What are the antecedents of change and what are the
consequences of change on the link between past sales
and future sales
Student: Maikel Lust Student Number: 10001560 Date: 26th June 2015
Education: MSC Business Administration
This document is written by Student Maikel Lust who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
ABSTRACT: This study examines the relationship between the performance of the present edition of a series and the performance of it’s predecessor and the
moderating role of change within a series. Also the antecedents of change are examined. To find all the relations, a quantitative study is performed based on archival data within the creative industry, more precisely the games industry. The sample consists of 1478 observations of games in the period 2000-2012 which are part of a series. This study finds, that there is a positive relation between the performance of the present edition and the performance of it’s predecessor. The moderator change can weaken this relation. Further this research found that there are not significantly more changes when the sales of the predecessor are low. This study also doesn’t found support for a relation between changes and competition. This study found support that there is a negative relation between changes and the number of editions in a series. This is the opposite of what I was expecting.
Table of contents
1. Introduction ... 5
2. Literature review ... 7
2.1. Sequels... 7
2.2.1. The prospect theory ... 8
2.2.2. Signaling theory... 10
2.2.3. The irrational behavior theory ... 10
2.2.4. Consumer socialization theory ... 11
2.3. Endogeneity ... 12
2.4. The antecedents of changes ... 14
2.5. Conceptual framework ... 15 3. Research design ... 16 3.1 Sample ... 17 3.2. Research models ... 18 4. Results ... 22 4.1. Descriptive statistics ... 22 4.2. Analyses ... 26 4.3. Robustness check ... 30
5. Discussion and conclusion ... 32
1. Introduction
Nowadays series are well common in almost every industry. In the phone industry there are series like the Iphone, and the Samsung Galaxy and in the software
industry there is for instance Windows. In the creative industry there are sequels like the Hunger Games, and Grand Theft Auto and in the car industry there are car series like the Renault Espace. These series are pretty successful, for example in the creative industry, in 2011, 9 of the 10 top-grossing films worldwide were sequels (Kurtzleben, June 26, 2012). The major reason of the success of these series is that when customers have good experiences with the original in the series, they expect to have good experiences with the successors as well (Völckner & Sattler, 2006).
In the article of Situmeang et al. (Situmeang, Leenders, & Wijnberg, 2014), the
aim of the authors was to test a model that explores whether there is a connection between the reviews of both consumers and experts of the original and their reviews of the successors. Their second aim was to explore if a lack of consensus in the
reviews of the past editions have an effect on the reviews of the future editions. To explore these two aims, the researchers have analyzed a dataset of 577 video game releases in the period 2000-2009. They found that there is a connection between the reviews of the past editions and the reviews of the future editions. They also found that this connection is lower when there is a lack in consensus in the reviews of the past editions. Since reviews have a strong signaling function (Connelly, Certo,
Ireland, & Reutzel, 2011a), they found that there is a strong positive relation between past sales and future sales. Possible reasons for this positive relation are that the reviews of the new editions may include some expectations that are based on the original edition. This implies that in the case that the reviews and sales of the original edition were well, customers expect that the new editions are of high quality to, which results in high sales (e.g., (Keller, 1993); (Hennig-Thurau, Houston, & Heitjans, 2009).
There are several theories that might explain the positive relation between past sales and future sales. These theories are the prospect theory, the signaling theory, the irrational behavior theory of consumers, and the consumer socialization theory. These all will be explained in the literature review. The aim of these
paragraphs is to better understand the relation which is the basis of this article.
However the relation between past sales and future sales is not desirable when
there’s an unsuccessful product, because this relation indicates that when the past sales are bad, the future sales will be low to. The goal of an organization is to maximize profit. Sales are an important variable of profit on which managers and companies are judged on by their shareholders. It is clear that in this case, low sales, managers and companies have to do something. They have to change something to improve the revenues in the future.
This study tries to extend the earlier study of (Situmeang et al., 2014) by
extending the provided model. (Situmeang et al., 2014) found that there is a positive relation between past sales and future sales within series. This present study tries to examine if changes in the product attributes could weaken this relation. Therefore I would like to examine the following research question: what drives the antecedents of change and what are the consequences of change on the link between past and future performance?
This study contributes prior literature by responding to a call by Situmeang et
al. (2014) for improving their model. I will improve their model by adding a moderator in the relationship, namely changes.
This study contributes to practice at several ways. In the first way it
contributes to managers by examining if changes can help when their company is suffering low sales at the moment. For example when the sales of Windows Vista are low, the company should not just continue with doing the same for the new version of Windows but it should try to change something to make sure that there is no positive relation between the sales of Windows Vista and Windows 7.
This study also contributes by examining what these changes drives. When managers know in advance that changes are needed, they can quickly respond to it.
The research is structured as follows. In chapter 2 a literature review is
provided. In this chapter, different theories will be described with the aim to better understand the findings of this thesis. At the end of chapter 2, the different
hypotheses are developed which will be tested in this research. Also the conceptual framework is shown. In chapter 3, the research method is explained, of which the results are presented in chapter 4. In the last chapter of this thesis, there is a general discussion of the results is provided on which a conclusion will be based. This is presented together with the limitations and suggestions for further research.
2. Literature review
In this chapter different theories will be addressed. The aim of this chapter is to better understand the positive relation between past sales and future sales, and also the findings of this article. During this chapter, several hypotheses will be addressed which will be examined in this thesis.
2.1. Sequels
When a company wants to increase it’s revenues it can use different ways: it can attract new customers, it can create new ways of using it’s existing products and it can charge higher prices (Taylor, 2003). A firm can create a new product to attract new customers as part of a brand extension (the use of an established name to enter a new product category) (Aaker & Keller, 1990). However new products call for big investments and managing effort, for instance the introduction of a new brand in the consumer market has an estimated cost range of $ 50 million to more than $ 100 million. A less expensive way to increase growth is to use sequels.
A Sequel is, according to the Oxford Dictionary (2015), a published,
broadcasted, or recorded work that continuous the story or develops the theme of an earlier one. Or simply a similar kind coming one after another. Series making is a successful strategy implemented by a lot of companies in almost every industry: the creative industry, the software industry, the car industry, and so on. In the past decade the revenues related to sequels in the movie industry have more than
doubled, from $ 718 million in the 1990s to $ 1.9 billion annually in the 2000s (Sood & Drèze, 2006).
There are different theories describing the success of series. The most important theories are described in the next paragraphs: the prospect theory, the signaling theory; the irrational behavior theory; and the consumer socialization theory.
2.2.1. The prospect theory
As described in the introduction, one of the main theories that describes and explains the positive relation between past sales and future sales is the prospect theory. The prospect theory is created in 1979 by Daniel Kahneman (Kahneman & Tversky, 1979). The prospect theory was created as a critique on the present expected utility theory, which was the dominant theory in 1979. Expected utility theory was a descriptive theory used to describe the economical behavior of people. The main message of the expected utility theory was that the decisions of people are based on the different statistical expectations of valuations. The highest value of the expectations is the preferred option (Rabin, 2000). However when people are risk averse, they will just take the option in which no risk is involved. However, according to Kahneman, this theory can’t be applied in every section to describe individual choices. Therefore he has created the prospect theory which is further developed in 1992 (Tversky & Kahneman, 1992). In this thesis, the focus will be on this updated prospect theory. In literature there is referred to this updated theory as being the cumulative prospect theory.
The prospect theory can be distinguished in two phases: the framing phase, and the valuation phase. In the framing phase, the decision maker is trying to make a clear picture of the different options. Everything that is relevant for the final decision is set in a certain heuristic. They frame every option by using a reference point, an anchor. Outcomes which are higher as the reference point are seen as a gain and outcomes which are lower as the reference point are seen as losses. This is different as the expected utility theory under which no reference point is used. The downside effects of losses are seen to have a bigger impact as the positive effects of gains. Figure 1 represents this framing phase.
Figure 1: the framing of decisions
Source: (Tversky & Kahneman, 1986)
In the valuation phase, the decision maker is calculating the value of the
different options and choose ultimately the option with the highest expected
outcome. In contrast to the expected utility theory, risk aversion and risk seeking are determined together by the value function and the capacities of the different options (Tversky & Kahneman, 1992).
As mentioned earlier, this prospect theory can be applied to the positive relation between past sales and future sales. When people are satisfied with the ‘original product’ in a series, they set this ‘original product’ as a reference point on which their further decisions are based. For series it is relatively easy to respond to this reference point, because there are a lot of similarities between the original and
the successor. Also the risk related to this ‘follow up product’ in seen as lower, because the decision maker is already common with the product. Therefore the probability that the decision maker will choose to buy the ‘follow up product’ is high.
To conclude, the prospect theory basically means that consumers set a
reference point on which their further decisions are based on. In the case of series this implies that when customers have positive experiences with the Iphone 4, they made a reference point of this. When they now faces the Iphone 5 they still have these positive feelings and they will buy it. This basically means that they have bought the Iphone 5 based on their experiences with the Iphone 4.
2.2.2. Signaling theory
Another theory that explains the positive relation between past and future
performance within a sequel is the signaling theory, which is related to the prospect theory. The signaling theory is useful when two parties have access to different information (Connelly, Certo, Ireland, & Reutzel, 2011b). Which basically means that different parties know different things. Signaling can help to reduce this information asymmetry. In the case of sequels this means that the consumer doesn’t know in advance what his or her experience will be with a certain product. However his or her experience with the previous edition in the series has a signaling effect. A good experience give him or her the idea that the experience with the new edition will be good as well. Therefore, signaling can help the consumer to reduce their risk of making a bad decision. Consumers are risk averse.
Overall, the experience of the customer with the first edition signals what kind of experience the customer will have will the second edition which explains the
positive relationship with the performance of those two.
2.2.3. The irrational behavior theory
A third theory that describes the success of series is the irrational behavior of consumers. Consumers are risk averse which have an influence on their decision
making (Bao, Zhou, & Su, 2003). This risk aversion basically means that when consumers face a bunch of options, they will choose the option with the lowest risk. The option with the lowest risk for them is choosing for the series. A series can be seen as a small brand, for example the game series Grand Theft Auto can be seen as a sub brand of Rockstar Games. The most important advantages that consumers link towards a brand are risk reduction, saving of search costs, and social and emotional expressive values (Fischer, Völckner, & Sattler, 2010). With risk reduction is meant that a brand can reduce the risk of buying a mistake by buying from a brand the customer already knows and has good experiences with. This applies for series as well, in the case that when consumers have good experiences with the Iphone 4, they will buy the Iphone 5 too, rather than choosing for the Samsung Galaxy S5, because they already know that the Iphone is fulfilling their needs. With saving of search costs is meant that when consumers know that the brand is good, they don’t have to look for another brand. When consumers know that the Iphone 4 is good, they don’t have to look for other phones that fulfill their needs by looking to all the attributes the different telephone providers deliver. When choosing for the familiar brand, they save these costs. The social and emotional expressive values is that consumers buy the Iphone 5, because their friends and family buys it too. This perceived advantage of brands will be more explained in the next paragraph in which the consumer socialization theory is explained.
2.2.4. Consumer socialization theory
Another theory that describes the relation between past sales and future sales is the consumer socialization theory. Due to the popularity of the original edition in their micro-environmental environment, people are very excited and they want to be just like their peers, resulting that they want to buy the second edition. They buy this to get favorable reaction from them (Brown, Clasen, & Eicher, 1986).
According to theory it can be expected that there is a positive relation between
the performance of earlier editions in a sequel and the future edition. This is
empirical tested and supported by Situmeang et al (2014). However there isn’t fully consensus about this in scientific literature. For instance Basuroy and Chatterjee
(2007) found that there is a negative relationship. They found that sequels perform worse than the parent film in terms of revenues. This could be explained by Sood and Drèze (2006). They say that consumers prefer dissimilarity over similarity and desire something new, especially when goods are experiential and intangible in nature. The reason is that experiential attributes such as the story line and genre tend to satiate such that consumers prefer something different in the sequel. Satiation is bad and lead to lower sequel evaluations.
Hypothesis 1: Performance of past editions in the series have a positive relationship with performance of the new editions.
2.3. Endogeneity
A positive relation between past and future performance indicates that there is a problem when the performance related to the original edition was bad, because this means that the performance of the new edition will be bad as well. This is a relevant case, because the number of disappointments in series is much larger as the number of successes (Sood & Drèze, 2006). Sequels like Spiderman 2 were quite successful, whereas sequels like Miss Congeniality 2 and Windows Vista were a
disappointment. Therefore it’s important for a company to change something in the series to influence the relation.
As explained in the previous paragraph, Sood and Drèze (2006) explain that people desire dissimilation over simulation. Consumers may be more attracted to series that include a new genre relative to a series that simply continuous the previous theme. The reason behind this is that consumers desire something new. This will be more explained in paragraph 2.4.
Several changes are possible for a company to implement in their series, some of these changes will be tested in this thesis. One of the possible changes is the name of the sequel. The significant effect of name on the evaluation of brand extensions are widely tested in the academic literature. The study of Sood and Keller (Sood &
of the consumer about a product which is part of a brand extension. The reason behind this difference is that consumers process this information differently in their mind. They found that family branded extensions, like Tropicana Cola, have more relation which each other in the mind of the consumer than subbranded extensions like Quencher by Tropicana Cola. A study of Sood and Drèze (2006) has also looked to the influence of name on brand extensions. They found that numbered sequels, like Daradevil 2 are more influenced by similarity than named sequels (Daredevil: Taking it to the Streets). They explained that when perceived similarity is high, extensions are assimilated with the parent brand and affect is transferred from the parent brand to the extension. With perceived similarity they meant that there is a feature overlap. When the feature overlap is high then the categorization is similar and evaluations improve compared to when feature overlap is low and the extension category is dissimilar. Perceived similarity can also be defined in terms of intangible attributes and brand-specific associations unrelated to the category, which is the case with the change in name (Sood and Drèze, 2006).
Based on this reasoning is the second possible change, namely change of genre. When the genre of the sequel change, feature overlap is low compared to the case within genre isn’t changed. This indicates that affect is less transferred from the original to the successor.
The last indicator of change that will be tested in this thesis is the difference in review score between the past and the present in the series, which is more the
outcome of the reasoning of Sood and Drèze (2006). When feature overlap is high, affect is transferred from the parent brand to the extension. This indicates that consumers will evaluate both versions almost equal, with the same review score. Affect is less transferred when feature overlap is low. Therefore I assume, that there will be a difference in the review score of the original and the present edition after a change has occurred.
Hypothesis 2: Changes have a negative effect on the positive relationship between the performance of past editions in the series with performance of the new editions.
Hypothesis 3: There is a negative relationship between the performance of past editions in the series and changes.
2.4. The antecedents of changes
There are several reasons possible why the sequels are less successful as the originals. Two of these possible reasons will be tested in this thesis.
One possible reason is competition in the market. Competition is hurting revenues in almost every industry. Competition is described by Adam Smith as the following: competition will immediately begin among buyers, when the supply is excessive. The price will sink more, the greater the competition of the sellers, or according as it happens to be more or less important to them to get immediately rid of the commodity. Competition is here (and usually) used in the sense of rivalry in a race, a race to get limited supplies or a race to be rid of excess supplies. Competition is a process of responding to a new force and a method of reaching a new
equilibrium (Stigler, 1957). Perfect market competition means that rivals will
continue entering the market as long as extraordinary gains are achieved. At the long run a market can therefore not be profitable. However, the aim of firms is to
maximize their profits. From this it can be concluded that in the case of game developers, that they will move to other markets (genres) when rivalry increases.
The more supply there is, the more consumers can choose from. Recent research shows that the market for games is stabilizing. (SOURCE).
Competition leads to lower revenues and profits due to increased rivalry. Therefore, competition can be a reason for change.
Hypothesis 4: There is a positive relationship between competition and changes.
Another possible reason I would like to examine is aging. Series can’t exist forever. At the moment, there are no game series which are now available, which have their existence in the beginning of the console gaming. A reason for this fact, might be that consumers are getting bored with a certain series and are searching for
new experiences. Psychological theory suggests that people are always looking for new experiences. That they always desire something new. There are three motives for the desire for something new. First, consumers want it because they see
something new as the fresh or newly created, Second, they see the new as the improved or innovative. Last, they see the new as the unfamiliar or the novel. In society, this phenomenon can be seen in the price setting of new things. New things often demand a higher price than similar product which are already in the market, although they perform the same function (Campbell, Hirsch, & Silverstone, 1992).
This can be applied to the case of series. People basically chose to buy the next edition of a series to reduce their risk of making a bad buying decision. However on the long run, their desire for something new will come greater. Sood and Drèze (2006) found this effect in experiential goods, namely the movie industry. They found that after a time, consumers prefer to experience something new. They become
satiate.
Therefore I assume that there is a positive relation between the number of series and change. The longer a series exists, the bigger the chance that consumers decide to buy a different game. Game developers can counterattack this by adding a chance to their series. By this, they can give the consumers the idea that it’s
something new.
Hypothesis 5: There is a positive relationship between age in series and changes.
2.5. Conceptual framework
An overview of the different hypotheses of this thesis is covered in the model below. The arrows show the direction of the relationship. The plus symbol in the boxes implies a positive relationship and the minus symbol implies a negative relationship.
3. Research design
This chapter will outline the research method used in this thesis. To examine my research question and mine different hypotheses I will conduct a quantitative database analyses. The main reason why I’m conducting a quantitative research is the generalizability. The outcomes of quantitative research can be applied in many settings while the outcomes of qualitative studies can mostly only be applied in the settings in where the research is conducted (Maanen, 1983). This makes it able for me to generalize the outcomes of my findings to more than the creative industry, namely to series in every product category. To examine my different hypotheses I will use two kinds of tests. For hypothesis three, I will use a probability test to test whether the probability of changes in series is greater in the case that there are weak
performances in the past edition. For all my other hypotheses I will use a regression analysis. The first paragraph of this chapter will describe the sample of the thesis and the applied databases. The second paragraph will explain the different variables that
will be used in the models. The last paragraph will describes the models which are used to test the hypotheses.
3.1 Sample
The sample that will be used in this study is the creative industry, more specifically the game industry and will include a time horizon from 2000 till 2012. The data is archival secondary data. The dataset covers games from the three major games consoles: Microsoft Xbox, Nintendo Wii, and Sony Playstation. There won’t be any kind of geographical or size constraints. Using the creative industry as sample in my study is in line with earlier studies (Situmeang et al., 2014). It’s also a very suitable setting in the sense that the use of series is very popular in the creative industry (Tschang & Szczypula, 2006).
I will merge two different databases to capture my research data:
metacritic.com and vgchartz.com. To capture sales data I will take use of the online database vgchartz.com. I will use metacritic.com for the other data, like reviews. These databases have also been used by other studies towards the relationship between past sales and future sales (Situmeang et al., 2014).
The sample started with 4980 unique observations. First, the observations were deleted if they are not part of a series. 877 observations were deleted and 4103 remained. Furthermore the observations which have missing sales data were deleted together with the other observations with missing data. This resulted in a final
sample of 1478 observations which will be used to conduct the examination of the research models. The average amount of editions in a series are 12. A distribution of the sample based on its genre is presented in the figure below. Most of the games are sport games and action/adventure games.
games. The metacritic.com database will also be used to know whether there are changes within a product series. The same sales figures and review scores will be used to test the second hypothesis which can also be applied for the third hypothesis. To examine my fourth hypothesis, I will include the variable market share in my regression analysis. The higher the market share, the more competition there is in the market (Boone, 2008). For my fifth hypothesis I will use the database metacritics.com again. I will measure age by looking to the time between the introduction date of the first edition and the introduction date of the last edition.
3.2. Research models
In this paragraph an overview is provided of the measures, variables and models which are used in this thesis. The variables described in paragraph 3.2. are applied in these models.
In order to test hypothesis 1, I will use a regression model. The performance of the games will be measured by sales. Using sales as a performance measure for games is used by more researchers in scientific research (Situmeang et al., 2014). I will capture the sales figures from the database vgchartz.com. The sales figures show the number of legal copies sold, this makes sense because the number of second hand sales and illegal copies is a small proportion of the legal sold games. The
performance of the past edition will be measured using the same way.
Sports 35% Shoorters 13% Strategy 4% Action/adve nture 21% Other 27%
SAMPLE DISTRIBUTION
The sales of games are not only influenced by the sales of the previous edition. Therefore I will include some control variables in the model. One of the control variables I will implement in the model is the review score. Review score is one of the most important measurable determinant of sales. A study of Deloitte (2007) find that 82% of the respondents make their purchase decisions based on user reviews. Another study showed that 62% of the consumers read reviews (Lightspeed
Research, 2011). Merging these two studies together implies that more than half of all the consumers make their purchase decisions directly based on product reviews. The product reviews will be measured in this thesis by two variables, namely meta
review score and user review score. Both scores are derived from the database metacritic.com. The meta review scores are the opinions of experts (professionals) in the game market. The user review scores are the opinions of other consumers who give their opinion based on their experience with the game.
I will also control for games in the sport genre which represents 35% of the total sample. These games might disrupt the relationship, because consumers will buy the next edition of these games anyways, a change of the game is therefore not required. For instance, the difference between Fifa 14 and Fifa 15 is not big, however people will buy Fifa 15 although they have Fifa 14 already. This is the case, because the most recent edition is better reflecting the ‘real world’. The dummy variable SPORTS equals 1 when the game belongs to the sport genre and 0 otherwise.
This results in the following regression model for hypothesis 1:
SALES = α + β1 PREV + β2 M.SCORE + β3 U.SCORE + β4 SPORTS + ε Where:
SALES = the sales of the present edition measured by the number of legal sold copies worldwide.
PREV = the sales of the previous edition measured by the number of legal sold copies worldwide.
M.SCORE = the meta review score. This is the average review score given to the games by experts.
U.SCORE = the user review score. This is the average review score given to the games by the consumers.
SPORTS = this dummy variable equals 1 when the game belongs to the sport genre and 0 otherwise.
Hypothesis 2 tests the moderating role of change in the model of hypothesis 1. Change will be measured, as discussed in paragraph 2.3., by three measures. These three measures are name change, genre change, and review score change. A change has occurred when at least one of these measures identifies a change.
I will use the fuzzy matching score to identify whether a change of the name has occurred or not. Fuzzy matching is an advanced mathematical process that determines the similarities between data sets. A fuzzy matching score of 100% means that the name of observation a is exactly the same as the name of observation b. A name change has occurred when there is a fuzzy matching score between in the range 0,25-0,75.
I will use the same method to measure whether a change of the genre has occurred. A fuzzy matching score lower as 100% identifies a change.
The third measure of change, review score change, is measured by looking to the average review score of the present edition and the average review score of the edition before. A change has occurred, when the difference of the average review score is higher as 1.
Including the moderator change in the model results in the following modified regression model:
SALES = α + β1 PREV + β2 CHANGE + β3 PREVCHANGE + β4 M.SCORE + β5 U.SCORE + β6 SPORTS + ε
Where:
SALES = the sales of the present edition measured by the number of legal sold copies worldwide.
PREV = the sales of the previous edition measured by the number of legal sold copies worldwide.
CHANGE = the dummy variable CHANGE equals 1 when at least two of the three measures of change equals 1 and 0 otherwise.
PREVCHANGE = the moderating variable. This variable is calculated by multiplying the variable PREV with the variable CHANGE.
M.SCORE = the meta review score. This is the average review score given to the games by experts.
games by the consumers.
SPORTS = the dummy variable SPORTS equals 1 when the game belongs to the sport genre and 0 otherwise.
To test hypothesis 3, I will divide all the observations in two groups based on the dummy variable DIF. This dummy variable equals 1 when the sales of the previous edition are higher as the sales of the present edition and 0 otherwise. A student t-test will test whether the difference in the coefficient of the variable change is higher for the lower sales group as it is for the higher sales group.
In order to test hypothesis 4 and 5, I will use a regression model which is different from those of hypothesis 1 and 2. Hypothesis 4 and 5 tests what determines change. The dependent variables in this model are competition and time.
Competition is determined by the number of games of the same genre. Time will be measured by the number of different editions that are of the same series.
I will include the same control variables as under hypothesis 1 and 2 with the same kind of reasoning. I will also include the control variable PREV with the same reasoning as hypothesis 3. Low sales indicate that something has to change.
The regression model I will use for hypothesis 4 and 5 is the following:
CHANGE = α + β1 COMP + β2 TIME + β3 PREV + β4 M.SCORE + β5 U.SCORE + β6 SPORTS + ε
Where:
CHANGE = the dummy variable CHANGE equals 1 when at least two of the three measures of change equals 1 and 0 otherwise.
PREV = the sales of the previous edition measured by the number of legal sold copies worldwide.
COMP = the competition variable. This calculate the number of games of the same genre that are sold worldwide.
TIME = the variable TIME represents the number of edition in a series.
M.SCORE = the meta review score. This is the average review score given to the games by experts.
U.SCORE = the user review score. This is the average review score given to the games by the consumers.
SPORTS = the dummy variable SPORTS equals 1 when the game belongs to the sport genre and 0 otherwise.
4. Results
In this chapter the results of my thesis are covered and analyzed. The first paragraph exists out descriptive statistics which provide information about the sample and an examination on the data of the sample. The second paragraph will provide an examination of the three research models by performing a regression analysis and a probability analysis related with hypothesis 3. The third paragraph will provide a robustness check.
4.1. Descriptive statistics
After creating all the different variables, the descriptive statistics, skewness, kurtosis and normality tests were performed to test whether the different variables are
normally distributed.
The mean of the variable sales is 1,770, while the mean of the variable
prevsales is only 1,578. This indicates that in general the sales of the earlier editions in the series are lower as the sales of the next editions. These variables are also highly correlated with each other (0,505). When the prevsales are high, the sales will be, in general, high as well. This is already basically a support for my first hypothesis.
Another interesting finding is that the mean of user review score is higher as the mean of the meta review score (7,217 compared to 7,094). This implies that in general, consumers are more satisfied with the games as the experts. A reasons might be that experts are reviewing games more critically. It could also be that consumers are more willing to give a review when they are satisfied compared to if they are not. This makes the consumer reviews less reliable.
Table 1 shows a Pearson Correlation Matrix for all the applied variables. The coefficient shows how much the different variables are related with each other. The coefficients ranges between -1 and +1. A correlation of +1 implies that as one variable moves, either up or down, the other variable will move in the same direction. Perfect negatively correlation means the opposite. When one variable moves a certain way, the other variable will move in the opposite direction. A correlation of 0 means that
there is no correlation between the two variables and they move completely random. The coefficients are significant when they are below a p-value of 0,01. The different coefficients are divided into three significance levels: 0,05; 0,01; and 0,001. Gujarati (1988) mentions that there could arise an issue of multicollinearity when there is a correlation coefficient between two independent variables of at least 0,8. This means that one variable can be linearly predicted from the other variable with a non-trivial degree of accuracy. Table 1 shows that this is the case with the correlation of the variable average review score with meta review score (0,889) and the variable different review score and review change (0,800), however in this case it’s normal that they behave the same since the one variable is directly originates from the other. This is not a problem since they are not part of the same formula.
Table 1: Descriptive statistics and pearson correlation matrix VARIABLE MEAN SD 1 2 3 4 5 6 7 8 9 10 11 12 1. SALES 1,770 3,762 1,000 2. PREVSALES 1,578 2,774 ***0,505 1,000 3. M. SCORE 7,094 1,342 ***0,247 ***0,194 1,000 4. U.SCORE 7,217 2,405 ***0,087 ***0,124 ***0,384 1,000 5. CONC 188,593 126,527 0,012 -0,002 *-0,054 ***0,091 1,000 6. DIF. SCORE 0,763 0,782 **-0,073 ***-0,173 **-0,137 ***-0,102 0,038 1,000 7. SERIES 13,670 11,457 0,049 *0,063 *0,066 ***-0,089 ***-0,126 ***-0,164 1,000 8. GEN. CHA. 0,563 0,496 0,031 **0,079 ***0,157 0,040 ***-0,090 ***-0,226 -0,031 1,000 9. NAM. CHA. 0,522 0,500 0,013 0,014 0,007 0,000 *0,053 0,017 ***-0,158 ***0,117 1,000 10. REV. CHA. 0,260 0,439 **-0,072 ***-0,146 ***-0,120 **-0,080 0,028 ***0,800 ***-0,125 ***-0,207 0,004 1,000 11. CHANGE 0,845 0,362 -0,003 -0,004 **0,069 0,021 -0,039 **0,173 ***-0,165 ***0,486 ***0,447 ***0,254 1,000 12. SPORTS 0,361 0,480 0,042 **0,071 **0,071 0,012 ***-0,328 -0,051 ***0,180 ***0,202 ***-0,097 -0,032 0,045 1,000 *Significant at <0,05 **Significant at <0,01 ***Significant at <0,001
I will use a variable inflation factor (VIF) test to further check for multicollinearity issues in the variables. The results of this test are presented in table 2, 3 and 4. A variable could be considered as a linear combination of other independent variables when there is a VIF value of at least 10. All the table show that there are no VIF values higher as 10, this implies that there are no issues regarding multicollinearity. Table 2: VIF test (sales formula)
Table 3: VIF test (moderator change formula)
Table 4: VIF test (change formula)
VARIABLE VIF 1/VIF
PREVSALES 1,21 0,828282 M.SCORE 1,18 0,850009 U.SCORE SPORTS 1,05 1,01 0,956278 0,991206 MEAN VIF 1,11
VARIABLE VIF 1/VIF
PREVSALES 6,55 0,152650 CHANGE 6,32 0,158314 PREVCHANGE 1,33 0,749435 M.SCORE 1,21 0,824437 U.SCORE 1,18 0,849831 SPORTS 1,01 0,989139 MEAN VIF 2,93
VARIABLE VIF 1/VIF
COMP 1,14 0,874768
TIME 1,06 0,945040
PREVSALES 1,05 0,953348
26 Table 5 presents the results of a Skewness and Kurtosis test. This test tests the
symmetry of this distribution. A value around zero means a normal distribution. Table 5: Skewness/Kurtosis test for Normality
VARIABLE PR (SKEWNESS) PR (KURTOSIS)
SALES 0,0000 0,0000 PREVSALES 0,0000 0,0000 M.SCORE 0,0000 0,0018 U.SCORE 0,0000 0,0000 AVE. SCORE 0,0000 0,0000 DIF. SCORE 0,0000 0,0000 SERIES 0,0000 0,0000 GEN. CHA. 0,0001 - NAM. CHA. 0,1723 - REV. CHA. 0,0000 0,0000 CHANGE 0,0000 0,0000 SPORTS 0,0000 - 4.2. Analyses
In order to test hypothesis 1, 2, 4 and 5 a regression analysis is conducted. The results of these tests are presented in the tables of this paragraph. Also the results related for hypothesis 3 are presented in one of the tables of this paragraph, namely table 9.
U.SCORE 1,21 0,825883
SPORTS 1,15 0,867014
27 Hypothesis 1 tests the relation between sales and the independent and control variables prevsales, metascore, userscore and sport games.
Table 6: results regression hypothesis 1
SALES = α + β1 PREVSALES + β2 M.SCORE + β3 U.SCORE + β4 SPORTS + ε
SALES Coef. Std. Err. t P>t [95% Conf. Interval] PREVSALES 0,646804 0,030691 21,07 0,000 0,586601 0,707006 M.SCORE 0,474368 0,068201 6,96 0,000 0,340587 0,608148 U.SCORE -0,058533 0,037553 -1,56 0,119 -0,132193 0,015128 SPORTS -0,027793 0,174092 -0,16 0,873 -0,369287 0,313700 _CONS 0,688818 0,460002 -4,75 0,000 -3,087862 -1,283202
The results of this regression supports my first hypothesis. There is a significant positive relationship between the sales of the predecessor in a series and the sales of the new edition in a series. This effect is significant, since the p-value is lower as the selected value of alpha which is 0,05. The sales of the previous edition is a good predictor of the sales of the new edition. The power of the model is pretty high (an adjusted R-square of 0,277). This means that the model have a high amount of
explanatory power. The control variable SPORTS has not a significant effect on sales, therefore this variable can be removed from the model. This provides the results presented in table 5. The coefficients are slightly changed.
Table 7: results regression hypothesis 1
SALES Coef. Std. Err. t P>t [95% Conf. Interval] PREVSALES 0,646636 0,030621 21,12 0,000 0,586571 0,706701
28 U.SCORE -0,058180 0,037525 -1,55 0,121 -0,131787 0,015428
_CONS -2,198679 0,458283 -4,80 0,000 -3,097635 -1,299724
However, as explained in the literature review of my thesis, this relation is bad for the producer of the series when the sales of the original edition is bad. This is, because this implies that the sales of the new edition in the series will, probably, be bad as well. Therefore I will put the variable changes in the model, to test whether including this can weaken the positive relation. The results of this regression are presented in table 8.
Table 8: results regression hypothesis 2
SALES = α + β1 PREVSALES + β2 CHANGE + β3 PRESALECHANGES + β4 M.SCORE + β5 U.SCORE + ε.
SALES Coef. Std. Err. t P>t [95% Conf. Interval] PREVSALES 0,520423 0,075285 6,91 0,000 0,372745 0,668101 CHANGE -0,367616 0,265098 -1,39 0,166 -0,887626 0,152395 PRESALECHANGES 0,149582 0,081713 1,83 0,067 -0,010704 0,309867 M.SCORE 0,479807 0,068160 7,04 0,000 0,346105 0,613508 U.SCORE -0,057370 0,037508 -1,53 0,126 -0,130944 0,016204 _CONS -1,929664 0,496145 -3,89 0,000 -2,902891 -0,956437
The output of table 6 is already an indication that change moderates the
relationship between the sales of an earlier edition and the sales of the present edition (0,520423 compared to 0,646636). The power of the test is almost the same (R-square of 0,2810). This means that the explanatory power of the model is high. In addition a t-test is performed on the coefficient of the variable prevsales to test
29 whether the change of the coefficient is significant. The t-test is performed with 1477 degrees of freedom and an alpha of 0,05. The corresponding t-value is -64,4514 and the corresponding p-value is 0,0000, which is less than 0,05. Therefore I can conclude that the moderator change significantly weakens the relation between the sales of an earlier edition and the sales of the present edition.
I’ve made a new dummy variable to test my third hypothesis. As explained in the research design chapter, this dummy variable equals 1 when the sales of the previous edition are higher as the sales of the present edition and 0 otherwise. The total sample will be divided in two groups based on this variable. A t-test will then test whether the mean of the variable changes is higher in the low sales group compared to the high sales group. This gives the results as provided in table 9.
Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] 0 792 0,852273 0,012616 0,355054 0,827507 0,877038 1 686 0,836735 0,014122 0,369877 0,809007 0,864462 Combined 1478 0,845061 0,009415 0,3619691 0,826592 0,863530 Diff 0,015538 0,018881 -0,021499 0,052575
In order to test hypothesis 4 and 5 I will use the regression model of changes, which is: CHANGE = α + β1 COMP + β2 TIME + β3 PREV + β4 M.SCORE + β5 U.SCORE + β6 SPORTS + ε
The results of this regression are shown in table 10.
CHANGE Coef. Std. Err. t P>t [95% Conf. Interval] CONC -0,000101 0,000078 -1,30 0,194 -0,000255 0,000052
TIME -0,005931 0,000830 -7,14 0,000 -0,007560 -0,004303 PREV -0,001408 0,003415 -0,41 0,680 -0,008105 0,005290
30 M.SCORE 0,023386 0,007627 3,07 0,002 0,008424 0,038348
U.SCORE -0,003823 0,004232 -0,90 0,366 -0,012124 0,004478 SPORTS 0,046999 0,020677 2,27 0,023 0,006439 0,087559 _CONS 0,792321 0,054964 14,42 0,000 0,684504 0,900137
The results of this regression doesn’t support my fourth hypothesis. There is not a significant positive relationship between the sales of the predecessor in a series and the sales of the new edition in a series. The results of the regression doesn’t support my fifth hypothesis, although the p-value is higher as the selected value of alpha, which is 0,05. The coefficient is negative, which implies that there is a significant negative relation between changes and the number of editions in a series. This means that the more editions there are, the less chance of a change. The explanatory power of the model is low, an r-square of only 0,0406.
An interesting additional finding is that the meta score has a significant positive effect on changes which is very counter intuitive. The better the games are according to the expert, the more likely that a change will occur.
4.3. Robustness check
I will perform a robustness check to test the results on their robustness. The purpose of this test is to identify if the results are not the result of influential observations. This test is performed by removing the 25% lowest and the 25% highest sales. This is to remove all the possible outliers. The results of this test are shown in respectively table 11, 12 and 13. The coefficients are almost the same.
The robustness check of hypothesis 1 shows a much lower coefficient for PREVSALES, however the relation is still significant and positive.
Overall, the relations found in the regression analysis are robust meaning that conclusions can be drawn of these findings.
31 Table 11: robustness check hypothesis 1
SALES Coef. Std. Err. t P>t [95% Conf. Interval] PREVSALES 0,025684 0,010690 2,40 0,017 0,004697 0,046671
M.SCORE 0,053368 0,013857 3,85 0,000 0,026165 0,080571 U.SCORE -0,001349 0,007043 -0,19 0,848 -0,015175 0,012478 _CONS -0,412428 0,096590 4,27 0,000 -0,222803 0,602053
Table 12: robustness check hypothesis 2
SALES Coef. Std. Err. t P>t [95% Conf. Interval] PREVSALES 0,017551 0,027822 0,63 0,002 -0,0370703 0,072172 CHANGE 0,020751 0,055938 0,37 0,711 -0,089066 0,130568 PRESALECHANGES 0,009678 0,030195 0,32 0,749 -0,049600 0,068955 M.SCORE 0,052981 0,013881 3,82 0,000 0,025730 0,080233 U.SCORE -0,001124 0,007038 -0,16 0,873 -0,014941 0,012694 _CONS 0,395339 0,106875 3,70 0,000 0,185521 0,605156
Table 13: robustness check hypothesis 4 and 5
CHANGE Coef. Std. Err. t P>t [95% Conf. Interval] CONC -0,000141 0,000101 -1,40 0,162 -0,000339 0,000057
TIME -0,005519 0,001408 -3,92 0,000 -0,008283 -0,002755 PREV -0,003428 0,006814 -0,50 0,615 -0,016806 0,009950
32 M.SCORE 0,018236 0,009927 1,84 0,067 -0,001252 0,037725
U.SCORE -0,007916 0,005361 -1,48 0,140 -0,018440 0,002608 SPORTS 0,038785 0,026106 1,49 0,138 -0,012467 0,090037 _CONS 0,890637 0,077825 11,44 0,000 0,737852 1,043423
5. Discussion and conclusion
In this paper the relation between the performance of the present edition and the performance of the past edition is investigated together with the determinants of change. It’s hypothesized that there is a positive relation between the performance of the present edition and the performance of the past edition which is based on the paper of Situmeang et al. (2014). The performance of the past edition plays a
signaling role. High performance in terms of high sales might indicate that the game is good. At the same time, low sales might indicate that the quality of the game is bad. It’s hypothesized that a positive relation has a downside, therefore the moderating role of change is investigated. Examined is whether implementing change in the model can help to reduce the impact of past sales on the present sales. It’s also hypothesized that the possibility of a change is higher in the case of low past sales. Besides this relationship, the determinants of change are investigated. It’s investigated whether the variables competition and time have an impact on whether a change will occur or not. When there is a lot of competition, sales will probably become lower in future. It might therefore be profitable for firms to change to another genre. At the same time it might that consumers are getting bored after a time with a certain series. A firm will then probably change their name to give consumers the idea that they have developed something new.
The sample of this study consists of 1478 games of the three major consoles,
namely Nintendo WII, Microsoft Xbox, and Sony Playstations within the time period 2000-2012.
33 The empirical results support hypothesis 1. There is a positive relation
between past and future performance which confirms with the existing literature and theoretical contention. The literature review chapter of this thesis showed that
different theories explains this relationship, namely the prospect theory, the signaling theory; the irrational behavior theory; and the consumer socialization theory. The main message of these different theories is that the previous edition can help the consumer in making their purchase decision, because without direct
experience there is not much more to rely on.
The empirical results also support hypothesis 2. Change has a moderating role in the relation between the sales of the present edition and it’s predecessor. This is an interesting addition to the model of Situmeang et al. (2014) who found that there is a positive relation between past and future sales.
The empirical results of my thesis doesn’t found support for hypothesis 3, 4
and 5. I didn’t found support that the chance that a change occur is bigger when the past sales are low. I also didn’t found evidence that changes are more likely when there is more competition. Furthermore, I didn’t found support that changes are more likely when the number of editions in series are higher. Results of my thesis even found the opposite. I found evidence that the chance of a change decreases when the number of editions grow.
A limitation of my research method is the sample I will use. In my research I
will only include games in my sample. Although other studies also only include the game industry in their sample to examine product series, it’s never tested whether this sample is appropriate to draw conclusions for series in general. This undermines differences between series in different product categories. Further research should be on other industries, for example the movie industry, or the car industry.
Another limitation of my research is more an overall disadvantage of database
research. This is the use of secondary data which is not specifically selected in order to examine my different hypotheses.
34 Also the use of online data collection has some disadvantages which should not be overlooked. Some of these disadvantages are unreliability and that the data is not as universal as should be. In the setting of the reviews in the games industry it might be that people with negative thoughts about a game are more willing to write a review than users with positive thoughts (Lefever, Dal, & Matthiasdottir, 2007).
Further research should try to expand my research by looking to more
antecedents of changes. This research has only looked to two antecedents of changes, namely competition and time. A qualitative field study should investigate more antecedents.
6. Bibliography
Bao, Y., Zhou, K. Z., & Su, C. (2003). Face consciousness and risk aversion: Do they affect consumer decision‐making? Psychology & Marketing, 20(8), 733-755.
Boone, J. (2008). A new way to measure competition*. The Economic Journal, 118(531), 1245-1261.
35 Brown, B. B., Clasen, D. R., & Eicher, S. A. (1986). Perceptions of peer pressure, peer
conformity dispositions, and self-reported behavior among adolescents. Developmental Psychology, 22(4), 521.
Campbell, C., Hirsch, E., & Silverstone, R. (1992). The desire for the new. Consuming Technologies: Media and Information in Domestic Spaces, , 48-64.
Connelly, B. L., Certo, S. T., Ireland, R. D., & Reutzel, C. R. (2011a). Signaling theory: A review and assessment. Journal of Management, 37(1), 39-67.
Connelly, B. L., Certo, S. T., Ireland, R. D., & Reutzel, C. R. (2011b). Signaling theory: A review and assessment. Journal of Management, 37(1), 39-67.
Deloitte (2007). Most consumers read and rely on online reviews; companies must adjust. Available at www.marketingcharts.com/interactive/most-
consumersread-and-rely-on-online-reviews-companies-must-adjust-2234/. (accessed 24
June, 2015).
Fischer, M., Völckner, F., & Sattler, H. (2010). How important are brands? A cross-category, cross-country study. Journal of Marketing Research, 47(5), 823-839.
Hennig-Thurau, T., Houston, M. B., & Heitjans, T. (2009). Conceptualizing and measuring the monetary value of brand extensions: The case of motion pictures. Journal of Marketing, 73(6), 167-183.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica: Journal of the Econometric Society, , 263-291.
36 Keller, K. L. (1993). Conceptualizing, measuring, and managing customer-based brand
equity. The Journal of Marketing, , 1-22.
Kurtzleben, D. (June 26, 2012). Why sequels have taken over the box office. Retrieved from
http://www.usnews.com/news/articles/2012/06/26/why-sequels-have-taken-over-the-box-office
Lefever, S., Dal, M., & Matthiasdottir, A. (2007). Online data collection in academic research: Advantages and limitations. British Journal of Educational Technology, 38(4), 574-582.
Lightspeed Research (2011). Consumers rely on online reviews and price comparisons to make purchase decisions. Available at http://
www.lightspeedresearch.com/press-releases/consumers-rely-on-onlinereviews-and-price-comparisons-to-make-purchase-decisions/. (accessed 24
June 2015).
Maanen, v. J. (Ed.). (1983). Qualitative methodology. London: Sage.
Rabin, M. (2000). Risk aversion and expected‐utility theory: A calibration theorem. Econometrica, 68(5), 1281-1292.
Situmeang, F. B., Leenders, M. A., & Wijnberg, N. M. (2014). History matters: The impact of reviews and sales of earlier versions of a product on consumer and expert reviews of new editions. European Management Journal, 32(1), 73-83.
Sood, S., & Drèze, X. (2006). Brand extensions of experiential goods: Movie sequel evaluations. Journal of Consumer Research, 33(3), 352-360.
37 Sood, S., & Keller, K. L. (2012). The effects of brand name structure on brand extension
evaluations and parent brand dilution. Journal of Marketing Research, 49(3), 373-382.
Tschang, F. T., & Szczypula, J. (2006). Idea creation, constructivism and evolution as key characteristics in the videogame artifact design process. European Management Journal, 24(4), 270-287.
Tversky, A., & Kahneman, D. (1986). Rational choice and the framing of decisions. Journal of Business, , S251-S278.
Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297-323.
Völckner, F., & Sattler, H. (2006). Drivers of brand extension success. Journal of Marketing, 70(2), 18-34.
