A Choice-Based Conjoint Analysis for the effect of movie-related attributes on customer uncertainty

(1)

A Choice-Based Conjoint Analysis for the effect of

movie-related attributes on customer uncertainty

(2)

A Choice-Based Conjoint Analysis for the effect of

movie-related attributes on customer uncertainty

Master Thesis Willem Visser

Faculty of Economics and Business University of Groningen

Completion date: 23-06-2014

Tweede Willemstraat 41A 9725 JH Groningen 06-33834394

willem_visser@outlook.com S1741306

Supervisor : dr. F. (Felix) Eggers

(3)

i

Management summary

This study is looking for the best strategy to reduce customer uncertainty regarding the choice of movies. This will be done by means of four different attributes. The attributes are studio information, word of mouth, reviews, and awards. Until now there is done research towards these attributes, because of the interest of people, and their preferences by moviegoers. No research is done facing the combination of the different attributes together. So far, studio’s do not have any idea how to approach moviegoers with these attributes. Therefore, it is of great importance to properly allocate the attributes, because taken them together will result in a decrease of customer uncertainty. The contribution of this study is to provide new insights in the movie world by means of the importance of the attributes. Thereby a choice-based conjoint analysis is used for determining the preference of people.

(4)

ii

Preface

This thesis forms the last part of my graduation. I started studying with the bachelor

Bedrijfseconomie. Hereafter I chose the Master Marketing Intelligence. During this Master a lot of research had to be done. In this thesis also much research is done towards the subject movies. I enjoyed writing this study and I think this will give a nice overview of my study career in the city of Groningen.

First of all, I would like to thank my supervisor Felix Eggers for his support, time, and great feedback. Moreover, I thank my fellow students who assisted me during this thesis. I also would like to thank my family, friends, and other people who filled in the questionnaire. Without this data I could not write this thesis. Lastly, I really would like to thank my parents, sister, girlfriend, and roommates for their great support.

(5)

iii

1. Introduction

The motion picture industry is an important industry, because this industry gets bigger and bigger. Today there are millions of people who go to the cinema. Over the last two decades, the quantity of investigation in the motion picture industry has been increased. This industry is the main supplier in the United States with regard to the delivery of entertainment products (Eliashberg et al 2006). Less research is known about the effect of movie trailers on people, or especially on moviegoers, knowing that trailers are an influential form of motion picture promotion (Devlin et al 2011). Furthermore, trailers have a great influence on the selection of a movie (Faber & O’Guinn 1984) and the success of a movie is mainly determined in the weeks before the release (Stapleton & Hughes 2005). This seems that movie trailers are an important method for advertising movies. They are seen on television, internet and before movies in cinemas. Making a trailer is not easy, because a number of shots have to be selected which has to trigger the moviegoer in to paying for that specific movie (Smeaton et al 2006). Most movies have a trailer, teaser, poster, or a combination to showcase the movie and to attract as much as possible moviegoers. But, because of the large offer and information about movies, moviegoers find it difficult to select the right choice which meet their expectations and therefore became uncertain. This choice depends on a number of attributes which I will describe in this study.

There are many definitions about movie trailers. According to Finsterwalder et al (2012), a movie trailer is a brief film 1- 3-min cinematic experience that usually displays images from a specific feature film while emphasizing its quality; it is created for the purpose of screening in theatres to promote a film’s theatrical release. In contrast to Finsterwalder et al (2012), in his study Wasko (2003) claims the run time of movie trailers is anywhere from 30 seconds to 4,5 minutes. Other people suggested that trailers can be no longer than two minutes (Bloombergcurrent.nl). Hughes & Stapleton (2005) define movie trailers as; a richly compacted passive audio-visual montage of the movie shown prior to current release. Therefore, movie trailers are important, because people are influenced by them when going or not going to a particular movie (Hixson 2006). Trailers are a commonly used technique to generate interest in a movie by ‘’directly targeting moviegoers at a time when they have already expressed interest in attending at least one movie’’ (Adams & Lubbers 2000). To summarize: trailers are a main technique used to introduce a movie to the public with the purpose of building expectations about an upcoming film by providing actual scenes (Hughes & Stapleton 2005).

(7)

2 seconds and are defined as; an image of the movie. This study has the main focus to target on these three forms of advertising.

Important is that customers make a deliberate/considered choice for a movie in which they reduce their uncertainty. Uncertainty is defined as: a lack of sure knowledge about the course of past events (Weitz 1989). Not only past events, but future events should also be more predictable. Therefore, this study is looking a manner to reduce uncertainty when choosing a movie. Transforming the definition above for this study leads to; customer uncertainty is a lack of sure knowledge about the choice of a movie.

Over the years, researchers have done much research towards movies. However, there are several reasons why the upcoming interest in the motion picture industry necessities additional research. First, the industry has high importance in the global industry. Second, the industry has a big cultural understanding and attracts attention. Third, the motion picture industry may help to better understand industries that share certain characteristics (Eliashberg et al 2006). This study should give a closer look in the motion picture industry by using combinations of attributes which are never explained before.

This research is focused on the relation between the attributes studio information, word of mouth (WOM), reviews, and awards and their influence on customer uncertainty whereby the attention lies on reasoning from a studio perspective. I chose for a studio perspective, because studios have a big influence how to promote a certain movie. The effect of this study is in particular intended on film marketers/film makers, because they can anticipate on the important attributes of moviegoers when they chose a movie. This research will also be beneficial for moviegoers, because their choice of movies became easier which results in a reduction in customer uncertainty. Finally, this study may contribute for the whole motion picture industry.

The academic improvement of this study is to provide new insights and new ideas regarding movies, whereby I will use a choice based conjoint analyses. This analysis makes use of utility estimates which are used to explain and/or predict consumer behavior. The aim of this analysis is that novel attribute combinations can be created whereby the effect of these attributes on uncertainty will be researched. Additionally, existing literature around movies is mostly about the general content (Finsterwalder et al 2012) whereby one object is viewed like the amount of music, actors/main characters, distributors etcetera (Gazley et al 2011). This paper will give more insight in a combination of attributes. These attributes together play an equally important role as the single attributes in existing literature. The purpose of this study is to research which attribute combinations are the most important and therefore will lead to a reduction in uncertainty by moviegoers whereby attention on influential moderators is given. Finally, segments can be created which makes it easier for studios to see which customers are related to which attributes and which groups. Furthermore, existing research on movie trailers did not make use of a conjoint analysis, in which these various attributes are examined. This achieves the following research questions:

What is the best strategy (combination of attributes) to reduce customer uncertainty with respect to the choice of movies?

(8)

(9)

4

2. Literature review

This section explains the different attributes. These attributes are found in literature and up to now there is no research done towards these attributes together. Because, there is not always information available on movies some literature is applied on other branches.

2.1 Motivation and choice of attributes

The introduction of this paper already mentioned the four attributes. Before I will clarify the chosen attributes I will first explain the concept of push and pull strategies, because this paper has chosen to use attributes which are all scaled under pull strategies.

Push/Pull strategies Until 2000, push media was the main form of communication. Push media is the

sending of commercial messages without the benefit of any major interaction with the receiver (Gobé 2001). An example is the traditional television, in which you turn on the TV, select a channel and watch the program or commercial delivered by the maker. Pull media is media where the content is requested by an individual. So, the difference between push and pull media is that the user requests each item individually by pull media. A good example of pull media is the world wide web which is a vast opportunity for engaging people in a dialogue (Gobé 2001). The reason for the choice of pull media in this research is that it is extremely easy to share information about movies with other people (Yong 2006). Therefore, the chosen attributes in this study are all pull strategies. This paper has chosen the following four attributes: Studio information, WOM, Reviews, and Awards. These attributes are based on literature. Below, the attributes are explained.

I will make a difference between studio related information or studio controlled information and non-studio controlled information (the descriptive attributes). The next sub header will display studio controlled information, thereafter the descriptive attributes are explained.

Studio controlled information

Studio information The first attribute in this research is studio information. Studios can perform

many alternatives to promote their movies. One way is to make a poster of the movie. Posters have been very important as a feature for motion picture marketing (Rhodes 2007). Posters exists out images that act as embodiments of their movies for collective memory (Rhodes 2007). Another alternative is to show a trailer. A trailer is a richly compacted passive audio-visual montage of the movie shown prior to current release (Hughes & Stapleton 2005). Lastly, studios can promote their movie by using a teaser. In the introduction is made clear that moviegoers do not always like trailers, because they are giving away too much information about the movie (Mackenzie 2007). Gazley (2011) concluded that moviegoers prefer a trailer above a poster, because active images are more popular than static views.

(10)

5 make the assumption that trailers, teasers and posters all have a time dimension. These time dimensions are expressed in levels which are mentioned in the research design part.

Non-studio controlled information (descriptive attributes)

WOM The second interpreted attribute in this study is WOM. The effects of WOM on product

evaluation and adoption are frequently described in literature (Awad et al 2004). Holbrook & Hirschman (1982) concluded that information posted by ‘regular’ moviegoers influences the mass tastes of going or not going to a movie. According to Chakravarty et al (2010), a source of pre-release movie-related information can be WOM. They also state that information about movies in the form of WOM on the internet has experienced an explosive growth in recent years. WOM exists in two forms. These are positive and negative WOM (Chakravarty et al 2010). Ahluwalia et al (1997) researched that recipients place a greater weight by negative WOM compared to positive WOM. It appears that moviegoers often rely on WOM through social media when they will watch a movie (Forrester Research 2000). For movie makers, it is therefore important to put down a good movie and attract the right people. Examples are the numerous sites that provide movie information or discussion boards. Shankar & Batra (2009) researched the effect of WOM on purchases and came with the result that WOM nowadays has a big influence on purchases. Translated the concept WOM to the internet reveals whether the movie is worth watching (Chakravarty et al 2010). The text above applies to WOM in social media. Examples of social media are Facebook, Twitter, Linkedin, Instagram, and MySpace. There is also WOM in the form of personal communication. Eliashberg et al (2006) states that personal WOM is an important driver of the success of entertainment goods. According to Yong (2006) personal WOM influences moviegoers. Examples of personal communication are the opinions of family and friends. So, the question arises if moviegoers prefer personal WOM or online WOM. Beside personal communication and social media, IMDB is another form of WOM. IMDB is a form of social media, but I will gather this concept alone and not under social media, because IMDB is a very popular item in the movie world whereby descriptions about movies are given (in this way IMDB could be ranged under Reviews, but as IMDB is a form of social media it is decided to put it under this header). IMDB is a form of WOM, because of the passing of information from one person to another person(s). It has become a favored online forum where moviegoers can interact with each other and contains useful and up to date information about all movies (Dodds 2006).

Reviews The next attribute in this study are reviews. According to Chakravarty et al (2010),

(11)

6 (experts) who provide reviews which helps consumers make good choices (Kirmani & Rao 2000). These critics have specialized knowledge about a product category, in this case movies (West & Broniarczyk 1998). Another manner to measure movie reviews is by the number of ratings a movie gets on a particular site (Marshall et al 2013). Hereby, a 1 till 5 rating review can be used for a movie. In this study consumer communities and professional critics are used for measuring reviews.

Awards The last attribute that can be deduced from literature are awards. Providing awards exists in

many industries. Examples are the automobile, hotel, and wine industries (Zhuang et al 2014). The motion picture industry is also giving awards. However, until now literature paid little attention to the influence of movie awards on product performance (Zhuang et al 2014). Nelson et al (2001), researched the effects of awards on sales. They concluded that awards have a significant influence on sales. These awards are given to movies who deserve this. Examples are the MTV movie awards and the Academy awards. According to Gemser et al (2008), an award may function as a signal of quality that helps consumers in their product selection process. The amount of awards a movie receives can be crucial for moviegoers to go to that particular movie or not, because people are influenced by awards in their choice (Nelson et al 2001). Hennig-Thurau et al (2004) researched awards in combination with movies and concluded that the success of a movie depends on the number of awards. Therefore, potential moviegoers are looking for believable signals, like awards, to determine a movie before they actually see it (Basuroy et al 2006). This raises the question: Are moviegoers sensible for the amount of awards a movie has?

Because it is not easy to find one word that captures all attributes together, from now on when I talk about movie-related attributes I mean all attributes (Studio information, WOM, Reviews, and Awards) together.

2.2 Customer uncertainty

In November 2013, Li et al, published an article about uncertainty theories. They state that uncertainty is everywhere and happens in any place in the world. Customer uncertainty can be categorized into two groups (see the figure below). These are aleatory uncertainty and epistemic uncertainty. Aleatory uncertainty, like the flipping of a coin, exists naturally without any human interaction or knowledge. This kind of uncertainty cannot be eliminated or reduced by knowledge. Epistemic uncertainty is the lack of knowledge and or providing the wrong information towards people. It is a type of uncertainty that can be reduced and sometimes even be eliminated (Li et al 2013). Therefore, this study only focuses on epistemic uncertainty whereby the main goal is to make people more certain regarding their choice of movies. Certainty, indicates that people are able to predict a situation or make the good choice for a situation (Tversky & Kahneman 1974).

(12)

7 There are several definitions of customer uncertainty in literature. Customer uncertainty related to choice means not being sure which alternatives moviegoers prefer (Chung Hun & Cranage 2010). Homburg et al (2012) state that giving the right information leads to a reduction in customer uncertainty. However, Grice (1989) claims that more information leads to more uncertainty, because customers may interpret this information incorrect. According to Kardes (1994), unfavorable information increases customer uncertainty and moviegoers with customer uncertainty may not know what the attributes and alternatives are which influences the choice they make. Film makers can change their minds by anticipating on these attributes and alternatives (which is done in this study). Therefore, moviegoers need a number of clear alternatives, provided by film makers, which should lead to a reduction in customer uncertainty relating to their choice of movies.

Uncertainty about demand, information, and choice is the most recognizable distinctive characteristic in the motion picture industry (Basuroy et al 2003). The following quote supports this; ‘’No-one can tell you what a movie is going to do in the market place…. Not until the films opens in a darkened theater and sparks fly up between the screen and the audience’’ (Gemser et al 2008). Therefore, it is difficult to know in advance whether a movie meets the requirements of a moviegoer in his or her choice (Sawhney & Eliashberg 1996). There is thus an incentive for potential moviegoers to look for credible signs, like the mentioned attributes under 2.1, that helps them in their movie choice to determine uncertainty (Basuroy et al 2006). The ultimate definition for customer uncertainty in this study, which is already used in the introduction, is: a lack of sure knowledge about the choice of a movie.

2.3 Movie related attributes on customer uncertainty

In literature, much research has been done towards the relationship between the movie related attributes in this study and customer uncertainty. However, these attributes are examined individually in combination with customer uncertainty. In this research the attributes studio information, WOM, reviews, and awards are analyzed together in their influence on customer uncertainty. In their study Dubois et al (2011), found that communication between movie related attributes affects customer uncertainty. They also mentioned that people interpret this communication as facts causing consumers to be more uncertain, because of the amount of different interpretations.

(13)

8 Figure 2: Communication (between movie related attributes) and uncertainty reduction

2.4 Moderators

In this part the moderators of the study will be described. These moderators will be used for the CBC-analysis (which is explained in chapter 3.1), but are no input of the CBC-analysis. The input of the CBC-analysis are the attributes described under 2.1. It is to be expected that the moderator variables are important for segmenting the moviegoers with respect to their choice of a movie.

2.4.1 Emotional responses/mood

For hedonic products (like movies), consumers describe their needs and expectations in the form of specific emotions (Shanker & Olshavsky 1995). According to Frijda (1993), emotions refer to subjective feeling states that are directed toward a specific stimulus. The emotional state of a person plays an important role in the choice (Bagozzi et al 2000). Therefore mood or emotional state are important triggers for structured thinking (Baas et al 2012) whereby emotions can be divided in two dimensions. These are positive and negative emotions (Bloemer & Ruyter 1999). Examples of positive emotions are happy, comfortable, relaxed, active, and proud (Bloemer & Ruyter 1999). Examples of negative emotions are anxious, sad, angry, depressed and guilty (White 2010).

(14)

9

2.4.2 Emotional responses on customer uncertainty

Much research has been done towards the relationship between emotions and customer uncertainty (Oliver & Rust 1997). Individuals have to feel good when buying products or using a service of a firm. If this is not the case there will be a loss for the company, because of a reduction in customers. This is reflected in unfavorable profits and negative WOM (Das 2013). During the time, a pattern can arise between emotions and uncertainty (Lerner & Keltner 2000). People who are in a state of positive emotions, like happiness, feel more certain about the choices they made and the upcoming things. On the other side, people who expresses negative emotions feel more uncertain about the choices and upcoming choices they made (Baas et al 2011). Moreover, emotional suppression can result in poorer outcomes (Butler et al 2003) and poorer relationships (Wang & Groth 2014). According to Shanker & Olshavsky (1995), consumers describe their desires and expectations for movies in terms of specific emotions, which has an influence on customer uncertainty. A common way to measure emotions in a study is through a survey (Bagozzi et al 2000). In the study of White (2010) respondents complete a questionnaire to capture their emotional status. After this, conclusions can be drawn about the influence of emotional responses on uncertainty. Wang & Groth (2014) states that emotional responses can change customer outcomes. In their study the effect of emotions on customer outcomes, like uncertainty, is researched. Wang & Groth (2014) also researches relationship strength and service personalization, but this is not the intent in this study. This research will deal with the effect of the emotional state on uncertainty when looking at the moviegoer.

2.4.3 Socio-demographic variables

The relation between the attributes (studio information, WOM, reviews, and awards) /movie related attributes and customer uncertainty can be influenced by the type of customer. Socio-demographic variables are included in this study to see if there are any differences between moviegoers. Variables that are applicable for this study are age, gender, and education. According to Moller & Karppinen (1983), age and education do influence movie behavior. They concluded that high educated moviegoers are extremely sensible for critics, whereas low educated moviegoers have the opposite view. They also found that moviegoers between the ages of 21-30 have another feeling about the popularity of the movie than moviegoers over 30. To summarize the conclusion of Moller & Karpinnen (1983); consumers attending different types of movies and belonging to different education and age categories have distinctly different motivation bases underlying their movie choices. There is not much literature known about the combination of gender and choice of movies. It is to be expected that gender does have an influence on the choice moviegoers made when watching a movie. Therefore, it can be predicted that men made different choices than women. This result has also an impact on uncertainty.

2.4.4 Genre preference

(15)

10 genres. This ensures that moviegoers do not have to wait long for their favorite genre (Wierenga 2006). In contrast, Eliashberg & Sawhney (1994) argue genre is not a variable for predicting enjoyment when choosing a movie.

2.4.5 The relation between genre and customer uncertainty

(16)

11

2.5 Conceptual model

This paper includes four attributes which will be input for the choice-based conjoint analysis. To give an overview of the study a conceptual model (figure 3) is established. This study has customer uncertainty as dependent variable and the four attributes as independent variable. The moderators emotional responses, socio-demographic variables, and genre preference are also shown. To keep the model simple, it is decided not to include the levels of the attributes in the conceptual model. The levels of the attributes are displayed in the research design chapter.

(17)

12

3. Research Design

3.1 Method

To find out the best way to reduce customer uncertainty, on the basis of the choice customers make about a certain movie it makes sense to look at the preferences of moviegoers.

An appropriate method to test this is the conjoint analysis. In the middle of the 70's, the CBC analysis received much attention. In the 80’s this increased tenfold and during the nineties almost every field makes use of this type of analysis (Hair et al 2010). Conjoint analysis is a multivariate technique developed specifically to understand how respondents establish preferences for any type of object (products, services, or ideas). Therefore in this study a choice-based conjoint analysis is used (from now CBC-analysis). There is many literature available about the CBC-analysis. In CBC-analysis, respondents are shown a set of combinations of attributes (i.e. profiles) and are asked to indicate which of the combinations of profiles they would chose (Akaichi et al 2013). These profiles exists out attributes and levels. Reason for the existence of attributes and levels is the representation of total worth or overall preference of an object. This representation is also called the utility of an object. The utility can be estimated by summing up the product parts of the object, also called part-worths (Hair et al 2010). The part-worth is measured on a common scale whereby the relative importance of a factor can be estimated. This importance is calculated through the difference between the highest and lowest values of the levels divided by the sum of the ranges (Hair et al 2010).

In this way, the preference of the respondent can be determined whereby the choice of a movie is the dependent variable. Choosing a preferred profile is similar to daily life, where people also have to make their favorite choice when shopping, doing sport etcetera. So, CBC-analysis is a widely used method to figure out consumers specific preferences. According to Hair et al (2010), there are three types of conjoint analysis. These are CBC-analysis, Traditional Conjoint analysis and Adaptive Choice analysis. CBC-analysis is used by six or fewer attributes, Traditional Conjoint is used by fewer than ten attributes and Adaptive Choice is used by ten or more attributes. This study uses four attributes, so a CBC-analysis is appropriate. There are three other reasons why I chose for the CBC-analysis. Firstly, choice is the behavior of interest, and models measured on choice have an advantage in forecasting choice behavior. Secondly, choice set tasks can be characterized to study the effect of choice set configuration on choice. Lastly, models grant direct forecast of choice shares which makes it easy to estimate (Elrod et al 1992). This study also contains a none-choice option. This is an option whereby respondents have the choice of choosing none of the specified alternatives (Hair et al 2010). A none-option adds realism to the alternatives, because in real life respondents do not have to chose or buy something that is compulsory (Johnson & Orme 1996). The none-option is also interesting for the researcher, because it creates absolute as well as relative effects (Hair et al 2010). After choosing an alternative the researcher tries to recognize the preference structure of a respondent (Hair et al 2010).

3.2 Attributes and levels

(18)

13 in biased outcomes. But the difference between two and three levels is small which made it not a big problem. Table 1 expresses the attributes with their corresponding levels whereby all attributes are nominal.

Table 1: attributes and levels

Another important aspect is the identification of attribute levels which means that each attribute level is assumed to be mutually exclusive of the others. So, a product or service, like a movie, has one and only one level of that attribute. This is visible in table 1 where each attributes contains out unique levels. The maximum number of attributes that can be used in a CBC-analysis is six. Therefore this study meets the requirements to perform a CBC-analysis (Hair et al 2010). Furthermore, the general amount of levels per attribute and the number of alternatives per choice set is 2-5. This study also satisfies these requirements.

Because this study is written from a studio perspective, the first attribute in this study is called Studio information and exists out the levels poster, trailer and teaser. Since little information is known about the effect of time dimensions on studio information, this study examines what the effect of such a time dimension is. A poster has a time dimension of zero seconds. A teaser has a length of 60 seconds and a trailer has a length of 2,5 minutes. In the survey these time dimensions are further explained.

The second attribute in this study is WOM. There is much literature on WOM available. This study assumes WOM is composed of the levels personal communication, social media and IMDB. In the survey is mentioned that this attribute is about short information of a movie.

(19)

14 The last attribute in this research are awards. This attribute consists of two levels. These are yes (information about awards) and no (no information about awards). In the survey the meaning of awards is described, because awards can have a large impact on the choice of a movie. The reason why this attribute is chosen is due the worldwide attention for events , like the MTV movie award or the Academy awards.

3.3 Questionnaire

This study makes use of a survey in the form of a questionnaire. Nowadays there are many ways to collect a survey. In this research data is collected from the internet by means of a survey website. There is chosen for the internet, because almost every person has the ability to use internet which made it easy to gather data. The program called preferencelab is used for making this survey. Moreover, the program is also used for putting the questionnaire online.

The questionnaire starts with an introduction (appendix A) where after a couple of demographic questions are asked followed by a question which asks for the emotional mood a respondent is currently on. Hereafter, a question about the favorite genre is asked followed by a question which tests the customer uncertainty about the choice of a movie (appendix B). Genre preference is measured by a question which incorporated the 10-most important genres. These are: Action, Comedy, Horror, Romance, War, Adventure, Drama, Fantasy, Thriller, and Science-fiction. The emotional mood of people is measured by 4 positive feelings and 4 negative feelings. The positive feelings are happy, proud, energetic, and comfortable. Sad, anxious, angry, and guilty are the negative feelings. Socio-demographic variables are estimated by looking at age, gender and education. Table 2 is showing an overview of the moderators and their measurement levels.

After the ‘general’ questions a short text is written about the explanation of the different attributes. Next, several questions (the conjoint questions) about the preference for attributes are displayed by showing the respondents attributes and their corresponding levels (appendix C and D). It became clear that respondents had to chose one of the three alternatives. The first two alternatives can be seen as strategies. The last alternative is the none-choice option. This option is clicked if respondents prefer a short description of the movie. Lastly, the survey contains a question in which respondents had to answer their preference for all four attributes (appendix E).

Table 2: moderators and measurement levels

3.4 Sample

(20)

15 Family, friends, and colleagues were asked by mail or other social media for responding this survey whereby distribution via WOM, Facebook, Twitter, Email, and other stuff was used to reach as much people as possible.

3.5 Choice design and holdout task

After specifying the method, attributes and levels, questionnaire, and sample the next step is to determine the data collection. Data collection exists out the type of presentation method (trade-off, full-profile, or pair wise comparison), and the type of response variable. The represented presentation method in this study is the full-profile method. This approach is describing alternatives (in this case three) in which all four attributes are mentioned. The advantage of this approach is the realistic description by showing profiles in terms of a level for each factor (Hair et al 2010). The second approach of data collection is the type of response variable. Because this study exists out ten levels and four attributes, respondents cannot pass through all possible profiles. The number of profiles with four attributes and 10 levels would become (3 levels x 3 levels x 2 levels x 2 levels) 36 and therefore not practical. For this reason a fractional factorial design is chosen which means that a fraction of the full number of alternatives is shown to respondents. The next question is of the number of profiles is appropriate for calculating the part-worths and what the minimum number of profiles is that must be evaluated by each respondent (Hair et al 2010). This minimum number of profiles is calculated as:

Minimum number of profiles = Total number of levels – Number of attributes + 1

According to this study the minimum number of profiles is (10 – 4 + 1) 7. There is chosen for nine choice sets consisting of three different alternatives for estimation of the model which ultimately results in 30 profiles ((9 + 1) * 3). Beside this appropriateness, the number of profiles has to satisfy two other measures. The first measure is orthogonality, which means that there is no correlation among the levels of an attribute. The second measure is the balanced design (each level in a factor appears the same number of times). This study meets both measurements. Within the three alternatives it is chosen to incorporate also a no-choice option (3.3).

The asked question regarding the choice sets is: which of the three options informs you the most about a movie you want to watch (see figure 4 for a screenshot of a conjoint question). The number of choice sets in a CBC-analysis is usually between the 8-16, but after 12 questions respondents’ are confronted with fatigue effects. These are effects that depends on the complexity of the choice situation. A good way to reduce this fatigue effect is to motivate people (Hair et al 2010). This study contains nine choice sets, therefore there will be no fatigue effects. Results are further estimated through the program Latent Gold Choice whereby the aim of the survey was to collect 200 respondents to get the most efficient results.

(21)

16 Figure 4: screenshot of a conjoint question

3.6 Describing the procedure

The main goal of this survey is to test the conceptual model. Hereby the impact of the moderators on the relation between the attributes and customer uncertainty will be researched. The advantage of this type of analysis is that it can be seen if there are any similarities between certain respondents or groups. These groups or respondents are than pooled together to form a segment. This estimation representing segments is one method. Another method, also known as an aggregate level analysis, is a method which is based on a one class model. Advantages of this method are a reduction in the data collection task for reducing the number of evaluations per person, second this method can estimate interactions between attributes and third a greater statistical efficiency is achieved through the use of more observations (Hair et al 2010). Both analyses are used in this study.

One of the first steps in a choice based conjoint analysis is the calculation of parth-worths. Part-worths are composed of three types or models. Hereby it is possible to see how the levels of the different factors are related to each other. The three models are; a linear model, a quadratic or ideal form, and a part-worth model. A linear model is the simplest of the three. Only one coefficient has to be calculated which is than multiplied by the value of the levels. A more complex method is the quadratic or ideal form where more parameters have to be estimated. The part-worth model is the most complex model, because in this model many parameters have to be calculated whereby each level has an individual parameter which takes lot of time to determine. This study makes use of the part-worth model.

(22)

(23)

18

4. Results

This chapter will discuss the results of this study. First the characteristics of the survey will be described. Second a choice based conjoint analysis is done on aggregate model. Third a choice based conjoint analysis is done based on segmenting the respondents whereby the hit rate of the model is predicted. This hit rate is compared with the hit rate of the aggregate model (4.3). Lastly, the effect of the moderators on the attributes is researched.

4.1 Characteristics of the survey

In total 229 people have completed the survey. At an early stage two people left the survey which ultimately results in 227 respondents filled in the questionnaire. A possible reason why these two respondents were leaving is that they were distracted and therefore not completed the survey. Another reason why respondents early left this survey are the conjoint questions, because they did not understand the questions, probably because the question is always the same but the alternatives are different, or found it too much reading. It should be noted that 32 respondents constantly chose the third option or no-choice option (a short description about the movie). That is the reason why these people are also excluded in this study, otherwise the analysis is not efficient anymore. Ultimately this has lead to go further with the final 195 respondents.

The average time of completing the survey was 5 minutes and 56 seconds. This is a respectable number, because people get the feeling that the survey is about something since they are not leaving immediately. On the other side the length of the survey was good indicating that there are only two respondents who left the survey at an early stage. The fastest respondent completed the questionnaire in 77 seconds. The slowest respondent completed the questionnaire in 1969 seconds. It was decided to include these respondents in the analysis, because they produced useful data. The covariates (age, gender, education, emotional mood, genre preference, and the ‘difficult’ choice of a movie) are well distributed. The average age of the respondents who have completed the survey was 31.1 years where the median is 25 years. The youngest respondent who completed the questionnaire is 12 years old. The oldest participant was 67 years. It is interesting to know that most of the respondents were between 23-25 years old. This is a good distribution, because most movies are watched by relatively young people (Cuadrado & Frasquet 1999). The gender of people is also good divided between man and woman. 53,33% (104) were men and 46,67% (91) women have completed the survey. Regarding the educational level, the conclusion can be drawn that most respondents are from HBO (40,79%) and WO (36,40%). 13,60% of the respondents have filled in the MBO level. 8,34% answered secondary school and 0,88% of the respondents filled in primary school. Figure 5 below shows an overview of the education level of the respondents.

(24)

19 When checking the emotional mood of people at the moment, it is remarkable that almost 90% of them are in a positive mood. One out of ten people is in a negative mood. The respondents who are in a positive mood are most of the time in a comfortable/relaxed mood (almost 60%). This mood is followed by energy/active and happy/enjoyment. Respondents who are in a negative mood are mainly anxious/worried (56,50%).

When looking at the favorite movie genre of people it can be seen that respondents mainly chose for four film genres. These are thriller (18,34%), comedy (17,03%), drama (15,28%), and action (13,10%). The genres romance, war, adventure, fantasy, horror, and science-fiction were between the 2%-10%. The question; ‘’to what extent is it difficult to choose a movie?’’, shows remarkable results. On the one hand, about half of the respondents (53.7%) said it is not difficult to choose a film. On the other hand, 46.3% of the respondents finds it difficult to choose a certain movie. Table 3 below shows a summary of the variables with their percentages.

Table 3: Summary of the variables

4.2 Choice-based conjoint analysis for the aggregate model

(25)

20 Whereby,

U = utility

4.2.1 Effects of the part-worth/nominal model

The first step to be done in predicting the aggregate model is to determine the type of model (linear, quadratic, or part-worth (see 3.5)). Latent Gold has the function to classify the variables into nominal (Part-worth) level or into numeric (linear) level. This study needs a part-worth model, because the attributes studio information, WOM, reviews, and awards are all nominal. Figures 6 and 7 and table 4 are showing part-worth models, since these figures do not show a straight line or quadratic form. The explanation is logical, because people have no clear preferences regarding the attributes studio information and WOM.It is for example imaginable that certain people prefer trailers over posters. Other people are more sensitive for posters than teasers etcetera. The same could be said over the attribute WOM. Social media like Facebook and Twitter can have the preference over the communication with friends and family, or IMDB is preferred over social media. In short, all scenarios are possible and therefore a part-worth model is the ideal case.

Figure 6: Utility studio information Figure 7: Utility WOM

LL npar Hit rate R²

Nominal -1483,4026 7 54,9% 0,0507

Table 4: Nominal model

The title of this chapter suggests the used model is an aggregate model. An aggregate model is based on one class. Therefore no segmentation is used. In addition, this model contains all 195 respondents. The R² of the nominal model is 0,0507. This indicates that the model is explaining 5,1% of total variation of the dependent choice variable. The hit rate of the model is also calculated. Hit rates are a form of predictive validity and calculates the number of percentage of observations the model hits. For the nominal model the hit rate is 54,9% ((472+384)/1560) which is good (appendix F).

4.2.2 Explanation aggregate model

(26)

21 a 5% level, except the attribute awards (P-value = 0,79). This indicates that an increase in the utility is associated with an higher preference for the corresponding level.

Attributes Utilities Wald-value Range Importance P-value

Studio information Poster Trailer Teaser -0,3421 0,2715 0,0706 63,31 0,61 50,8% 0,00 WOM Personal communication Social media IMDB 0,0535 -0,2617 0,2082 37,36 0,47 38,9% 0,00 Reviews Experts Non-experts -0,0533 0,0533 3,91 0,11 9,2% 0,048 Awards 0,07 0,015 1,2% 0,79 Yes No No-choice -0,0073 0,0073 -1,2851 267,39 1,2851 0,00

Table 5: Overview of the aggregate model

The most important attribute is studio information (50,8%). This illustrates that respondents prefer this attribute over all other attributes. The second most important attribute is WOM (38,9%) which expresses that respondents are sensible for the distribution of information by other people. Reviews are the third most important factor (9,2%). Lastly, the least important factor are awards (1,2%). Notable is the large preference for the attributes studio information and WOM. The attributes Reviews and Awards are not considered important enough by respondents.

When looking at the individual attributes it is relevant to interpret the utilities to figure out the preferences of the respondents. The attributes studio information exists out poster, trailer, and teaser. Respondents prefer trailers above teasers due to the higher utility. Posters are less preferred to the other levels since the utility of posters are negative (-0,3421). The second attribute WOM consists of the levels Personal communication, Social media, and IMDB. Consumers prefer the level IMDB followed by personal communication. The level social media is least preferred since this level has a negative utility. Next the attribute Reviews comprises the levels Experts and experts. Non-experts are preferred above Non-experts. The last attributes Awards exists out the levels Yes (information about awards) and No (no information about awards). No information about awards are preferred the most since the level ‘’yes’’ has a negative utility estimate.

(27)

22

4.3 Segmenting the respondents by CBC

As already said respondents can also be segmented which has the advantage that people can be divided into homogenous groups. The program Latent Gold can accomplish this by dividing people in latent classes or segments. The formula for the segmented model can be given as:

Uj = constant + β1 * j X1 + β2 * j X2 + β3 *j X3 + β4 * j X4 Whereby, U = utility X1 = Studio information X2 = WOM X3 = Reviews X4 = Awards j = segment 1,2 …..n

To perform this analysis by means of a CBC-analysis there is chosen for a maximum of five classes, because more classes will result in estimates that are inefficient in the sense of classes that are similar to each other. Furthermore it is difficult to divide 195 respondents into more than 5 classes, because of the size of the different classes. These classes will be researched by several information criteria. These are Log likelihood (LL), Bayesian information criteria (BIC), Akaike information criteria (AIC), and the Akaike information criteria 3 (AIC3). Differences between BIC, AIC and AIC3 is that BIC is preferred by large sample sizes. AIC and AIC3 are preferred for small sizes. A second difference is that BIC gets higher penalties for the complexity (i.e. more latent classes), while AIC, and AIC3 gets smaller penalties. Because the sample of this study is around 200, BIC is a good factor to interpret the model fit.

It is good to mention that there is chosen to describe only the covariates gender, age, emotional mood, and genre preference, because these covariates will show a good representation of this study. It could be possible to add the covariates education and difficulty of choice later to it. Latent Gold has the possibility to set the covariates as active or inactive. Firstly, all covariates (age, gender, genre preference, and emotional mood) are set to active needing to test if they are significant to the model. Setting all covariates as active will result in a 2-classes model. This is based on the lowest BIC and AIC3. A 3-class model is also possible due the lowest AIC. Table 6 below is showing the results of the 1-5 classes when all covariates are set to active.

Classes LL BIC(LL) AIC(LL) AIC3(LL) R²

1-class

2-classes 1466,6453 2671,2018

3-classes 2669,9042

4-classes 2689,5268

5-classes 2751,0468

(28)

23 The resulting 2-classes model is not perfect since a larger number of clusters can better describe between segments. However as suggested earlier a large number of classes will result in a less efficient estimation. By calculating the numbers of table 6 there is looked at significance for these covariates at all five classes. The 2-3 classes do not have any significant covariates. The 4-5 classes only have gender as a significant variable. Therefore, there is chosen to set all variables, except gender, as inactive. These inactive covariates are only used for describing the segments and are therefore not estimated for predicting class enrollment. An advantage of these inactive covariates is that the unweighted estimates are more stable.

Table 7 shows the 1-5 classes with their different information criteria after all covariates, except for gender, are set as inactive. This table suggests, based on the BIC, that there are 3 classes. Therefore, it has been decided to select ultimately 3 clusters, because this BIC is lower than the BIC of table 6. Another reason for continuing with 3 clusters instead of 2 is that 3 clusters have more differences between the classes.

The sizes of the segments are respectively 69%, 19%, and 12% which is indicating that segment 1 is by far the largest followed by segment 2 and segment 3. A good and interesting point to mention is to look at the R², LL, and Hit rate again. The R² of the model is now 0,2388, which is indicating that the model is explaining 23,9% of total variation of the dependent choice variable. The hit-rate is also increased from 54,9% to 64,8% ((467+390+154)/1560). An overview of these hit rates are given in appendices F and G.

Classes LL BIC(LL) AIC(LL) AIC3(LL) R²

1-class

2-classes 1466,6453 2626,1354

3-classes 2596,3327

4-classes 2577,6443

5-classes 2566,9468

Table 7: Comparison between the classes (inactive covariates except for gender)

After specifying the number of classes and calculating the LL, Hit rate, and R² it is useful to look at the importance of the attributes and their levels for the different classes. Table 8 is showing an overview of the different classes and their importance for the attributes and levels.

Importance (%) Class-1 Class-2 Class-3

Studio information 56,9 3,0 6,4 WOM 29,1 36,9 51,2 Reviews 13,5 46,8 39,0 Awards 0,44 13,3 3,4 Totaal Class size 100,0 69 100,0 19 100,0 12 Table 8: Overview of the importance by class and their class size

(29)

24 information. The third segment attaches the highest importance to WOM followed by reviews and studio information. Striking is the low preference for studio information and awards. Another interesting thing to look at is the importance of the levels per segment. Table 9 is showing an accurate describing of the segments. Segment 1 and 2 prefer the level trailer above poster and teaser whereas segment 3 has a preference for teasers. Segment 1 and 3 have a preference for IMDB regarding the attribute WOM. Segment 2 has a preference for personal communication. Reviews are the third attribute. Segment 1 and 2 desire non-experts and segment 3 desires the opinion of experts. The last attribute awards are playing a dominant role in the second segment. The 1st and 3rd segment do not have a preference for awards. Lastly, the no-choice option (the short description about the movie) is only chosen by people of segment 2. Persons of segment 1 and segment 3 negatively valued the no-choice option.

In the preceding section the preference for levels of the different attributes per class is mentioned. Now the segments will be exactly described (table 9). This will be done by using the betas of the different levels per attribute per class. Starting with the attribute studio information. Class-1 has the highest preference for trailers (40,6%), followed by teasers (7,4%). Class-1 response about posters is negatively valued indicating the negative sign (-0,4800). Concluded can be that this class has less preference for posters. Class-2 has the highest preference for trailers. This class has less desire for posters and teasers regarding their negative beta’s. The third class has the highest preference for teasers followed by posters. This class has less preference for trailers (utility = -0,1665). The second attribute WOM also shows some remarkable results. The most important level of class-1 is IMDB (0,18) followed by personal communication (0,10). This class has less preference for social media (utility = -0,2762). The most important level of class-2 is personal communication whereby this class has less preference for social media and IMDB regarding their negative utilities. Class-3 has the biggest preference for IMDB. Personal communication and social media are less desired by this class. The third attribute are reviews. This attributes exists out the levels experts and non-experts. Class 1 and 2 like the opinion of non-experts regarding this attribute whereas class 3 has a preference for the advice of experts. The last attribute are awards. Again class 1 and 2 agree on their preference for this attribute. These classes prefer awards above non-awards. Class 3 has a preference for no-awards meaning that they do not like information about awards when choosing a movie.

4.3.1 Covariates of the different classes

(30)

25

Attributes Class-1 Class-2 Class-3

Betas Betas Betas

Studio information Poster Trailer Teaser -0,4800 0,4058 0,0742 -0,0132 0,0345 -0,0213 0,0730 -0,1665 0,0936 WOM Personal Communication Social Media IMDB 0,0996 -0,2762 0,1766 0,3728 -0,3169 -0,0558 -0,8080 -0,4767 1,2847 Reviews Experts Non-Experts -0,1053 0,1053 -0,4383 0,4381 0,7969 -0,7969 Awards No Yes -0,0034 0,0034 -0,1247 0,1247 0,0691 -0,0691 No-Choice -3,6937 0,9905 -1,4474 Covariate Gender (p-value = 0,019) Man Female Beta -0,0931 0,0931 Beta -0,4269 0,4269 Beta 0,5200 -0,5200 Table 9: Accurate describing of the segments

Age The mean age of class-1 is 29,87 years old. This class is the youngest class of the three, because

class-2 has an average mean age of 35,03 and class 3 has an average mean age of 32,09 years old.

Gender is more dispersed over the three classes. The first class is almost identical divided between

men and women. 51,28% is men whereas 48,72% is women. The second class contains many more women than men. 65,1% of this class is female whereas 34,9% is men. The third class is showing the opposite. This class contains more men than women (78,5% over 21,5%).

Genre preference Some remarkable results in table 10 are bold. The genre comedy is the most

important in class-1 and class-2, whereas respondents in class-3 have a high preference for the genre drama. Class-2 is the class where the genres war and science-fiction are least preferable. On the other hand this class has the highest preference for the genre adventure. The genre fantasy is the least popular in class-3.

Emotional mood For this covariate there are also some remarkable results shown in bold (table 10).

(31)

26

Covariate % Class-1 Class-2 Class-3

Age Mean 29,87 35,03 32,09 Gender Men Women 0,5128 0,4872 0,3492 0,6508 0,7852 0,2148 Genre Comedy Action Drama Horror War Science-fiction Adventure Fantasy Thriller Romance 0,1905 0,1277 0,1418 0,0409 0,0577 0,0699 0,0567 0,0301 0,1797 0,1050 0,1529 0,1662 0,1108 0,0587 0,0002 0,0001 0,1886 0,0276 0,1396 0,1552 0,0948 0,1520 0,2610 0,0541 0,0519 0,0663 0,0964 0,0034 0,1594 0,0607 Emotional mood Energy/active Happy/enjoyment Comfortable/relaxed Anxious/worried Proud/self-complacent Regretful/guilty Sad/depressed Angry/evil 0,1782 0,1135 0,5727 0,0459 0,0433 0,0140 0,0174 0,0150 0,1393 0,0500 0,6507 0,0277 0,0771 0,0552 0,0001 0,0000 0,1325 0,1192 0,6242 0,0359 0,0546 0,0064 0,0254 0,0020

Table 10: Covariates of the different classes

4.4 Predicting validity of the hit rate on segmented level

In chapter 4.2.1 the hit rate of the nominal model is calculated. The hit rate of the segmented model is 64,8% (appendix G). This is an increase with respect to the hit rate of the aggregate model. Furthermore it is useful to look at the hit rate of the holdout sample, because this hit rate is showing the reliability of the model. The hit rate of the holdout sample is calculated by looking if the model predicts the right alternative of the holdout question. If this is the case, a hit is noticed. The conjoint questions consisted of nine questions where the holdout set was the fifth question in these nine questions. The external hit rate or hit rate of the holdout sample is 48,5% which is higher than the random choice of 33% and therefore good.

4.5 Effect of the moderators on the attributes

(32)

27 significant; Age*Trailer, Age*Teaser, Age*Reviews, Gender*Trailer, and Gender*Reviews. So, if age is going to increase with one year, than respondents are less into watching trailers. The opposite happens with teasers. If age is increasing with one year people are going to watch more teasers. The same holds for reviews. People find reviews more important if they are getting older. Looking at the attribute gender there are also some remarkable results. Women prefer to watch trailers over men. The reverse is true for reviews where women attach less value to reviews than men. To see if a covariate, like age, as a whole has a significant or non-significant effect on the four attributes there is chosen to measure this by means of comparing the BIC of the original model and the model with the moderator. If the BIC of the model with the moderator is lower than the BIC of the original model than the moderator plays a significant role. Table 11 shows the results (the bold numbers are for the moderator model whereas the other numbers are for the original model). Concluded can be that none of the moderators are significant, since all BIC for the moderator model are higher than the original model.

Covariates Original model Moderator model

Gender 1746,52 1774,45

Age 1746,52 1779,86

Genre preference 1746,52 1788,11

Emotional mood 1746,52 1785,35

(33)

28

5. Conclusion

This chapter will discuss the conclusion of this research. First, the purpose of the investigation will be discussed. After this the results will be discussed. Next, managerial suggestions will be made. Finally, constraints and propositions for the future will be given.

The aim of this study is to compose strategies to reduce customer uncertainty regarding the choice of the movie. It has been found that about half of the respondents have difficulties by choosing a suitable film. Therefore, this study will present appropriate strategies for all segments. This will be done both at aggregate and at a segmented level. Until now there is not much research done for movies, but this area will be more important in the future since considerable interest of people. Research regarding attributes is done, but the combination of the four attributes in this study is new. Ultimately, there should be looked if some homogeneous groups can be created. These groups can be placed under certain segments. Data of this research is collected by using an online survey where the respondents are found via social media, like Facebook, Twitter, and Linkedin and through asking family, friends, other students, etcetera. To analyze, a CBC analysis is used whereby the desires of respondents are researched. Finally, research to the moderator effects is done.

5.1 Results discussion

First, the model on aggregate level shall be discussed. This model contains all 195 respondents. The most important attribute is studio information in which trailers, introduction about the movie of 2,5 minutes, are the most relevant level. A reason for this is that trailers are the most specialized method of a movie promotion (Devlin et al 2011). Teasers, introduction about the movie of 60 seconds, are also important by the choice of respondents for a movie. Posters are less important. A possible reason for this is that most of the people prefer moving images instead of static pictures. Another reason is that moving images show more about a movie than posters do.

The second attribute which is valued most is WOM whereby the level IMDB is by far the most important between the respondents. This attribute is followed by personal communication, like the communication with family and friends. Social media, such as Facebook or Twitter, is least relevant. A remarkable result since social media is very important for customer decisions (Bronner & Hoog 2014). Nowadays millions of people all over the world are using Facebook and other social media to communicate with each other. Therefore it is good to do the same research again in the future to see how social media can influence this results. Interesting too is the size of the importance of the attributes studio information and WOM. These attributes have an importance of 50,8% respectively 38,9%. Concluded can be that most attention should be placed towards these attributes.

The third attribute are reviews. Non-experts, like family and friends are preferred whereas experts are less important. A possible reason for this is that respondents attach more value to friends and families because they are more trustful.

(34)

29 The no-choice option is not preferred. This implicates that the attributes are able to reduce uncertainty more than the text description of the movie.

Second, the segmented model should be analyzed. This model exists out three classes or segments whereby the segments are estimated by setting all covariates, except for gender, as inactive. The first segment is the biggest segment which contains 69% of the respondents. The second class contains 19% of the respondents followed by 12% of the respondents in class-3. Hereby it is interesting to see that the classes 1 and 2 prefer the level trailer of the attribute studio information in comparison with class-3 which is preferring teasers. A reason could be that men prefer teasers more than women. The second attribute WOM is also showing some differences between the classes. Class 1 and 3 prefer the level IMDB where class-2 chooses for the level personal communication. A possible explanation for this is that this is the segment with the oldest people in it. Older people prefer personal communication above using the internet. Another explanation could be that class-2 exists mainly out women. In general, women are using computers less than men do.

Class 1 and 2 prefer non-experts regarding the attribute reviews. Class-3, which consists mostly of men, desires the opinion of experts. An explanation for this is that women and younger people attach more value to persons who they know whereas men prefer people who are specialized in giving reviews.

Class-3 is also the segment which prefers awards above no-awards. Therefore is it possible that men are sensible for awards and have the tendency to call for the advice of experts. Striking is that the attribute awards is scoring very low in all three segments which indicates that this attribute is less important than the other attributes.

The importance of the different attributes over the three segments shows that every segment has a different preference of their most important attribute. Class-1 is preferring the attribute studio information. Class-2 desires the attribute reviews and class-3 has a preference for the attribute WOM. A reason why class-1 prefers studio information the most is that this segment is the youngest segment. Young people like to watch movies, television etcetera and are less sensitive for the opinion of someone else. Class-2, which mainly consists out women, is preferring reviews. Argumentation for this is that women rely more on other people like family, friends, and experts when they chose a movie. Class-3 is preferring WOM. This class is composed for the most part of men. An explanation for this can be that men are more on the internet (IMDB and Social media) than women. The only segment that prefers the no-choice option (a short description of the movie) is class-2. Therefore, segment 1 and 3 always try to chose between the two other options that were given.

(35)

30 Respondents in the classes 1 and 3 always try to chose an option that is other than the no-choice option, because of the higher negative numbers related to the aggregated model.

Lastly, the covariates are discussed. The first class has the lowest average age of all segments. The average age of this class is 29,87 years old. Men and women are equally dispersed and the favorite genre of this class is comedy. Fantasy is the least liked genre of this segment where most people in this class are in a comfortable/relaxed mood and least people are in a regretful/guilty mood.

The second class is the oldest segment. The average age of this class is 35,03 years old. A remarkable result is that there are many more women in this group. The favorite movie genre of these people is adventure. The least preferred genre is war. Also, in this segment most of the respondents are in a comfortable/relaxed mood.

The last class is the third segment. Respondents in this group have an average age of 32,09 years old. In contrast to class-2 this class contains more men than women. Drama is the most preferred movie genre of this respondents followed by thriller and action. In this segment the genre war is also least preferred. Again most of the people are in a comfortable/relaxed mood and fewest people are in a angry/evil mood.

A remarkable finding is that none of the covariates play a significant effect on the attributes, since the moderator models are less performing with regard to the original model.

5.2 Suggestions for managers

This study is interesting for studios. For them it is useful how different people react on the four attributes and what the best strategy is to reach this people. Thereby it is meaningful to see the contrasting homogeneous classes or segments. The study is showing that Dutch people find Trailers, IMDB, Non-experts and No information about awards the most important levels of the different attributes. Thereby it is nice to see that the attribute studio information attaches the greatest value. This is the attribute studios have control over. Therefore studios can come with a good strategy regarding this attribute. It is not easy to satisfy everybody by reducing their uncertainty. Therefore the descriptive attributes are also useful for explaining the differences between moviegoers, because they give a more detailed description from another view. Hereby, other strategies can be practiced. The attributes reviews and especially awards are less important, but it is good to consider them as some people have a big preference regarding these attributes.

A Choice-Based Conjoint Analysis for the effect of movie-related attributes on customer uncertainty

A Choice-Based Conjoint Analysis for the effect of

movie-related attributes on customer uncertainty

A Choice-Based Conjoint Analysis for the effect of

movie-related attributes on customer uncertainty

Management summary

Preface

Table of Contents

1.

Introduction

2.

Literature review

Studio controlled information

Non-studio controlled information (descriptive attributes)

3.

Research Design

4.

Results

5.

Conclusion