainly focused on the
celebrity (Metha, 1994). A celebrity overshadowing the brand in an ad is known as the vampire effect (Erfgen, Zener & Sattler, 2015). Assuming that the vampire effect will distract consumers from following the story line of a narrative ad, the use of celebrity endorsers might lead to less absorption into the narrative. Kutiva, and Karlíček (2014) found that the vampire effect is less likely to occur when non-celebrity endorsers are used. In accordance with the E-ELM, this would mean that consumers will be more absorbed in a narrative that uses non-celebrity endorsers, be-cause the story line appeals to them more than the endorser, leading to more positive brand atti-tudes (Slater & Rouner, 2002). Furthermore, the E-ELM assumes that people become more ab-sorbed by a narrative when they experience homophily, which means that consumers feel that the character in the narrative, is similar or equal to the consumers’
self-image. As similarity identification is more likely to occur when people are exposed to a non-celebrity endorser (Bhuthada & Rollins, 2015; Sood, 2002) the following hypothesis was drawn up:
H1. Consumers who are exposed to a non-celebrity endorser in narrative sports ads will have a more positive attitude towards the brand, than consumers who are exposed to a celebrity
endorser. Purchase intention
Purchase intention can be described as a consumers’ behavioral intention to buy a brand after being exposed to persuasive message (Daneshvary & Schwer, 2000). When the purchase intent is high, a person will be more inclined to buy products from a certain brand. A connection exists
between brand attitude and purchase intention. This connection can be explained using Ajzens’ Theory of Planned Behavior (TPB) (Ajzen, 1985). In the TPB, Ajzen (1985) assumed that peoples’ behavioral intentions can be influenced by three factors, namely attitude, the subjective norm, and the perceived behavioral control. The attitude, according to the TPB, is influenced by two things; the beliefs of the consequences of a behavior, and the evaluation of the consequences of a behavior. This means that when a person positively evaluates what a brand does, and what it stands for, he or she is likely to have positive beliefs about the brand, which leads to a more positive brand attitude. As a result the person will be more likely to be planning to buy that particular brands’ products. Findings by Laroche, and Brisoux (1989) confirm the TPBs’ assumption. According to their findings positive brand attitudes positively influence consumers' purchase intentions. In addition, more recent research by Huang, Chou, and Lin (2010) has shown that brand attitudes are the most essential determinants of purchase intentions. Based on the TPB, and the literature the following hypothesis was drafted:
H2. The higher consumers’ brand attitudes, the higher purchases intentions.
Level of Identification
People are said to identify themselves with an endorser, when they feel that they know the endorser, or when they recognize their own characteristics in the endorser (Carlson, Donovan & Cuminsky, 2008). One could thus say that the identification with a character in a media message relates to a personal concept (Cohen, 2001). In the process of identification, a person often
distinction can be made between two types of identification; similarity identification, and wishful identification (Sood, 2002). Similarity identification arises when one feels that he or she is similar to the endorser of an ad, and wishful identification occurs when a person wishes to be like the endorser of an ad (Sood, 2002). The level of similarity identification is expected to mediate the effect of non-celebrity endorsement on brand attitude, and the level of wishful identification is expected to mediate the effects of celebrity endorsement on brand attitude. Thus, it is assumed that the level of identification with the endorser (similarity- vs. wishful
identification) mediates the effect that endorsers have on brand attitude, and indirectly influences purchase intention. Some theories, and models already exist, that can help explain the influence of the level of identification on the aforementioned effect. They are elaborated on, below.
Wishful identification
The wishful identification of consumers with celebrity endorsers can be explained using the Meaning Transfer Model (MTM). This model assumes that endorsers can transfer cultural and symbolic meanings to consumers (Erdogan, 1999). When, for example, a celebrity endorser has a specific cultural meaning for a specific group of consumers, the endorser will give a meaning to the message (i.e. LeBron James for basketball players, Cristiano Ronaldo for soccer players). According to McCracken (1989) a meaning transfer can lead to a construction in the self-image, which can lead to consumers identifying themselves with celebrity endorsers. The use of
inspirational models enhances consumers’s self-image, and can thus be seen as a form of wishful identification, which is likely to occur when the receiver of a message desires to actually be like,
or behaves in similar ways as the endorser. If LeBron James wears Nike shoes that were
designed for him in a Nike ad, consumers could think that if they buy Nike’s LeBron shoes, they become more like LeBron and will thus improve their basketball game. Based on their
attractiveness, and the cultural meaning that celebrity endorsers have, it is more likely that consumers will experience more wishful identification after being exposed to celebrities compared to when they are exposed to non-celebrities. In accordance with this assumption, McCracken (1989) presses the fact that celebrity endorsers have great cultural meanings, which can easily lead to meaning transfer. Also, research has shown that higher identification with celebrity endorsers leads to stronger relationships between the endorser, and the viewer (Boon & Lomore, 2001). Moreover, Funk, and Pritchard (2006) found that high levels of identification with an endorsers lead to more consistent attitudes. Based on the MTM and the literature, the following hypotheses were drafted:
H3a. Consumers that are exposed to celebrity endorsers will experience higher levels of wishful identification than consumers that are exposed to non-celebrity endorsers.
H3b. Consumers who experience high levels of wishful identification with the celebrity endorser will form more positive brand attitudes than consumers who experience low levels of wishful
identification with the celebrity endorser
Similarity Identification
The E-ELM can also be used to explain how identification can influence the brand attitude. According to the model, identification with characters is influenced by the perceived homophily
with the characters (Slater & Rouner, 2002). This means that when people feel that the characters in the media message are similar to themselves, they identify themselves more with the
characters, leading to more positive attitudes. Soods' similarity identification (Sood, 2002), is much like the homophily factor that Slater, and Rouner (2002) incorporated in the E-ELM. Homophily is in fact the same thing as similarity identification. Eyal, and Rubin (2003) confirm this assumption. They wrote that homophily measures the extent to which people who interact are similar in beliefs, social status, and how they are alike in other ways. A theory that can support the assumptions of homophily, is the Social Cognitive Theory (SCT) (Bandura, 1977). According to this theory, people learn how to behave from both the consequences of their of their own behavior, and by observing others with whom they identify themselves (Bandura, 2004; Slater & Rouner, 2002). This means that when people see the behavior of identifiable others in media, they imagine behaving in the same way as the identifiable other (Bandura, 2001; Cohen, 2001). Thus, in case of an endorser in narrative sports ads, exposure should lead to identification according to the E-ELM. Based on both the E-ELM, and the SCT consumers are thus more likely to form a more positive brand attitude when they experience a high level of homophily, or
similarity identification after being exposed to an audiovisual, narrative, sports ad. High degrees of homophily, and similarity identification are more likely to occur with non-celebrity endorsers, who are perceived as being equal and similar than with celebrity-endorsers. This led to the fol-lowing hypotheses:
H4a. Consumers that are exposed to non-celebrity endorsers will experience higher levels of similarity identification than consumers that are exposed to celebrity endorsers.
H4b. Consumers who experience high levels of similarity identification with a non-celebrity endorser will form more positive brand attitudes than the consumers who experience low levels
of similarity identification with a non-celebrity endorser.
METHOD Design
A between subjects factorial design has been used, with endorser type as main factor. ‘endorser type’ consists of two levels; the use of celebrity endorsers, and the use of non-celebrity endorsers in narrative, audiovisual, sports ads. This experiment uses a between subjects design, because the two groups, were exposed to two different condition (celebrity endorsement, or non-celebrity endorsement) are compared with each other. This means that two experimental groups arise, group 1 (celebrity endorser), and group 2 (non-celebrity endorser). Experimental designs have the advantage that two or more groups can be examined under controlled conditions (De Goede, 2009). By manipulating the two conditions, the hypotheses of this experiment can be researched. Some advantages of experimental designs is that they are cost efficient, they can reach many pe-ople in a short period of time, and pepe-ople voluntarily participate, leading to more accurate results (Reips, 2000). On the other hand, it is hard to control the surroundings of the participant during the participation of online experiments. Participants could, for example, be distracted by other people while participating. Furthermore, the chance of dropouts in online experiments is higher (Reips, 2000). Nonetheless, with the limited time available, the advantages of online experiment weigh up to the disadvantages.
Sample
For this online experiment a sample of 156 Facebook-using participants has been gathered. Participants had to be eighteen years or older to participate. If participants were younger, they were redirected to the end of the experiment and their data were excluded from the analyses. For two weeks, Facebook was used as a primary way to gather a convenience sample. The advantage of a convenience sample, is that it is easy to reach many people in a short period of time (Fricker & Schonlau, 2002). Disadvantage can be, that the people who are reached do not meet the experiments’ criteria, and there is little control over who participates in the experiment (Reips, 2000). In addition, this type of convenience sample also increases the risk of having participants that dropout, because they do not really care about the experiment. Still, with the limited time in mind, Facebook was used for gathering participants. On Facebook, participants could click on a link that redirected the participant directly to the experiment. The experiment was conducted using the online survey program Qualtrics, a program that randomly, but evenly, distributed the participants over the two conditions. After the exposure to one of the two conditions participants filled out a questionnaire (see Appendix B). Of the 156 participants that participated, 29
participants were excluded from further analyses due to missing data (n= 127). The manipulation check has been conducted using these 127 remaining participants.
Procedure
After clicking on the link, participants were led to the experiment. First, participants were asked to agree to an informed consent form, which stated that participants’ anonymity was ensured; the exact nature of the experiment was withheld from them to prevent biased data. Once the
experiment started, participants were first asked a couple of demographic questions (gender, age, education), and a question that checked whether participants were familiar with the Nike brand. After these questions were answered participants were exposed to one of the two conditions. One group was exposed to the celebrity condition, the other group was exposed to the non-celebrity condition. Following the ad exposure, participants were asked about their brand attitudes, and their purchase intentions. Subsequently, the levels of wishful identification, and similarity
identification were measured. The experiment was concluded by a manipulation check, to ensure that people recognized the endorser in the ad, and a question that asked participants whether they had seriously participated in the experiment while answering the questions. The scales that were used to measure ‘brand attitude', ‘purchase intention’, and the two forms of identification are described below. An oversight of all the dependent variables and their descriptives has been made (see Table 1.1).
Table 1.1
Descriptive statistics for the dependent, and the mediating variables across the conditions.
Variables Condition 1 (celebrity endorser) (n=35) Condition 2 (non-celebrity endorser) (n=47) Brand attitude M= 5.46 (SD=0.76) M= 5.23 (SD= .84) Purchase intention M= 5.22 (SD= 1.06) M= 4.8 (SD= 3.3) Wishful identification M= 3.42 (SD= 1.07) M= 2.70 (SD= 1.00) Similarity identification M= 3.37 (SD= 1.12) M= 3.35 (SD= 1.36)
Material
The variable ‘endorser type’ was manipulated to research the main effect of endorsers on brand attitude. Two different narrative, audiovisual, sports ads were used, to reflect the two
manipulated conditions (celebrity vs. non-celebrity endorser). Both narrative ads were Nike brand ads focused on basketball. The celebrity endorsement condition (condition 1) shows an ad of NBA (National Basketball Association) player LeBron James. James is seen as one of the most influential, best players basketball has ever known. In the audiovisual narrative you see James working out in the gym in Nike clothing, alternated with various career highlights of James in the NBA. The condition with the non-celebrity endorsement shows an ad of an average looking young, African-American man. The young man wears worn down Nike shoes and a Nike sports outfit, and the video is shot in the slums of a city. The visual content of the two conditions is different, but both conditions contain the same audio, to prevent bias in the results. By
synchronizing the audio in the conditions, any results that were found can be ascribed to the type of endorser, and not to the audio (Figure 2, Appendix A).
Pilot Test
A pilot test was conducted to ensure that participants interpreted the manipulations as they were supposed to be seen. A study among 23 people showed that both manipulations were successful. People were shown one of both conditions, after which they were asked whether the person they had just seen in the ad was a celebrity or not. If they thought the person was a celebrity, they were asked if they knew the name of the celebrity, and if they did not think the person was a celebrity they were redirected to the end of the pilot test. Fourteen people were exposed to
condition 1 (celebrity endorser), and nine were exposed to condition 2 (non-celebrity endorser). Of the fourteen people that were exposed to condition 1, twelve recognized LeBron James as a celebrity, and all twelve were able to identify him (see Appendix D). Two people thought he was no celebrity, and could not identify hem, which means that the manipulation was successful for 85.7 % (see Appendix D). Of the nine people that saw condition 2, eight thought the endorser was a non-celebrity, and one thought the endorser was a celebrity. This means that for condition 2, 88.9% of the manipulations were successful. As a result, it was assumed that the manipulati-ons for both conditimanipulati-ons were successful.
Measurements Dependent variables
Brand Attitude. Brand attitude was measured using a measurement scale by Spears, and Singh (2004). The scale measured brand attitude with five items, using seven-point semantic differential scales. The five items measured the following: unappealing - appealing, bad - good, unpleasant - pleasant, unfavorable - favorable, unlikeable - likable. To ensure that the scale was valid, first a dimension reduction factor analysis was performed in SPSS Statistics with all participants that completed the experiment (n= 127). The Varimax-rotated principal component analysis (PCA) generated one component with an Eigenvalue above 1, with a component loading of 3.17. This component explained 63.4% of the variance within the component. Subsequently, a reliability analysis was conducted, which revealed a reliable scale with a Crohnbach’s Alpha of .
85 (M=5.43, SD= .91). The higher participants score on the scale, the more positive is their brand attitude (see Appendix C.1).
Purchase Intention. To measure purchase intention again a measurement scale by Spears, and Singh (2004) was used. The scale consisted of five items, from which one item was
excluded. The item was excluded because it was not deemed relevant to measure the type of purchase intention examined in this research. The four remaining items were asked on
seven-point semantic differential scales. The items are: definitely no intent to buy - definitely intent to buy, very low interest to buy - very high interest to buy, definitely not buy it - definitely buy it, and probably not buy it - probably buy it. A dimension reduction factor analysis was conducted for the scale measuring purchase intention. The results of the Varimax-rotated principal component analysis (PCA) showed one component with an Eigen value above 1. The component explained 84.1% of the variance within the scale, with a component loading of 3.36. Subsequently, a reliability analysis was conducted, which revealed a Crohnbach’s Alpha of .94, proving that the scale is reliable (M= 4.85, SD=1.81). The higher participants score on the scale, the higher is their purchase intention (see Appendix C.2).
Mediators
Wishful Identification. A scale that consisted of three items measured wishful
identification (Hoffner, 1996). Participants were asked to indicate to what extent they agreed, or disagreed with each item using a five point Likert measurement scale. The items stated “I’d like to do the kinds of things the character in the Nike ad does", “The character in the Nike ad is the sort of person I want to be like myself.", and “I wish I could be more like the character in the
Nike ad.". To ensure that this scale was valid, a dimension reduction factor analysis was conducted. The Varimax-rotated principal component analysis (PCA) revealed one component with an Eigen value above 1, with a component loading of 2.40, that explained 79.9% of the variance within the component. Then a reliability analysis revealed a Crohnbach’s Alpha of .87 (M= 3.01, SD= 1.48). This higher participants scores on the scale, the higher the level of wishful identification (see Appendix C.3).
Similarity Identification. The similarity identification was measured using a scale by McCroskey, Richmond, and Daly (1975), that originally measured homophily with public figures, a scientific concept that means the same as similarity identification. The original scale consisted of 16 items that were measured using seven-point semantic differential scales. For this research five items were believed to measure the type of similarity identification that this
research examined most accurately. Participants were asked how they felt about the endorsers they had just seen in the narrative ad. Examples of used items are “the character in the Nike ad does not think like me thinks like me”, “the character in the Nike ad does not behave like me -behaves like me”, and "the character in the Nike ad is different from me - is similar to me”. The Varimax-rotated principal component analysis (PCA) generated one component with an Eigen value above 1, with a component loading of 3.54, that explained 70.7% of the variance within the component. The reliability analysis revealed a Crohnbach’s Alpha of .89 (M= 3.36, SD= 2.27). The higher participants score on the scale, the higher is their level of similarity identification (see Appendix C.4).
Gender. This variable was operationalized with one question: ‘What is your gender?’. Participants could choose between either 'male' or ‘female’ (50% male, and 50% female).
Age. Again, this control variable was operationalized with one question: 'What is your age?'. For this question participants were asked to indicate their age on a slider going from 0-90. In front of the slider it said ‘I am .. years old’, to prevent false answers. The average age of the participants was 26 (M=26.34, SD= 4.06).
Level of education. The control variable was operationalized with a singular question: “What is your highest level of education?”. For this question, the participants could choose between seven options: VMBO, HAVO, VWO, MBO, HBO, WO. Of all participants 90 % has followed HBO (22.5%) or WO (67.5%). The remaining 10 % was educated on lower levels. For an oversight of the descriptives of the control variables, see Table 1.2.
Manipulation Check Table 1.2
Descriptive statistics for the control variables across the conditions.
Variables Condition 1 (celebrity endorser) Condition 2 (non-celebrity en-dorser) Total Gender 45.7% man 54.3% woman n= 35 50% man 40% woman n= 46 n=81 Age M= 27.31 (SD= 5.39) n= 35 M= 26.43 (SD= 5.94) n= 47 n=82 Level of education 34.3 % low
65.7% high n= 35 31.1 % low 68.9% high n= 47 n= 82
A manipulation check was conducted for both conditions. Two variables measured whether the manipulation was successful. The first question asked: 'Was the person you have just seen in the ad a celebrity?'. Participants could answer 'yes' or 'no'. If yes was answered, the participant was redirected to the second question that asked the following: 'Who was the celebrity you have just seen?'. Participants could fill out the name of the celebrity here. Of the 59 people that were in the celebrity condition, nine participants failed to recognize the endorser as a celebrity. These
participants were excluded from the data. 50 participants were left, of whom fifteen were not able to name the celebrity correctly. These fifteen participants were also excluded from the data, and 35 participants remained in the celebrity group. Of the 69 participants that were exposed to the non-celebrity condition, 22 identified the non-celebrity endorser as an endorser. These participants were excluded from the data analysis, after which 47 participants remained in the non-celebrity condition. The following statistical analyses have been conducted using the data of the remaining participants (n= 82).
Analyses
The dataset of the experiment was downloaded as an SPSS file. First the incomplete data, and the participants with whom the manipulation did not succeed were deleted from the dataset.
Subsequently, the scales were constructed by performing factor analyses and reliability analyses of the measurement scales of ‘brand attitude’, 'purchase intention’, ‘level of wishful
identification’, and ‘level of similarity identification’. The manipulation check was performed with cross tabs, and a Chi Square test. Prior to any further analyses, randomization checks were
independent samples T-test for 'age'. To test the first hypothesis (H1), another independent samples T-test was conducted. To test the second hypothesis (H2) a singular regression analysis was executed. This type of analysis was chosen because both the dependent (purchase intention), and the independent variable (brand attitude), were measured on a numeric level, and both experimental group consisted of more that 30 participants. H3a was tested by using an
independent samples T-test with 'endorser type' as the independent variable, and 'level of wishful identification' as dependent variable. Likewise, H3b was also tested using an independent samples T-test. Only the participants from condition 1 were selected, and the T-test used a dichotomized 'level of wishful identification' as independent variable (low vs. high level) based on a median split, and 'brand attitude' as the dependent variable. The mediating role of wishful identification was examined by conducting three specific singular regression analyses. The following regressions were conducted: 1. endorser type and the level of identification, 2. endorser type and brand attitude, 3. level of identification and brand attitude (Baron & Kenny, 1986). With the results of these three regressions a Sobel test was conducted when the guidelines of Baron, and Kenny (1986) had been met, to statistically check the possible mediating role of 'level of wishful identification’. For the fourth hypothesis, again two sub-hypothesis have been examined. Both hypotheses were comparable to the the hypotheses from H3 and analyzed similarly.
RESULTS Randomization check
In the celebrity condition (condition 1), 45.7% was men (n= 16), and 54.3% was woman (n= 19). The average age of participants in condition 1 was 27.3 years (SD= 5.39), of whom the majority
studied on college education (WO) level (n= 23). In the non-celebrity condition, the group consisted of 24 men (53.3%), and 21 women (46.7%), with an average age of 25.6 years (SD= 2.42). Furthermore, of the participants in condition 2 the majority had studied on WO level (n= 31). Randomization checks were performed to ensure that the participants in this research were randomly assigned to the two conditions. First, the cross tab with ‘endorser type' and ‘gender’ showed that the distribution of men and women was equal between the two experimental conditions (see Table 1). Additionally, a Fisher Exact test revealed no significance (Fisher-exact p= .824) (see Table 2.1). The T-test for ‘endorser type' and continuous variable ‘age’ revealed that there is no significant difference between participants in condition 1 (M= 27.31, SD= 5.39), and participants in condition 2 (M= 26.43, SD= 5.94) regarding age, T(1,80)= .70, p= .488, CI [-1.65, 3.43] (see table 2.2). Subsequently, a cross tab for 'level of education’ (high education vs. low education) showed an equal distribution (see Table 1), and the Fisher Exact test showed no significant difference between conditions (Fisher-exact p= .718) (see Table 2.3). 90 % of all participants followed or had followed high education (HBO or WO), which means, a significant outcome of the Chi Square test was very improbable. Still, these outcomes show that the control variables were distributed evenly across the two experimental groups, thus the randomization was successful.
Manipulation check
A manipulation check has been conducted with all the participants who completed the
experiment (n= 127). Of the participants that were exposed to a celebrity (n= 59), 50 participants Table 2.1
Fisher’s Exact Test for ‘endorser type’ with ‘gender’.
Value df Asymptotic Significance (2-sided) Exact Sig. (2-sided) Exact Sig. (1-sided) Chi Square Test 0.068 1 0.795 Fisher’s Exact Test 0.824 0.487 Table 2.2
Independent Samples T-test for ‘endorser type’ with ‘age’.
t df Sig. (2-tailed) CI 95% Lower CI 95% Upper Equal Vari-ances As-sumed 0.696 80 0.488 -1.65154 3.42905 Table 2.3
Fisher’s Exact Test for ‘endorser type’ with ‘education’. Value df Asymptotic Significance (2-sided) Exact Sig. (2-sided) Exact Sig. (1-sided) Chi Square Test 0.194 1 0.660 Fisher’s Exact Test 0.718 0.468
(84.7%) recognized the celebrity as being a celebrity. 35 (70.0%) of the 50 remaining participants from the participants that were exposed to the celebrity endorser identified the endorser as “LeBron James”, or “LeBron”. 47 of the participants (67.6%) that were exposed to a non-celebrity (n= 69) said that the endorser was not a famous person. Two experimental groups remained for which the manipulation was successful; a celebrity group (n=35), and a
non-celebrity group (n=47).
Hypotheses testing Hypothesis 1
The first hypothesis assumed that non-celebrity endorsers will induce more positive brand attitudes than celebrity endorsers (H1). The results indicated that participants that were exposed to the celebrity condition (M= 5.55, SD= .71), did not have significantly less positive brand attitudes than participant that were exposed to the non-celebrity condition (M= 5.30, SD= .80), T(1,80)= 1.48, p= .144, CI [-.09, .59] (see Table 3.1). These results show that non-celebrities endorsers do not produce more positive brand attitudes among consumers, than celebrity endorsers. Therefore, H1 is rejected.
Hypothesis 2 Table 3.1
Independent Samples T-test for ‘endorser type’ with ‘attitude scale’ (H1).
t df Sig. (2-tailed) CI 95% Lower CI 95% Upper Equal Variances Assumed 1.475 80 .144 -.08782 .59134
The second hypothesis assumed that the more positive the brand attitudes the higher the purchases intention. To test this hypothesis, a singular regression analysis was conducted. The regression model was significant, F(1,80)= 54.60, p< .001. This means that the regression model can be used to predict the purchase intention of consumers based on their brand attitudes The model is strong in predicting purchase intentions of consumers (R2 = 0,41). This means that 41% of the differences in purchase intentions can be predicted based on the brand attitude of consumers. Brand attitude, b*= .64, t= 7.39, p< .001, CI [.75, 1.30], has a strong coherence with the purchase intentions of consumers (see Table 3.2 & 3.3). Meaning that brand attitude is a significant predictor of purchase intention, thus H2 is adopted.
Hypothesis 3
The third hypothesis consisted of two sub-hypotheses. The first hypothesis (H3a) assumed that people who are exposed to celebrity endorsers will experience higher levels of wishful
identification than people who are exposed to non-celebrity endorsers. The hypothesis was tested using an independent samples T-test. The test revealed that equal variances could be assumed,
Table 3.2
Model Summary of singular regression ‘attitude scale’ with ‘purchase intention scale’. (H2)
Model R R Square Adjusted R
Square
Std. Error of the Estimate
1 .637 .406 .398 .95722
Table 3.3
Coefficients table of singular regression. (H2) Model Unstandard-ized B Unstandard-ized Std. Error Standardized Beta t Sig. (Constant) -.673 .754 -.892 .375 AttitudeScale 1.021 .138 .637 7.389 .000
Levene’s Test for Equal Variances showed F= .11, p= .740. Furthermore, there is a significant difference in the experienced level of wishful identification between the two conditions, T(1,80)= 3.15, p= .002, CI [ .27, 1.19]. Participants that were exposed to the celebrity condition (M= 3.42, SD= 1.07) experienced higher levels of wishful identification than participants who were
exposed to the non-celebrity condition (M= 2.70, SD= 1.00) (see Table 1.2 & 3.4). Thus, hypothesis H3a was confirmed. Hypothesis H3b assumed that people who experience higher levels of wishful identification with the celebrity endorser have more positive brand attitudes than people who experience lower levels of wishful identification with the celebrity endorsers. The dichotomous variable ‘level of wishful identification’ was created using a median split. Levene’s Test of Equal Variances revealed no significant differences in variances between the groups (low vs. high identification), F(1,33)= .217, p= .644, thus equal variances were assumed. The T-test itself revealed no significant results of the wishful identification on the brand attitude for participants from condition 1 (n= 35) , T(1,33)= -.99, p= .329, CI [-.72, .25]. Participants that experienced higher levels of wishful identification did have slightly higher brand attitudes (M= 5.67, SD= .66) than participants who experienced lower levels of wishful identification (M= 5.43, SD= .76) (see Table 1.2 & 3.5), but the differences was not significant. Therefore, H3b was rejected.
Table 3.4
Independent Samples T-test for ‘endorser type’ with ‘wishfulID’. (H3a)
t df Sig. (2-tailed) CI 95% Lower CI 95% Upper Equal Vari-ances As-sumed 3.151 80 .002 .26767 1.18521
The mediating role of wishful identification
The first regression model is significant, F(1, 80)= 9.93, p= .002. This means that this model can be used to predict the level of wishful identification, based on the endorser type that consumers have seen (R2 = .11). 11% of the differences in the level of experienced wishful identification can be predicted based on the endorser type that a person was exposed to. The endorser type, b*= -. 33, t= -3.15, p= .002, CI [-1.19, -.27], has a weak coherence to the level of wishful identification (see Table 4.1 & 4.2). The second regression showed that the regression model is not significant, F(1, 80)= 2.18, p= .144. Therefore, this model can not be used to predict the brand attitudes of consumers, based on the endorser type that consumers have been exposed to (R2 = .03). Endorser type, b*= .-16, t= -1.48, p= .144, CI [-.59, .09], has no significant coherence to the brand
attitudes of consumers (see Table 4.3 & 4.4). Based on Baron, and Kenny (1986), and the non significant outcomes of the second regression it can be said that wishful identification does not play a mediating role. Still, the third regression was conducted. It showed that the regression model was significant, F(1, 80)= 8.55, p= .004. This means that the regression model can be used to predict the brand attitudes of consumers based on the level of experienced wishful
identification (R2 = .10). Ten percent of the differences in the brand attitudes of consumers can be predicted based on the level of experienced wishful identification. The level of wishful
Table 3.5
Independent Samples T-test for ‘wishfulID’ with ‘attitude scale’. (H3b)
t df Sig. (2-tailed) CI 95% Lower CI 95% Upper Equal Vari-ances As-sumed -.990 33 .329 -.72490 .25039
identification, b*= .31, t= 2.92, p= .004, CI [.07, .37], has a significant weak coherence to the brand attitudes of consumers (see Table 4.5 & 4.6). According to the guidelines of Baron, and Kenny (1986), there are a number of conditions that must hold, for a mediation. All three regressions must reveal significant affects of the independent variables on the dependent variables. Only then, can the mediating role of a mediator be assumed. Due to the fact that the second, regression did not show a significant affect of ‘endorser type' on ‘brand attitude’, it can be said that the level of wishful identification is no mediator to the studied affect of endorser type on the brand attitudes of consumers. To confirm the latter assumptions about the mediation, a Sobel Test was conducted. The test functions as a significance test, that tests for an indirect effect of an independent variable on a dependent variable via a mediating variable. The results showed a Sobel test score of -2.14, p= .032. Based on the assumptions made by Baron, and Kenny (1986) the level of wishful identification does not have a mediating effect, but the results of the Sobel test show that the level of wishful identification has a significant mediating on the affect that the endorser type has on brand attitudes of consumers.
Table 4.1
Model Summary of singular regression ‘endorser type’ with ‘wishfulID’. 1 of 3.
Model R R Square Adjusted R
Square
Std. Error of the Estimate
Table 4.2
Coefficients table of singular regression. 1 of 3. Model Unstandard-ized B Unstandard-ized Std. Error Standardized Beta t Sig. (Constant) 4.155 .380 10.929 .000 AttitudeScale -.726 .231 -.332 -3.151 .002 Table 4.3
Model Summary of singular regression ‘endorser type’ with ‘attitude scale’. 2 of 3.
Model R R Square Adjusted R
Square
Std. Error of the Estimate
1 .163 .026 .014 .76428
Table 4.4
Coefficients table of singular regression. 2 of 3. Model Unstandard-ized B Unstandard-ized Std. Error Standardized Beta t Sig. (Constant) 5.800 .281 20.613 .000 AttitudeScale -.252 .171 -.163 -1.475 .144 Table 4.5
Model Summary of singular regression ‘endorser type’ & ‘wishfulID’ with ‘attitude scale’. 3 of 3.
Model R R Square Adjusted R
Square
Std. Error of the Estimate
Hypothesis 4
H4a assumed that people who are exposed to a non-celebrity endorser will experience higher levels of similarity identification than people who are exposed to a celebrity endorser. To test this hypothesis an independent samples T-test was used. The results of the test showed that equal variances could be assumed according to the Levene’s Test for Equal Variances, F= 3.77, p= . 056. No significant difference was found in the levels of similarity identification between the two conditions, T(1, 79)= .08, p= .940, CI [ -.55, .59]. This means that participants that were exposed to the non-celebrity endorser (M= 3.35, SD= 1.36) did not experience significantly higher levels of similarity identification than participants who were exposed to the celebrity endorser (M= 3.37, SD= 1.12) (see Table 3.6). Therefore, hypothesis H4a was rejected. H4b assumed that people who experience higher levels of similarity identification with a non-celebrity endorser will have more positive brand attitudes than people who experience lower level of similarity identification. The independent samples T-test revealed that the Levene’s Test for Equal Variances was significant, F(1,45)= 3.92, p= .054, which meant that equal variances could be assumed between the high involvement group, and the low involvement group. The outcomes of the T-test showed no significant effect of the level of similarity on the brand attitude of the
Table 4.6
Coefficients table of singular regression. 3 of 3. Model Unstandard-ized B Unstandard-ized Std. Error Standardized Beta t Sig. (Constant) 4.742 .241 19.705 .000 WishfulID .220 .075 .311 2.924 .004
participants from condition 2 (n=45), T(1,45)= .48, p= .636. Participants who experienced high levels of similarity identification had slightly less positive attitudes (M= 5.17, SD= .63) than participants who experienced lower levels of similarity identification (M= 5.29, SD= 1.00) (see Table 3.7). Since the results were insignificant, H3b was rejected as well.
The mediating role of similarity identification
The first regression is not significant, F(1, 79)=.06 , p= .940. Therefore the model can not be used to predict the level of experienced similarity identification, based on the endorser type (R2 = .00). The endorser type, b*= -.01, t= -.08, p= .940, CI [-.59, .55], showed no coherence to the level of similarity identification (see Table 5.1 & 5.2). The second singular regression also revealed that the regression model was not significant, F(1, 80)= 2.18, p= .144. Therefore, this model could not be used to predict the brand attitudes of consumers, based on the endorser type (R2 = .03). There is no significant coherence between endorser type, b*= .-16, t= -1.48, p= .144,
Table 3.6
Independent Samples T-test for ‘endorser type’ with ‘similarityID’. (H4a)
t df Sig. (2-tailed) CI 95% Lower CI 95% Upper Equal Vari-ances As-sumed .076 79 .940 -.54548 .58879 Table 3.7
Independent Samples T-test for ‘similarityID’ with ‘attitude scale’. (H4b)
t df Sig. (2-tailed) CI 95% Lower CI 95% Upper Equal Vari-ances As-sumed .297 45 .768 -.40630 .54706
CI [-.59, .09], and the brand attitudes of consumers (see Table 5.3 & 5.4). Furthermore, the last regression was not significant, F(1, 79)= 2.17, p= .145, meaning that the regression model can not be used to predict the brand attitudes of consumers based on the level of experienced
similarity identification, and endorser type (R2 = .03). The level of similarity identification, b*= . 16, t= 1.47, p= .145, CI [-.04, .24], did not show significant coherence to the brand attitudes of consumers (see Table 5.5 & 5.6). Using the earlier mentioned guidelines of Baron, and Kenny (1986), non of the terms are met. Not one regression showed significant, meaning the level of similarity identification is not a mediator.
Table 5.1
Model Summary of singular regression ‘endorser type’ with ‘similarityID’. 1 of 3.
Model R R Square Adjusted R
Square
Std. Error of the Estimate
1 0.009 0.000 -0.013 1.26555
Table 5.2
Coefficients table of singular regression. 1 of 3. Model Unstandard-ized B Unstandard-ized Std. Error Standardized Beta t Sig. (Constant) 3.392 0.472 7.191 0.000 AttitudeScale -0.022 0.285 -0.009 -0.076 0.940 Table 5.3
Model Summary of singular regression ‘endorser type’ with ‘attitude scale’. 2 of 3.
Model R R Square Adjusted R
Square
Std. Error of the Estimate
CONCLUSION
This research aimed to examine the direct effect that different endorser types could have on brand attitudes, and the indirect effect on purchase intentions. Furthermore, the mediating roles of wishful identification and similarity identification have been examined. To answer the research question of this experiment, four hypotheses were drafted, and two additional model analyses have been performed, to determine possible mediators.
Table 5.4
Coefficients table of singular regression. 2 of 3. Model Unstandard-ized B Unstandard-ized Std. Error Standardized Beta t Sig. (Constant) 5.800 .281 20.613 .000 AttitudeScale -.252 .171 -.163 -1.475 .144 Table 5.5
Model Summary of singular regression ‘endorser type’ & ‘similarityID’ with ‘attitude scale’. 3 of 3.
Model R R Square Adjusted R
Square
Std. Error of the Estimate
1 .204 .042 .017 .80062
Table 4.6
Coefficients table of singular regression. 3 of 3. Model Unstandard-ized B Unstandard-ized Std. Error Standardized Beta t Sig. (Constant) 5.079 .243 20.876 .000 SimilarityID .100 .068 .163 1.472 .145
First, the results revealed no significant differences between the affect of celebrity, or non-celebrity endorsers (H1). Second, the results indeed revealed a significant relationship between brand attitude and purchase intention, more positive the brand attitudes lead to higher purchase intention (H2). Furthermore, brand attitude is not significantly influenced by the level of experienced wishful identification (H3b), but the level of wishful identification is higher when exposed to celebrity endorsers as compared to non-celebrity endorsers (H3a). Also, consumers who are exposed to non-celebrity endorsers do not experience higher levels of similarity identification than consumers that are exposed to celebrity endorsers (H4a), and there is no significant difference between the level of similarity identification that is experienced between consumers who are exposed to celebrity endorsers, and those who are exposed to non-celebrity endorsers H4b). Finally, the results show that wishful identification functions as a mediator between the endorser, and the affect the endorser has on the brand attitude of consumers, whereas similarity identification does not play a mediating rollin this affect. Altogether, the results of this study showed that endorser type does not affect brand attitudes of consumers, but brand attitudes do influence the purchase intentions of consumers. To finalize answering the research question, the level of wishful identification mediates, and the level of similarity identification does not mediate the affect that endorser type has on the brand attitudes of
consumers. Below, both the theoretical, and practical implications of this research will be set out. Thereafter, possible limitations of this research will be discussed, and research proposals for future research will be presented.
The findings, and the conclusions of this online experiment implicate a number of theoretical, and a number of practical implications. Because the findings of this research do not give
conclusive evidence of what works better, a celebrity or a non-celebrity. Future studies focusing on the subject could lead to more conclusive evidence about the affect of different types of endorsers on consumers. The results of endorser type on the brand attitude (H1) correspond to the findings of Metha (1994), but the findings differ from the conclusions by Gaied and Rached (2010). An explanation for the divergent findings, could be that Gaied, and Rached (2010) researched print ads, whereas this research examined audiovisual ads. Consumers could focus on different aspects when looking at ads in different media. Additionally, narratives might actually distract consumers from the celebrity in the ad, leading to less vampire effects (Erfgen et al., 2015). This assumption should however be examined for conclusive evidence. Furthermore, according to the outcomes a positive relation exists between brand attitude and purchase
intention, which matches findings by Huang et al. (2010), Laroche, and Brisoux (1989), and also confirms the assumptions of the TPB (Ajzen, 1985). Although, the results show that wishful identification has a mediating effect, and similarity identification has no mediating effect on brand attitude, more research must be done assess the consistency of the attitudes as a result of identification with the endorser, that Funk, and Pritchard (2006) found. The results showed that celebrities transfer meaning, leading to higher levels of wishful identification, confirming the conformations made by the Meaning Transfer model (Erdogan 1999). Higher levels of similarity identification as a result of homophily have not been found. This, in contrast to the assumptions of the E-ELM (Slater & Rouner, 2002). In the limitations sections some explanations are given that could have led to lower levels of identification.
This research is one of the first to research the influence of endorsers in narrative ads. In a commercial society, in which it is important to reach consumers, and most importantly to get the consumers involved with a brand, it is important to learn more about the combination of
endorsers and narratives, that are being used in combination more, and more often. This is important, because knowing which endorser most affects consumers when using narrative ads, could lead to more involvement with a brand, and more positive brand attitudes of consumers. By also researching the level of wishful-, and similarity identification as possible mediators, the reason why endorsers work, or do not work was studied in more depth. Furthermore, the results, and the conclusions of this study implicate that brands can anticipate on certain reactions from consumer while designing, and creating campaigns. According to the outcomes, brand attitudes of consumers are similar, regardless of the endorser type. Big, established brands could, based on the results for Nike, consider using non-celebrity endorsers, in stead of celebrity endorsers. By using cost efficient non-celebrities, a big part of the marketing budget could be spent on other forms of campaigning. Furthermore, marketeers who want to influence purchase intention, should focus on creating more positive brand attitudes, in stead of having to think of creative ways to increase purchase intention. After all, brand attitudes are essential for the purchase intentions of consumers (Huang et al., 2010). For marketeers it is important to efficiently reach their target groups, which makes knowing when to use which endorser type essential in the process of creating positive brand attitudes, and getting consumers to buy the brands' products. This study has shown that two different kinds of identification, do not aid a brand in getting the latter named result that marketeers, and brand aim for. Nonetheless, more research on the topic of endorser type and the possible mediating or moderating factors that influence brand attitude, and
purchase intention could provide brands with a more funded insight into what works or could work, and what does not on today's commercial society consumers.
Limitations
Following this experiment, it is important to mention and discuss some of the possible limitations related to this research. The discussed limitations can can be taken into account in future research, to help expand the scientific knowledge on the subject. The first limitation is the use of online surveys. 29 participants were excluded due to incomplete data. When using online tools for an experiment, there is no way for the researcher to control the surroundings of the participant, nor can the researcher stress the participants to complete the experiment. Thus, using online tools to can lead to incomplete, or unusable cases in datasets. In this research, for
example, 29 of the 156 initial responses had to be deleted due to missing data. It is possible, that more completed responses could have lead to different findings
Second, future experiments should consider using different stimulus materials. Although the pilot test indicated that the manipulation was successful (n= 26), the experiment produced different results. The manipulation failed among 46 participants. Future research aimed at Dutch consumers, should possibly use Dutch endorsers (celebrity, as well as non-celebrity). For this research, basketball playing endorsers were used in the stimulus material. The results however clearly showed that, although a pilot test was successful, the notoriety of LeBron James amongst consumers in the Netherlands was too low. Nevertheless, Dutch people may have less distinct opinions about basketball, whereas many Dutch people have funded opinions about for example
soccer, or field hockey. By using basketball, it was expected that the participants would watch the material carefully, without immediately disregarding the content of the material based on their existing opinions about a more familiar type of sport. By changing the endorsers and
changing the sport that is advocated in the future experimental materials, the manipulations could be more successful. Moreover, this experiment made use of male endorsers, but future research could use both male and female endorsers. It might be hard for a woman to identify herself with a male endorser, and the other way around for men, and female endorsers. Research by Boyd, and Shank (2004) for example found that endorsers of the same gender are found to more effective. By matching the gender of a participant with the gender of the endorser that the participant is exposed to, changing the sport that is advocated, and changing the endorsers the levels of identification could increase, leading to different results.
Third, for this experiment the brand Nike was chosen to be used in the experimental materials. In this research, Nike could be a little too familiar for Dutch people, meaning that the participants already had a strong existing opinion about Nike in the first place. In this experiment the existing opinions about Nike were not measured, though all participants indicated that they were familiar with the Nike brand. Future research could consider using a brand that is less renowned, or create fictional brands, to be used in the experimental materials. Also the existing opinion about the brand could be measured prior to exposure to the experimental stimulus.
Finally, future research should focus on possible mediators and moderators, such as the
credibility, the attractiveness, the reliability, and the expertise of the endorsers. The goal of this experiment was to attempt to expand the very little scientific knowledge about the comparison
phone use and overall access to information has much increased among consumers, it is
important for brands to know how to connect best to their audience. While consumers are already starting to accept, and believe more from peers, a possibility exists that in the near future,
non-celebrity endorsers will become more desirable than they used to be.
LITERATURE
1. Addis, M., & Holbrook, M.B. (2010). Consumers' identification and beyond: Attraction, reverence, and escapism in the evaluation of films. Psychology and Marketing, 27(9), 821-845. doi:10.1002/mar.20359
2. Ajzen, I. (1985). From intentions to actions: A theory of planned behavior. Berlin: Springer Berlin Heidelberg. doi:10.1007/978-3-642-69746-3_2
3. Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84(2), 191-215. doi:10.1037/0033-295X.84.2.191
4. Bandura, A. (2001). Social cognitive theory of mass communications. In J. Bryant, & D. Zillman (Eds.). Media effects: Advances in theory and research (pp. 121-153). Hillsdale: Lawrence Erlbaum. doi:10.1207/S1532785XMEP0303_03
5. Bandura, A. (2004). Health promotion by social cognitive means. Health Education & Behavior, 31(2), 143-164. doi:10.1177/1090198104263660
6. Baron, R.M., & Kenny, D.A. (1986). The Moderator-Mediator Variable Distinction in Social Psychological Research: Conceptual, Strategic, and Statistical Considerations. Journal of Personality and Social Psychology, 51(6), 1173-1182. doi:10.1037/0022-3514.51.6.1173
7. Boyd, T.C., & Shank, M.D. (2004). Athletes as Product Endorsers: The Effect of Gender and Product Relatedness. Sport Marketing Quarterly, 13(2), 82-93.
8. Boon, S.D., & Lomore, C.D. (2001). Admirer-celebrity relationships among young adults. Human Communication Research, 27(3), 432-465. doi:0.1111/j.1468-2958.2001.tb00788.x 9. Bhutada, N.S., & Rollins, B.L. (2015). Disease-specific direct-to-consumer advertising of pharmaceuticals: An examination of endorser type and gender effects on consumers’ attitudes and behaviors. Research in Social and Administrative Pharmacy, 11(6), 891-900. doi:10.1016/ j.sapharm.2015.02.003
10. Carlson, B.D., Donavan, D.T., & Cumiskey, K.J. (2008). Consumer-brand relationships in sport: brand personality and identification. International journal of retail & distribution man-agement, 37(4), 370-384. doi:10.1108/09590550910948592
11. Chang, C. (2013). Imagery Fluency and Narrative Advertising Effects. Journal of Advertising, 42(1), 54-68. doi:10.1080/00913367.2012.749087
12. Cohen, J. (2001). Defining identification: a theoretical look at the identification of audiences with media characters. Mass communication and society, 4(3), 245-264. doi:10.1207/
S15327825MCS0403_01
13. Daneshvary, R., & Schwer, R.K. (2000). The association endorsement and consumers’ intention to purchase. Journal of Consumer Marketing, 17(3), 203-213. doi:
10.1108/07363760010328987
14. De Goede, M. (2009). Het Experiment. In H. 't Hart, H. Boeije, J. Hox (Eds.), Onderzoeksmethoden (pp. 164-207), The Hague: Boom Lemma.
