Are Lengthy Movie Trailers Worth Fighting For? A Conjoint Analysis on Consumer Preferences

(1)

Are Lengthy Movie Trailers Worth Fighting For?

A Conjoint Analysis on Consumer Preferences

Tim Houweling

(2)

Are Lengthy Movie Trailers Worth Fighting For?

A Conjoint Analysis on Consumer Preferences

Tim Houweling

University of Groningen

Faculty of Economics and Business

Master thesis MSc Marketing Intelligence

23th of June 2014 Oostersingel 108 9711 XH Groningen Tel: +316 34 88 28 55 E-mail: t.houweling@student.rug.nl Student number: 1894943 Supervisor: dr. F. Eggers

(3)

Management summary

Multiple researchers have attempted to build marketing models predicting box office success for a movie prior to its release in order to reduce performance uncertainties. However, the impact of theoretical research on practice has been rather limited. In an effort to expand the theoretical foundation of box office success predictions, this research addresses the literature gap of the effectiveness of movie trailers. The issue of theatrical movie trailer has grown highly relevant due to recent disagreements between the National Association of Theatre Owners (NATO) and movie studios regarding trailer duration. New guidelines created by the NATO limit maximum trailer duration to 120 seconds and prevent the display of movie trailers for movies premiering in more than 5 months.

Movie studios have responded negatively to these guidelines and are unwilling to comply. Using a choice-based conjoint analysis this research identifies a positive relationship between trailer duration and movie uncertainty. In turn, movie uncertainty is found to be negatively related to movie preference, therefore providing theoretical support for the position of movie studios. Aside from these main findings, this article shows that pre-trailer movie preference and star power do not have a moderating effect on either the relationship between trailer duration and movie uncertainty or the relationship between movie uncertainty and movie preference. Not only does this prove the existence of a direct effect of movie trailers on movie preference, it also demonstrates that increased trailer duration positively affects movie preference.

(4)

1. Introduction

The box office revenues for all movies released across the globe totaled $35.9 billion in 2013 alone (Theatrical Market Statistics, 2014). Despite the fact that economic interests are high, predicting box-office success for a particular motion picture has proven to be a difficult and challenging problem for researchers and movie studios alike. Multiple researchers have attempted to build marketing models estimating ticket sales prior to the release of a movie in order to reduce performance uncertainties (e.g. Sharda and Delen, 2006; Sharda et al., 2007; Eliashberg & Shugan, 1997). However, the impact of this research on practice has not been as significant as in other industries (Eliashberg et al., 2006). Surprisingly, the motion picture industry is characterized by a relatively high economic importance and a richness of data available for research. As a result, the field of research predicting consumer preference among a wide range of movie choices provides an interesting area for improvement.

Recently, a discussion has emerged regarding the duration of cinematic movie trailers (Cooper, 2014). The Los Angeles Times reported that the National Association of Theatre Owners (NATO) is dissatisfied with the long duration of movie trailers preceding movies in their cinemas (Verrier, 2014). The NATO argues that long trailers have been a source of complaints among moviegoers, as they usually follow an already large set of paid advertisements. Verrier argues that “although studios and theater owners view trailers as a key way to market upcoming movies, exhibitors have grown increasingly concerned that long promotional spots consume valuable advertising space, reveal too many plot lines and can be ineffective if they are screened too many months ahead of the movie's release date.” Additionally, the NATO claims overly lengthy trailers are less effective than shorter versions of these trailers, particularly when the promoted movies do not premiere in the next five months. New guidelines were suggested by the NATO limiting movie trailers to a maximum of two minutes, appearing no more than five months before the movie’s premiere. The NATO claims these guidelines would improve the effectiveness and efficiency of movie marketing (Rothman, 2014).

(7)

for paid advertisements by limiting the maximum duration of movie trailers. As a result, movie theaters could reap greater benefits from paid advertising deals (Fritz, 2014). This would imply that NATO’s argumentation that trailer duration should be shortened due to consumer complaints regarding long advertisements is not their true cause of concern. Furthermore, studio executives contend that their biggest features would be more difficult to promote under the new guidelines. Movie trailers for these features often start running a year in advance of the movie’s premiere, thus conflicting with the new guidelines as suggested by the NATO. To overcome these issues, movie studios aim to ignore the guidelines and negotiate with theater owners individually in order to maintain their current line of practice (Fritz, 2014).

The heated arguments confirm that neither party is willing to budge. Their discussion highlights a knowledge gap which this article will address by attempting to establish a connection between trailer duration and movie preference. The lack of relevant findings regarding movie trailers is made undeniably clear by Elberse & Anand (2007), who investigate the effect of pre-release advertising for movies and in their research do not consider the theatrical trailer. The aim of this research is therefore to investigate the movie trailer and answer the following research question: does movie trailer duration have an

effect on movie uncertainty and/or movie preference of moviegoers?

1.1 Introduction to the movie industry

(8)

As mentioned during the introduction the motion picture industry is characterized by a richness of data available for research. The BoxOffice (2014) collects a vast amount of this data including ticket revenues in the United States, as shown in figure 1. The upward slope indicates continuously increasing box office ticket revenues. In figure 2, the amount of theater visits follows a progressively downward slope after its initial peak in 2002. The downward slope indicates that the increase in ticket revenues is due to rising ticket prices while the amount of tickets sold, and thus overall movie attendance, has actually decreased.

This suggestion is confirmed by the Motion Picture Association of America (MPAA) that provides an overview of the average ticket price per movie (Theatrical Market Statistics, 2014). Figure 3 shows the development of the average ticket price over time as well as the development of the ticket price following a consumer price index (CPI). The CPI resembles the increase in prices similar to the inflation rate (Theatrical Market Statistics, 2014).

(9)

movie to lose approximately $17 million at the box-ofﬁce, excluding DVD rentals and sales (Elberse and Anand, 2007). This is counterintuitive to the initial observation of increasing box office ticket revenues. The average loss can be explained by the rising amount of annual movie releases. Table 1 presents the number of movies released each year, which gradually increases (Theatrical Market Statistics, 2014). The growth in ticket revenues is therefore distributed between an expanding amount of movies, decreasing the average box office performance.

In conclusion, an increasing number of movies compete for a diminishing amount of moviegoers. This proves that competition for each customer increases across movies. As a result, each individual customer is becoming more valuable to the movie studio. Hence, the movie studios should aim to improve their marketing efforts to meet their potential customers’ individual preferences and demands. Currently, as mentioned by Zufryden (1996), movie studios are often clueless regarding the effectiveness of their marketing efforts. This research measures the effect of movie trailers, and in particular trailer duration, on movie preference.

1.2 Research on box office performance

(10)

new movie productions. They demonstrate that star power has a positive influence on short-term performance. However, a movie’s long-short-term success depends on the subsequent critical reviews of consumers resulting in either positive or negative word-of-mouth.

Elberse & Anand (2007) find a positive link between pre-release advertising and box office performance. Surprisingly, the possible influence of movie trailers was neglected in this research, as it solely focused on pre-release advertising in the form of other marketing activities. As a result the influence of movie trailers has remained unattended. Furthermore, Zufryden states that “the film diffusion pattern is typically characterized by a peak in box-office receipts at the time of initial film release which is then followed by a pattern of exponential decay over time” (p. 30, 1996). The movie trailer, which precedes initial film release, could potentially influence this peak.

(11)

2. Theoretical framework

The main goal of this research is to identify the effect of trailer duration on movie uncertainty, and consequently the effect of movie uncertainty on movie preference. These relationships strengthen the argument of either the movie studios or the NATO, regarding the new guidelines on trailer duration. Furthermore, this research provides a theoretical background for future trailer development. Hence, this section identifies the trailer attributes under consideration for the modelling process.

2.1 Trailer duration

Despite the disagreement between movie studios and the NATO, the effect of trailer duration on movie uncertainty and/or consumer movie preferences has not yet been studied. Nonetheless, some research has investigated the relationship between the length of an advertisement and commercial recall. Newell & Henderson (1998) state that the length of a promotional message significantly and positively affects brand recall. Preceding research by Singh & Cole (1993) demonstrates that prolonged commercial duration increases consumers’ learning of the product. Patzer (1991) found similar results claiming that 30 second-commercials were more effective than 15 second-commercials. The articles all agree that prolonged commercial duration positively influences consumer learning of a product. In this research, consumer learning can be seen as collecting movie information reducing the movie uncertainty faced by the consumer.

Movie uncertainty describes the uncertainty a consumer faces when considering to watch a movie. Consumers experience this uncertainty because the outcome of a purchase decision can only be known at some point in the future (Dowling & Staeling, 1994). The consumer may reduce this uncertainty by attaining additional information, or by watching a cinematic trailer. When the trailer runs for 0 seconds, there is no effect on movie uncertainty. Following this logic, showing the full movie will completely remove movie uncertainty and maximize consumer learning, resulting in the first hypothesis:

H1: Trailer duration is negatively related to movie uncertainty.

(12)

commercial message length is non-monotonically related (follows an inverted ‘U’ shape) to the effectiveness of the communication. For movie trailers, this would imply that the effectiveness of trailer duration increases until it reaches a certain peak. The marginal decrease in movie uncertainty is expected to decline with every additional minute following this initial peak. The influence of trailer duration on movie uncertainty is therefore expected to follow a nonlinear trend:

H2: The influence of trailer duration on movie uncertainty follows a nonlinear trend.

An often seen alternative to the theatrical trailer is the TV trailer, also known as a ‘teaser’ trailer. These trailers usually contain similar scenes as their lengthy counterparts, but mainly differ in length. The TV trailers are released by movie studios to air during commercial breaks, running for approximately 30 seconds.

Until now the conceptual model has focused solely on movie uncertainty. In order to predict how consumers will respond to movie trailers, these findings need to be related to the movie preference of consumers. Identifying what causes movie preference is useful in the establishment of improved movie trailers which are more likely to attract moviegoers.

2.2 Movie preference

According to Murray (1991) consumers will search for product information to reduce the perceived risk and uncertainty related to the service industry. Product evaluation can only take place after the purchase of a movie ticket, and as a source of information movie trailers can influence consumers in the decision making process. The Hollywood Reporter states that ‘half of moviegoers (49%) think trailers give away all the best scenes’ (Ford, 2013), following a self-conducted study among American moviegoers. Their study finds support for trailers as the biggest influence on consumers’ movie uncertainty. This supports the earlier finding that trailer duration is negatively related to movie uncertainty. In addition, this finding would prioritize movie trailers over other sources of information in the search for product information mentioned by Murray (1991).

(13)

pre-release advertisements on box office revenues. Their research states that advertisements draw consumers who would otherwise have opted for different movies.

Since movie trailers are part of pre-release advertisements for a movie, it is assumed that movie trailers contribute to the positive effect on box office revenues found by Elberse & Anand (2007). The information the consumer acquires when watching a movie trailer is hypothesized to be (partially) responsible for the increase in movie preference. This leads to the following hypothesis:

H3: Movie uncertainty is negatively related to movie preference.

The initial finding by Ford (2013) was that movie trailers are the biggest influence on consumers’ movie uncertainty. Subsequently, it was revealed that not all participants displayed the same reaction after watching a movie trailer for ‘Iron Man 3’. 19% of the participants were unwilling to watch the movie after they witnessed the trailer, while 24% of participants were more excited to watch the full movie. No information is provided regarding Ford’s selection of participants, and thus no information is available regarding pre-trailer preference of these participants. Initial prejudice regarding the particular movie for which the trailer was shown could explain why 19% of the participants were found unwilling to watch the entire movie. Similarly, initial preference for a movie could explain why 24% of the participants were more enthusiastic after watching the same movie trailer. This would be in line with the research of Muthukrishnan & Kardes (2001) who find that initial product preference influences product perception. Hence, pre-trailer movie preference could help explain some of these findings and is included as a moderating variable for movie preference estimation:

H4: Pre-trailer movie preference moderates the effect of movie uncertainty

on movie preference.

2.3 Star power

(14)

power, are likely to make consumers expect a high-quality movie. As mentioned in the introduction, Hennig-Thurau, Walsh & Wruck (2001) found that star power of actors is a significant influencer of short-term success after the movie’s premiere. The effect diminishes in importance when word-of-mouth and movie critics become available to consumers, which are considered a more credible source of information regarding movie quality. This indicates that star power can determine the level of uncertainty moviegoers possess when considering a new movie. Seeing that this could intervene with the effect of trailer duration on movie uncertainty the star power variable is added as a moderating variable. This leads to the following hypothesis:

H5a: The star power of actors that appear in a movie trailer moderates the effect

of trailer duration on movie uncertainty.

Furthermore, Bagella & Becchetti (1999) found strong support for actors and directors as influencers of overall box office success. This is in line with the hypothesis that star power reduces the movie uncertainty faced by consumers regarding a new movie. Additionally, they found that adding interaction effects using both variables (actors and directors) neglected the effect of all other predictor variables in their research. In order to control for a potentially moderating effect of star power on the effect of movie uncertainty on movie preference for this research, the following hypothesis is added to the conceptual model:

H5b: The star power of actors that appear in a movie trailer moderates the effect

of movie uncertainty on movie preference.

2.4 Movie genre

(15)

cannot predict any preferences stated by participants, since its value is identical across alternatives. Nonetheless, the potential differences between trailer duration preference across genres needs to be addressed. In order to analyze whether significant differences exist the one-sided ANOVA test statistic is selected (Malhotra, 2010). Movie genre is therefore not included in the conceptual model.

2.5 Conceptual model

The final conceptual model resulting from the hypotheses is presented in figure 4.

2.6 Omitted variable bias

Various alternative influences are present in a consumer’s choice process. Hennig-Thurau et al. demonstrate that movies based on existing intellectual property have greater certainty in the market over movies based on original screenplays as they have pre-existing consumer awareness (as quoted by Gazley et al., 2011). These movies will be excluded from this research to avoid a potential bias in favor of, or detrimental to, the concept the movie originated from.

(16)

3. Methodology

3.1 Method

As the aim of this research is to identify consumer preferences towards differences in movie trailer duration, participants are presented with movie trailers varying in length. Participants are asked to choose the trailer deemed most informative and preferred. Therefore, the dependent variable is a choice made by participants. In order to observe their selection of the most informative movie trailer and the most preferred movie, the conjoint analysis method is selected. Preceding the choice sets pre-trailer movie preference is measured on a 1-7 Likert scale, based on star actors and a plot description.

Following the pre-trailer movie preference measurements participants are presented with two choice sets per genre. Each choice set consists of two alternative movie trailers, of which the most informative movie trailer and the preferred movie is selected. The method is therefore choice-based, with a small number of attributes as presented in the conceptual model. The choice-based conjoint (CBC) analysis fits this description best. Additionally, Wierenga (2006) argues that preference structure measurement techniques should be used in research regarding the motion picture industry.

As mentioned by Eggers & Sattler (2011), a ‘no-choice’ option should be integrated into the research design for each choice set whenever price sensitivity or willingness-to-pay (WTP) is measured. This research does not aim to identify WTP or price sensitivity and intends to prevent the loss of information by leaving out the no-choice option.

3.2 Model specification

(17)

( 1 )

where

: the systematic utility component, or the ‘rational utility’. : the stochastic utility component, or the error term.

It is assumed that each movie trailer is a combination of trailer attributes; the consumers attaches a part-worth utility to each of these attributes; and the systematic utility ( ) is the sum of part-worth utilities given by:

( 2 )

where

: number of attributes.

: a dummy indicating the specific attribute level of movie trailer . : the part-worth utility of consumer for attribute .

The choice model assumes that participants select the alternative for which

( 3 )

Where A and B represent different movie trailers. The probability of selecting a movie trailer from choice set is calculated by means of a multinomial logit model:

( 4 )

where

: number of alternatives in the choice set. For this research . : the choice set being evaluated.

(18)

3.3 Choice design

3.3.1 Movie selection

For each movie genre, 4 movies were selected based on two key characteristics. First, the selected movie is not yet released in the cinema in order to prevent the influence of movie critics and word-of-mouth (section 2.1.4). Second, the movie is not based on pre-existing intellectual property, for example books or other movies, in order to prevent a potential bias in favor of, or detrimental to, the concept the movie originated from (section 2.1.4).

The movie trailers were selected for the three most popular movie genres measured by their box office revenue. These genres are Comedy (54,88mln), Adventure (42,33mln) and Drama (38,94mln), as can be seen in figure 5 measuring U.S. box office revenues (Statista, 2013). As the fourth most popular movie genre, the action genre closely follows drama (37,57mln).

However, there are multiple movies labeled both as ‘action’ and ‘adventure’ (IMDb). The trailers for these movies are therefore used to represent the action/adventure genre. In conclusion, based on these conditions a total of 12 movies, equally distributed across the 3 genres, were selected for this research (Appendix, table 23).

(19)

movies are never compared across genres. This prevents the potentially moderating effect of movie genre preference on movie selection.

3.3.2 Trailer duration

In order to perform the CBC analysis, two choices are presented in each choice set. For every choice set, one trailer is substantially longer than its counterpart. The originally developed cinematic movie trailers run for nearly two and a half minutes, and are henceforth referred to as lengthy trailers throughout this research. The lengthy trailer is currently the industry standard for cinematic trailers. The second group of trailers, referred to as the intermediate

trailer, consists of the original trailers edited to last roughly one minute and forty seconds.

These trailers were either readily available or were edited for the purpose of this research. The third type of trailer is commonly known as a teaser trailer, running for considerably less than a minute. These types of trailers are most commonly seen on television during commercial breaks. As was the case with intermediate trailers, the teaser trailers were either readily available or consist of a lengthy trailer edited for the purpose of this research. An overview of the trailer duration variable is provided in table 2.

Additionally, the trailers within a choice set should never have the same duration. For example, if the trailers for the movies ‘Belle’ and ‘Begin Again’ within 1 choice set are both intermediate trailers, the effect of trailer duration on movie choice could not be analyzed. Therefore, each participant is presented with 2 choice sets per genre, consisting of trailers of different duration based on random selection.

3.3.3 Star power

(20)

popular on the list. The IMDb rating is based on consumer ratings, where higher ranked actors are more likely to attract an audience while lower ranked actors are less likely to attract an audience (section 2.1.3). A continuous variable is developed based on the ranking of the actors from the movies used in the survey in order to measure their star power. The movie for which the star actor ranks highest on IMDb’s list was appointed the 12th rank in the star power variable, as 12 movies were selected for this research. The lowest ranked actor was labeled as a 1. This data is collected separately from the survey adding secondary data to the model estimation procedure (Malhotra, 2010).

3.3.4 Pre-trailer movie preference

A complication during trailer evaluations is that participants are likely to choose the movie they would like to see regardless of the movie trailer. As a result movie selection does not necessarily result from the movie trailer shown during the research. Therefore, any results can easily be said to be biased. In order to overcome this problem, a short description of each movie´s plot and star actors are provided preceding the trailer. Participants are asked to rate the movie based on the given descriptions on a 1-7 Likert scale measuring pre-trailer movie preference, as mentioned in the method section (section 3.1).

3.4 Survey exhaustiveness

While developing choice sets, each trailer is assigned to a certain movie. Each movie only appears once during each survey. There are 4 movies for the 3 genres, adding up to 12 movies or a maximum of 6 choice sets available per participant. Each choice set includes approximately 3 minutes of movie trailers. Therefore, under the assumption that participants will watch every trailer in full, participants face over 20 minutes of survey questions.

(21)

always belong to the same genre, eliminating the genre preference effect in individual choice sets.

3.5 Survey sample

The choice sets for this research are embedded in an online survey which was distributed by means of direct mailing and sharing on social media websites (Facebook and LinkedIn). Furthermore, by encouraging participants to spread the questionnaire among their network snowball sampling was used. This can result in a nonprobability sample because referrals have demographic and psychographic characteristics more similar to the people referring them than would occur by chance (Malhotra, 2010).

Before commencing the data analysis the data is explored in order to observe inconsistencies and remove corrupt data lines. Due to the nature of this research a minimum response time is established to ensure that the participants included in the analysis have watched the trailers. The data reveals that 18 of the 116 participants completed the survey within 400 seconds, which would not be sufficient to watch even the shortest trailers in full. Therefore, these lines of data are removed from the dataset, reducing the number of participants to 98.

(22)

4. Results

The main issue this research addresses is the influence of trailer duration on movie uncertainty and movie preference. The research can be labeled as causal, where the major objective is to obtain evidence regarding cause-and-effect relationships (Malhotra, 2010).

4.1 Descriptive statistics

The response set of 98 participants consists of 59,2% male and 40,8% female participants. The average age of the participants is 27, of whom 85% have attended or completed higher level education. Therefore, the response set is slightly biased toward educated moviegoers.

As a measure of consumer involvement, their frequency of theater visits is provided in table 3. Over half of the participants (52%) visit the cinema once every 3 months, while little over a quarter of the participants (26%) visit the cinema once a year. Only 4% of the participants declared that they never visit the cinema. Therefore, most of the participants are familiar with the movie theater and are expected to have sufficient movie-going experience to complete the survey.

As was shown in section 3.1.1, comedy is considered to be the most popular movie genre. For this research, 2 out of the 3 genres were selected by each participant. Comedy and action/adventure were each selected in 77% of the cases, while drama trails behind with 42% (table 4). This follows a similar distribution of genre preference as presented in figure 3.

4.2 Balance check

(23)

which is the expected average value. The lengthy trailers only appear in 28,8% of the choice sets, nearly 5% lower than the expected average value of 33,3%. The choice design is therefore not optimally balanced.

A possible explanation for this is that participants who were confronted with lengthy trailers faced increased fatigue levels, and consequently may have decided not to complete the survey. This presents a minor limitation to this research due to the fact that lengthy movie trailers are underrepresented in the collected data.

4.3 Movie uncertainty

The first dummy-coded variable for which the part-worth utilities are estimated is the ‘Most Informative’ variable (MostInfo), which indicates whether or not the trailer was selected as being the most informative movie trailer alternative within a given choice set. Consequently, the most informative movie trailer is the trailer which reduces movie uncertainty the most. Therefore, positive effects on the MostInfo variable are negatively related to the participant’s movie uncertainty.

Based on the conceptual model (section 2.5) trailer duration and star power are selected as the independent variables. The star power variable is included as an interaction variable moderating the effect of trailer duration on movie uncertainty. Therefore, an interaction variable was created as a function of the trailer duration ( ) and star power ( ) variables following TD * SP.

To determine the contributions of these predictor variables and their attribute levels in the determination of movie uncertainty, the data on each individual choice set is analyzed using LatentGold. Inserting the variables into equation 2 (section 3.2) for the MostInfo DV, the equation becomes:

(24)

where

: the rational utility component for selecting a movie trailer as ‘most informative’.

: the effect of trailer duration on .

: the effect of star power on .

: the interaction effect of trailer duration and star power on .

4.2.1 Trailer duration

In order to estimate the systematic utility for movie uncertainty the -estimates need to be calculated for each level of the attributes. As can be seen in table 6, the parameter estimates for trailer duration are highly significant ( ), indicating a significant relationship between the duration of a trailer and the probability a trailer is chosen to be most informative. The -value of -0,6612 for the teaser trailer is the part-worth utility assigned to the ‘teaser’ level for the attribute ‘trailer duration’. The teaser trailer is thus estimated to reduce the rational utility level for selecting a movie trailer as being most informative by 0,6612. The -values for intermediate (0,2373) and lengthy (0,4239) trailers display a positive relationship between trailer duration and the ‘MostInfo’ dummy variable.

As mentioned before, the most informative movie trailer is the trailer which reduces movie uncertainty the most. Hence, the trailer which offers the most information is negatively related to movie uncertainty. As a result, H1 stating that trailer duration is negatively related

(25)

4.2.2 Model linearity

As can be seen in table 2 (section 3.3.3) the average duration of a teaser trailer is 39,26 seconds, the average duration of the intermediate trailer is 97,74 seconds and the average duration of the lengthy trailer is 146,47 seconds. The linearity of the effect of trailer duration on movie uncertainty is tested by adding a new variable to the dataset. This variable contains the length of each individual movie trailer in seconds. Using this new variable as the single independent variable, instead of the trailer duration dummy, the model is re-estimated. In order to test H2 stating that the relationship between trailer duration and

movie uncertainty follows a nonlinear trend, the new linear model is compared to the earlier nonlinear model excluding the star power variable in order to isolate the effect of trailer duration on movie uncertainty.

First, a graph is plotted showing the -values for each attribute level of trailer duration in seconds using the dummy variable for trailer duration (figure 6).

The increase in the rational utility level is larger between teaser and intermediate trailers than the increase in the rational utility level between intermediate and lengthy trailers. As hypothesized by H2, the marginal increase in the -values for trailer duration appears to

(26)

Whereas figure 6 presents a visual representation of the relationship, additional statistical tests are required to determine the linearity of the model. As shown in table 7, shifting from a nonlinear model to a linear model decreases the model’s -value. The linear model has an of 0,2162, while the nominal model has an of 0,2185. Similar to the LL-statistic, the -statistic increases with every variable added to the model as it follows . The , with as the amount of parameters, punishes for the addition of insignificant variables and therefore provides a better estimation of the model fit. The variation of 0,0023, which represents a 0,23% increase in the variance explanation of the DV, does not indicate a large difference in model performance based on .

In addition to the -value the BIC, AIC3 and CAIC values complicate model evaluation even further. The BIC and CAIC provide favorable outcome for the linear model, while the AIC3 score indicates a better model fit for the nonlinear model (table 7). The best model is usually selected based on the largest adjusted and the lowest BIC, AIC3 and CAIC values. However, the adjusted is more commonly used in the evaluation of linear models while BIC, AIC3 and CAIC are more fit to evaluate both linear and nonlinear models (Blattberg, Kim & Neslin, 2008).

As a result, H2 stating that the relationship between trailer duration and movie uncertainty

(27)

4.2.2 Star power interaction

The moderating effect of star power was observed by adding the interaction term to the model. As could be seen in table 6, the inclusion of both the star power variable and the interaction variable are insignificant ( ). This effect is re-estimated for the linear model, adding star power as an interaction term with the trailer duration in seconds.

Table 8 shows that the star power variable has no significant effect on the selection of the most informative movie trailer ( ). Additionally, no significant interaction is found between the effect trailer duration has on the selection of the most informative movie trailer and the moderating impact star power may have on this particular effect ( ). Both the star power and the interaction variable have no significant effect and are removed from the model, thus finding no support for hypothesis H5A.

4.2.3 Model estimation

Following the exclusion of the star power variable from the model, a new model is estimated using trailer duration as the single predictor variable of the most informative movie trailer. As a result, equation 5 is reduced to:

( 6 )

(28)

The probability that a participant finds a movie trailer lasting 60 seconds more informative than a movie trailer lasting 55 seconds is calculated by inserting these values into equation 4 (section 3,2):

( 7 ) The odds of selecting the 60 second trailer instead of the 55 second trailer is calculated by dividing the probability value by 0,5, since there are 2 possible alternatives in each choice set: . An increase of 5 seconds in trailer duration improves the odds of a movie trailer being selected as most informative by approximately 4,3% up to the maximum trailer duration in this research, which is 157 seconds.

4.2.4 Model fit

Both the first model, including trailer duration and the interaction variable for star power, and the second model, using only trailer duration as a predictor variable, are analyzed in terms of model fit. Table 10 lists several of model fit indicators for both models.

(29)

The BIC, AIC3 and CAIC value all support the second model, where trailer duration is used as the only independent variable (Hair et al., 2013). The model fit comparison therefore does not provide any arguments to deviate from the previously selected model, which is model 2.

In order to further analyze the predictive power of the model the overall hit rate is observed. The prediction table (table 11) provides an overall hit rate of 75,8%, with a prediction error of . This is an increase in hit rate of = 51,6% compared to the null model. The model therefore increases the correct choice prediction of the DV by 51,6% compared to the null model.

4.3 Movie preference

In addition to selecting the most informative movie trailer, participants selected their preferred movie after watching the trailers. As can be seen in the conceptual model, the movie preference dummy is hypothesized to depend on movie uncertainty (measured by the most informative movie trailer dummy ) with a potentially moderating influence of star power ( and/or pre-trailer movie preference ( ). Furthermore, trailer duration is included as a linear variable in order to test whether it has any effect on movie preference in addition to its effect on movie uncertainty. Inserting these variables in equation 2 (section 3.2) provides the following equation:

( 8 )

where

: the rational utility component for selecting a movie as the preferred movie.

: the effect the most informative trailer dummy has on .

(30)

: the effect of star power on .

: the effect of pre-trailer preference on .

: the interaction effect of the MostInfo dummy and star power on .

: the interaction effect of the MostInfo dummy and pre-trailer preference on .

The model and its attributes will be tested in an order similar to the analysis of movie uncertainty.

4.3.1 Most informative movie trailer

(31)

4.3.2 Interaction variables

As was tested for the effect of trailer duration on movie uncertainty, any influencing effect may be (partially) explained by additional variables. Following the parameter estimates in table 12 the effect of movie uncertainty on movie preference is not explained through an interaction effect ( ). This finding holds for both the star power and the pre-trailer movie preference variable. Therefore, both interaction hypotheses H4 and H5B are not

supported in this research and are removed from the model.

Despite having no interaction effect, the pre-trailer movie preference variable is highly significant at . In addition, the star power variable closely resembles movie uncertainty when comparing their p-values (0,12 compared to 0,16) and Wald statistics (2,4107 compared to 1,9837). These statistics indicate that if the effect of movie uncertainty on movie preference is significant following the removal of the interaction terms, the effect of the star power variable on movie preference in itself could be significant as well.

In order to check whether the effect of movie uncertainty and star power on movie preference is significant, these variables are included in a model without the interaction variables. As can be seen in table 13 the dummy indicating the most informative movie trailer is significant at , while star power is significant at .

(32)

movie trailer demonstrates that movie uncertainty is negatively related to movie preference. Hence, the model finds support for H3, which is accepted.

The participants’ pre-trailer movie preference is based partially on the star actors of a movie, which in turn determine the star power variable. Therefore, interaction effects can occur between the star power and the pre-trailer movie preference variables. To measure this effect an interaction variable was created for SP * PTP, which was found to be insignificant with (table 14). All three variables are therefore included in the final model, and in the following sections model estimation and model fit will be discussed.

4.3.3 Model estimation

Unlike the prediction of movie uncertainty, the movie preference variable is predicted by more than one independent variable. The exclusion of the trailer duration variable, as well as the exclusion of the interaction terms for star power and pre-trailer movie preference, reduces equation 8 to:

( 9 )

Table 13 provides the -values for each variable, so that equation 9 can be rewritten as:

( 10 )

where

(33)

Therefore, the rational utility increases by 0,6894 when a trailer is selected to be most informative. The effect is significant at .

As an example, consider two movies with identical values for star power (6) and movie rating (5). For these values, and . These values are inserted into equation 4 (section 3.2) in order to obtain the probabilities for the preferred movie choice with the only difference between the two movie alternatives being the movie uncertainty variable, as presented in equation 11:

( 11 ) which provides . The probability of a movie trailer being selected as the preferred movie, without it being selected as most informative, is therefore given that all other variables remain constant.

(34)

The movie uncertainty variable is found to be more important than the star power variable in predicting movie preference. However, the pre-trailer movie preference of participants is relatively the most important predictor of (post-trailer) movie preference. This finding indicates that movie preference largely depends on the participant’s pre-trailer movie preference, and less on the subsequently shown movie trailer.

4.3.4 Model fit

The model used to predict the preferred movie choices provides an overall hit rate of 71,2%, as demonstrated by table 16. The prediction error of indicates that although movie uncertainty, star power and pre-trailer movie preference predict a significant part of movie preference, a lot of variance remains unexplained. In comparison to the null model the predictive accuracy of the current model is improved by .

As was done in the model comparison part of section 5.1.5, the and the are provided in order to analyze the percentage of variance explained by the model. Table 17 shows that the model explains 18,27% of the variance in the movie preference variable. The

indicates that due to an adjustment for the amount of parameters, this percentage is

(35)

4.3.5 Movie genre

As mentioned in section 2.4 the results of this research could differ between genres. Due to the research design there is no variance in movie preference that is explained by genre, but the possibility remains that for a particular genre increased trailer duration is more/less preferred than in other genres. The average trailer duration for the preferred movie trailers in each genre are presented in table 19.

These differences appear to be small, and in order to test this a one-way ANOVA was performed for the preferred movies in each genre. By selecting only the preferred movies, the preferred trailer duration can be compared between genres. The results of the ANOVA are presented in table 20, demonstrating that there is no significant difference and the findings can therefore be generalized across the 3 movie genres.

4.3.6 Trailer duration in seconds

(36)

(37)

5. Discussion

This research sought an answer to the question whether movie trailer duration has an effect on movie uncertainty and/or movie preference of moviegoers. The application of the choice-based conjoint analysis revealed some interesting findings regarding these relationships. First of all, trailer duration has a significant negative effect on movie uncertainty. Hence, increasing trailer duration decreases the movie uncertainty a consumer faces. In turn, the movie uncertainty faced by consumers negatively influences movie preference. In other words, if a movie trailer is evaluated to be more informative that movie is preferred over its less informative counterpart. Therefore, trailer duration influences movie preference through its influence on movie uncertainty.

In addition to the effect of trailer duration on movie uncertainty and movie preference, this research also supports the overall effectiveness of movie trailers as an advertising tool. Even though pre-trailer movie preference is established as the most important attribute in the modelling process predicting movie preference, it is shown that the movie trailer has a significant effect on movie preference as well. These findings hold for all 3 genres discussed in this research. This broadens the scope of box office performance predictors.

Recalling the discussion which highlighted the knowledge gap, the movie studios’ point of view is supported. This research proves that trailer duration is found to increase movie preference of consumers. This effect is explained via the negative effect of trailer duration on movie uncertainty, which is a significant predictor of movie preference. Therefore, the argument provided by movie studios that shorter movie trailers are not sufficient to persuade consumers to visit their movies receives theoretical support from this research. As a result, the movie studios have every right to be upset about the NATO´s guidelines limiting the maximum trailer duration to 120 seconds.

(38)

5.1 Star power

The star power variable measured by ‘actor rank’ did not interact with either the effect of trailer duration on movie uncertainty or the effect of movie uncertainty on movie preference. In the latter case, star power did have a significant effect on movie preference for p < 0,05 (table 13). The influence of star power on movie preference is in line with the findings of several prior research papers (Sochay, 1994; Gazley et al., 2011; Hennig-Thurau et al., 2001; Bagella & Becchetti, 1999). The findings seem to contradict earlier work of Desai & Basuroy (2005). They state that for more familiar movie genres the impact of star power on box office performance decreases. Following their findings, the selection of movies in this research based on the three most popular movie genres should decrease the effect of star power. However, the star power variable is significant in the prediction of movie preference.

5.2 Pre-trailer movie preference

(39)

6. Limitations & Recommendations for future research

This section discusses several limitations to the paper as well as recommendations for future research. The first limitation is related to the fact that participants were asked to select 2 out of 3 genres which they preferred. Gazley et al. (2011) identified genre to have a significant influence on movie preference. The assumption that both of these genres are equally preferred cannot be made, and therefore genre preference is not included in the prediction of box office success. Future research should aim to identify differences in trailer preference among participants preferring different movie genres, apart from the difference in trailer duration preference measured in this research.

Secondly, movies were excluded from the research based on existing intellectual property in order to isolate the effect of trailer duration on movie preference. Hence, the effect of existing intellectual property is not measured, and the trailers included in this research are compared to other movies which have no existing intellectual property. This is not a realistic representation of the marketplace. The effect of trailer duration could decrease when a movie is compared to movies based on already existing intellectual property. Future research should focus on differences in advertisement strategies for movies based on existing intellectual property and movies that are not based on existing intellectual property.

Thirdly, this research does not measure whether participants would visit the movie they have selected. The choices merely represent the preferred movie which was selected from 2 provided alternatives per choice set. Subsequently, the focus on movie preference did not provide results based on willingness-to-pay. Therefore, no conclusions are drawn on the behavioral consequences of watching a movie trailer in monetary terms. The research does establish a link between movie trailers and movie preference, and future research should focus on monetizing the added value these movie trailers have for moviegoers by comparing the willingness-to-pay for different movies. One of the attributes should then be whether or not a movie trailer is shown for the movies in a given choice set.

(40)

7. References

Bagella, M., & Becchetti, L., (1999). The determinants of motion picture box office performance: evidence from movies produced in Italy. Journal of Cultural Economics, 23 (4), pp. 237-256. Blattberg, R. C., Kim, B. D., & Neslin, S. A. (2008). Database Marketing: Analyzing and Managing

Customers. Springer New York.

Caves, R. E. (2000). Creative industries: Contracts between art and commerce. Harvard University Press.

Chang, B.H., & Ki, E.J., (2005). Devising a practical model for predicting theatrical movie success: focusing on the experience good property. Journal of Media Economics, 18 (4), pp. 247-269. Cooper, G.F., (2014). Theater owners want shorter, later movie trailers. NBC News, [online] January 27. Available at: <http://www.nbcnews.com/pop-culture/movies/theater-owners-want- shorter-later-movie-trailers-f2D11999565> [Accessed 25 May 2014].

Desai, K.K., & Basuroy, S., (2005). Interactive Influence of Genre Familiarity, Star Power, and Critics’ Reviews in the Cultural Goods Industry: The Case of Motion Pictures. Psychology &

Marketing, 22 (3), pp. 203-223.

Dowling, G.R., & Staelin, R., (1994). A Model of Perceived Risk and Intended Risk-Handling Activity. Journal of Consumer Research, 21 (1), pp. 119-134.

Eggers, F., & Sattler, H., (2011). Preference Measurement with Conjoint Analysis: Overview of State- of-the-Art Approaches and Recent Developments. International Journal of Research in

Marketing, 26(2), pp. 108-118.

Elberse, A., & Anand, B., (2007). The effectiveness of pre-release advertising for motion pictures: an empirical investigation using a simulated market. Information Economics and Policy, 19, pp. 319-343.

Eliashberg, J., & Shugan, S. M. (1997). Film critics: Influencers or predictors? Journal of Marketing, 61 (April), pp. 68-78.

Eliashberg, J., Elberse, A., & Leenders, M.A.A.M., (2006). The Motion Picture Industry: Critical Issues in Practice, Current Research, and New Research Directions. Marketing Science, 25 (6), pp. 638-661.

Ford, R. (2013). Study: Half of Moviegoers Think Trailers Give Away All the Best Scenes. The

Hollywood Reporter [online] 5 January. Available at:

<http://www.hollywoodreporter.com/news/study-moviegoers-think-trailers-give-450336> [Accessed 30 January 2014].

Fritz, B., (2014). Movie Theaters, Studios Get in a Tiff Over Trailers. Wall Street Journal, 28 Jan. Gazley, A., Clark, G., & Sinha, A., (2011). Understanding preferences for motion pictures. Journal of

(41)

Hair, J.F., Black, W.C., Babin, B.J., & Anderson, R.E., (2013). Multivariate Data Analysis. Pearson New International Edition.

Hennig-Thurau, T., Walsh, G., & Wruck, O., (2001). An investigation into the factors determining the success of service innovations: the case of motion pictures. Academy of Marketing Science

Review, 6, pp. 1-23.

Johnston, K.M., (2008). ‘The Coolest Way to Watch Movie Trailers in the World’. The International

Journal of Research into New Media Technologies, 14 (2), pp. 145-160.

Malhotra, N. K. (2010). Marketing research: An applied orientation. New Jersey: Pearson Education. Murray, K.B., (1991). A Test of Services Marketing Theory: Consumer Information Acquisition

Activities, 55 (1), pp. 10-25.

Muthukrishnan, A.V., & Kardes, F.R., (2001). Persistent Preferences for Product Attributes: The Effects of the Initial Choice Context and Uninformative Experience. Journal of Consumer

Research, 28 (1), pp. 89-104.

Newell, S.J., & Henderson, K.V., (1998). Super Bowl advertising: field testing the importance of advertisement frequency, length and placement on recall. Journal of Marketing

Communications, 4 (4), pp. 237-248.

Patzer, G.I., (1991). Multiple dimensions of performance for 30-second and 15-second commercials. Journal of Advertising Research, 31, pp. 18-25.

Rothman, L. (2014). Movie Trailers Will Get Shorter, But Won’t Become Interactive Anytime Soon. Time, [online] 27 January. Available at: <http://entertainment.time.com/2014/01/27/nato-

movie-trailers-change/> [Accessed 25 May 2014].

Sharda, R., & Delen, D., (2006). Predicting box-office success of motion pictures with neural networks. Expert Systems with Applications, 30 (2), pp. 243–254.

Sharda, R., & Delen, D., (2007). Movie forecast Guru: A Web-based DSS for Hollywood managers. Decision Support Systems, 43 (4), pp. 1151-1170.

Singh, S.N., & Cole, C.A., (1993). The effects of length, content and repetition on television commercial effectiveness. Journal of Marketing Research, 30, pp. 91-104.

Sochay, S., (1994). Predicting the Performance of Motion Pictures. Journal of Media Economics, 7 (4), pp. 1-20.

Statista, (2013). Most popular movie genres in North America by total box office revenue from 1995

to 2013. [online] Available at: <http://www.statista.com/statistics/188658/movie-genres-in-

(42)

The BoxOffice, (2014). Yearly Domestic Gross. The BoxOffice, [online] Available at: <http://www.boxoffice.com/statistics/yearly> [Accessed 5 February 2014].

Verrier, R. (2014). Movie theater owners call for shorter trailers. Los Angeles Times, [online] 28 January. Available at: <http://www.latimes.com/entertainment/envelope/cotown/la-et-ct- shorter-movie-trailers-20140128-story.html> [Accessed 25 May 2014].

Wierenga, B., (2006). Motion pictures: consumers, channels, and intuition. Marketing Science, 25 (6), pp. 674-678.

(43)

(44)

(45)

(46)

(47)

(48)

(49)

(50)

(51)

(52)

(53)

(54)

(55)

(56)

Are Lengthy Movie Trailers Worth Fighting For? A Conjoint Analysis on Consumer Preferences

Are Lengthy Movie Trailers Worth Fighting For?

A Conjoint Analysis on Consumer Preferences

Tim Houweling

Are Lengthy Movie Trailers Worth Fighting For?

A Conjoint Analysis on Consumer Preferences

Management summary

Table of Contents

1. Introduction

2. Theoretical framework

3. Methodology

4. Results

5. Discussion

6. Limitations & Recommendations for future research

7. References