Online Survey

72 expert had much more (practical) experience with the topic, wherefore a somewhat unstructured approach would be necessary, to allow the respondent to contribute with his knowledge and experience. Thus the interview was not controlled rigidly by the interviewer, but instead the respondent was given great opportunity to talk and comment as he preferred. This of course constituted a challenge for the interviewer, as much control of the interview was given to the respondent. It was thus necessary that the interviewer only directed the interview by asking follow up questions and referred to the topics that had been predefined as the ones that needed to be touched upon, as well as asking relevant enquiring questions. The interviewer thus needed to direct the respondent without imitating him, which was made difficult by the fact that the interviewer did not have the same experience as the respondent. It should thus be expected that the interview could have gone very differently depending on the interviewer and what things of what the respondent said he chose to follow up on.

73 The survey was kept fully structured and only included close-ended questions, as the purpose was to provide specific and quantifiable data pertaining to the proposed research questions.

The survey was structured into six main parts, which related to the conceptual model (see section 2.5 for the conceptual model). These parts pertained to:

1. Involvement 2. Personality

3. Attitude towards the ad 4. Attitude towards the brand 5. Effectiveness measures 6. Descriptive information

For a quick overview of the survey please see appendix D1 and for a full overview of how the survey appeared for the respondents please see appendix D2

The first two sections related to the relationship the respondent had with the celebrities and with Rolex, as well as how the respondents viewed their personalities as well as the personality of the respondent. Then in part three the respondents were faced with an ad showing different celebrities endorsing Rolex and asked to rate how they felt about the ad, an example of these ads are shown below in .

FIGURE 11 - ROLEX AD

Source: Own work

74 In section four they were then showed the same ads, but this time asked how they felt about the brand (Rolex) after seeing each of the ads. In section 5 the respondents were asked to rate how this would have affected their behavior (in terms of purchase, word of mouth and so on). Finally the respondents were asked in regards to their descriptive information.

In choosing the celebrities to include in the survey the aggregated results from the focus group assignment were used (see appendix B4). In this Roger Federer was clearly the celebrity that scored the highest, as however he already endorsed Rolex, he was discarded for fear that this would bias the answers of the respondents. As the celebrity scoring 2^nd highest were George Clooney, who then was chosen. George Clooney however already had an endorsement deal with Omega (Ambassadors - George Clooney), which also might bias the respondents. It was however believed that such would not be possible to avoid as all of the celebrities scoring high in the test had similar endorsement deals for other luxury watches. The ones with the lowest score were Mikkel Kessler and Jim Carrey, which were arguably because they are known for things (boxing and very silly comedy) that are very unrelated to luxury watches. Because of this they were discarded to likewise avoid biasing the respondents. Therefore Hugh Grant was chosen as the celebrity  whose  personality  did  not  fit  well  with  Rolex’s.      

Questions were measured by use of nominal, ratio and ordinal scales (Hansen K. , 2012) All ordinal scales were in the involvement and personality sections (sections 1 and 2) measured by use of a 7 point Likert scale. This to provide the respondents the opportunity to better distinguish their ratings, so as to provide a more nuanced picture. Attitudes (sections 3 and 4) were however measured by use of a 7 point semantic scale. This as such scale consists of opposed items that help the respondent to take a stand and answer the questions. At no point was “I  do  not  know”  or  “Not   of  relevance”  options  presented,  which  served  to  force  the  respondents  to  take  a  stand and answer all questions.

In designing the survey several measures were taken into account in order to keep the perception of burden down (Crawford, Couper, & Lamias, 2001). The point of this was to make it as instinctive, pleasant  and  “pain  free”  to  answer  the  survey  as  possible.  This  to  both  make  sure  the   respondents understood what they were answering and to ensure, that as many respondents as possible would finish the survey. This was particularly important as the survey would most likely

75 be somewhat longer than what most of the respondents would be used to. To keep the perception of burden down the phrasings was e.g. kept as short, precise and easily comprehensible as possible and arranged as systematically as possible. Further a progress indicator was implemented in order to allow the respondents to continuously keep track of their progress, which served to discourage respondent drop-offs.

Prior to launching the survey two independent pilot tests were conducted, where emphasis was put on whether the respondents had indeed understood the questions correctly. Both respondents had a couple  of  comments   in   relations  to   typo’s  and  the  likes,  but   nothing  to   the  overall  structure  and   both appeared to have understood all the questions correctly.

To  recruit  respondents  the  author’s’  social  networks  were  initially  utilized,  mainly  by  use  of  the   author’s   Facebook   account   and   direct   mails.  Next the survey was distributed through online forums such as urforum.dk, jubii.dk, trendsales.dk and Kandu.dk. It was deliberately stated that everyone would be liable to answer the survey.

3.2.1.1 Considerations

As the online survey will be the basis for testing the proposed hypotheses, many considerations have naturally been discussed prior to launching the survey. These considerations mainly pertained to designing the survey and recruiting the respondents.

Designing the survey:

As the first consideration it was discussed whether it would be appropriate for the survey to be fully structured or if the respondents should be able to provide some deeper insights to their answers by the use of open text boxes. As the purpose of the survey was however to test the proposed hypotheses, such text boxes would mainly serve to validate that the respondents had fully comprehended the question. As it was believed the survey would already be somewhat extensive it was argued that it would be better not to include such boxes, so as to limit the respondent drop-off. This to be able to get more respondents, which would potentially heighten the generalizability of the findings.

It was further discussed in what order the questions should be presented to the respondents, particularly in regards to the personality, attitude and effectiveness measures. At first it was argued

76 that all of such questions should be grouped together in regards to the given celebrity they would be concerning. So that first personality, attitude and effectiveness questions pertaining to George Clooney should be asked, and then the same questions for Hugh Grant and John Smith. However it was found that such structure was confusing for the respondents who, as a consequence, were not always sure of what they were answering. So to make the survey more intuitive the questions were instead grouped by question type, so that all personality questions would be asked first, then all attitude questions and lastly all effectiveness questions.

It has previously been laid out that personality-fit is interrelated with other factors that may come into play in terms of how effective a celebrity endorser will be. Thus it has naturally been discussed how such other factors may be excluded. This may however not be possible as they are very interrelated, but so as to be able to gauge their influence, the respondents were asked questions pertaining to these factors. As such the respondents were asked in regards to the physical attractiveness, expertise and credibility of each of the celebrities. Further, it was considered what questions to ask so as to make sure that the questions would be involving the relevant terms.

Scales:

It was considered whether it is prudent to use both a Likert and a semantic scale and whether this would have any influence if such results were to be cross-examined later on. It was chosen to use Likert scales for the brand personalities as this was how Geuens and Aaker had originally set it up.

Using different scales than the intended would thus arguably be misusing the concepts behind Geuens and Aakers brand personality concepts. Further a semantic scale was chosen for measuring the attitudes as this was believed to be the best way of getting the respondents to answer accurately. This as semantic scales essentially allow the respondents double the amount of items to consider, as they have two opposites to relate to each time they have to rate something. It was further considered whether using such different scales would later on influence the results if such constructs were to be cross-examined. This was however not found to give rise for concern as such tests would only focus on the connection between the two different constructs, wherefore they do not necessarily need to be using the exact same scale.

By   not   having   an   “I   do   not   know”   answer   one   risks   of   forcing   the   respondents   to   answer   something that they actually do not believe, but are merely answering because they have to. As it

77 however, in marketing, is generally believed that much of consumer reasoning happens subconsciously, it is argued that the subconscious of the respondents would be able to know what they were answering. Further, as the ordinal scales all have 7 points they have a natural middle, which the respondents could just answer if they truly did not have an opinion on the subject.

Choosing celebrities to include:

Considerations have been made in regards to what celebrities where to be included in the survey.

In order to make sure that the respondents would be basing their scoring on the personality of celebrities it was necessary to keep certain factors as constant as possible. As such much thought was put into finding celebrities that were of similar age, credibility, attractiveness and expertise (in regards to watches). Further it was strived that the two actual celebrities should be known for very much the same thing, as it was believed that if one e.g. was a golfer and the other a Formula 1 driver, the perceived image of their profession could possibly play a part. Further it was considered that the picture that was chosen for each celebrity may potentially have an influence if they are very different. To avoid such, pictures were chosen in which the celebrities wore a suit (without a tie) and smiled while they were looking a little away from the camera. Further it was strived to find celebrities that had not been involved in a scandal recently so that the respondents would not be fixating on that.

In picking the celebrities to use in the survey it was further considered that the celebrities current/former endorsement deals may also potentially affect the respondent. As such Roger Federer was e.g. not selected as he already endorses Rolex (see section 3.2.1). However, most of the celebrities that, in the focus group was, found to match Rolex well had previously endorsed certain brands of wristwatches. Brad Pitt and Tiger Woods have e.g. previously endorsed TAG Heuer (Celebrity wristwatch endorsements). However this only seems natural as there celebrities that may match Rolex well should also be expected to match other luxury wristwatch brands similarly well.

Further it was considered that the amount endorsement deals the celebrities have might also influence how the consumers view them. This as consumers generally tends to react more favorably to celebrities that only have a few exclusive endorsement deals (A new study of Exclusive Brand Endorsement By Celebrities, 2012).

78 Technical issues:

Because of unforeseen technical limitations in the used online survey generator it was not possible to arrange the questions fully as the authors would have liked. As such some of the questions who used the exact same rating scales were automatically made wider than others (see appendix D2).

This might make the respondents initially think that the scales were different, which may cause confusion and lead to untruthful replies or drop-offs. As however no way of altering this was found.

Further it was made so that respondents who answered that they did not know either of George Clooney, Hugh Grant or Rolex at all was automatically lead to the last page of the survey.

It was further made so that the respondents had to answer all questions, before they would be able to proceed. Further the survey was made so that the respondents were not able to go back and edit their responses, so as to get the immediate and unconscious decision.

Further it was made so that respondents could not answer their age as being more than 100 years old. This as it was argued that some respondents might by accident type one digit too many, and it was not believed possible that any person over 100 years would answer.

Recruiting respondents:

It was considered whether it would be necessary to recruit respondent   that   would   be   of   Rolex’  

target group (at least descriptively). This as it for Rolex would be most relevant to know how their target consumers behave and not how everyone else would behave. As however the focus of this thesis is not on Rolex, but Rolex has rather been used as the case company in order to test something, this was not believed to be necessary. Instead it would be more beneficially to not narrow down who could participate, so as to get as many respondents as possible, which could potentially strengthen the validity and generalizability of the findings.

79

4 Results

This section will present the results of the online survey. To set the scene the section starts by explaining how the data has been processed, so as to let the reader know how the results have been handled and how they have come to pass. Followingly the samples validity will be examined by comparing how normally distributed the data sets descriptive measures are. Then results pertaining to the conceptual model will be presented in three stages: first the measurement model will be evaluated, then the structural model, and at the end the relationships in the structural model will be tested and the estimated model presented. Lastly results beyond the testing of the conceptual model will be presented.