How to deal with negative OCRs
Overcoming the negative effect of negative Online Customer Reviews with the use of price promotions.
Michiel Zwama
June, 18 2018
University of Groningen Faculty of Economics and Business
2
How to deal with negative OCRs
Overcoming the negative effect of negative Online Customer Reviews with the use of price promotions.
Michiel Zwama S3272885
University of Groningen Faculty of Economics and Business
MSc Thesis MSc Marketing management June, 18 2018 Friesestraatweg332 9718 NT Groningen m.zwama.2@student.rug.nl Student number: 3272885
4
Management summary
In this study, it was investigated whether the negative effect of a negative OCR on purchase intentions could be outbalanced by the positive effect of a price promotion. Moreover, it was examined whether the regulatory focus of consumers interacts with the effects of both negative OCRs and price promotions. Thus far, research specified towards dealing with negative OCRs has been limited. Most research is centered towards complaint management. This is somewhat surprising since the negative impact of negative OCRs has been proven. Moreover, dealing with these negative OCRs is something companies, and especially marketing managers are facing more and more.
A choice-based conjoint analysis was performed, which lead to preference-based utilities for multiple attributes. Negative OCRs and price promotions were defined as attributes
containing multiple levels. For negative OCRs this means there were multiple levels of valence, for price promotions there were multiple levels of discounts. Three models were specified, one aggregate model, and a latent class analysis resulted into two different segments of consumers with similar preferences.
All models concluded that the more negative a review is, the more negative the effect on purchase intentions becomes. The same holds for price promotion, but the other way around: the more positive a price promotion is, the more positive the effect on purchase intentions. Whether the negativity of a negative OCR could actually be outbalanced by a price promotion and at what level of price promotion depends on the segment consumers are in.
5
Acknowledgements
I started the pre-MSc marketing two years ago, leaving the HBO as quite an average student with a bright mind, but a lazy attitude. The past two years have challenged me to continuously perform on a high level, not settling for average. I personally feel like two years of RUG has taught me more than four years of HBO. Moreover, it helped me in becoming an adult, and behaving like one. Play time is over now, welcome to the real world. I hereby present to you: my key to the real world.
I want to start by thanking Dr. J.A. Voerman, my first supervisor, I could not have wished for a better and, nicer, advisor. Thank you for always providing me with both feedback and support. I would also like to acknowledge second supervisor A. Onrust.
I am, thankful for everyone that has taken the effort to participate in my research by filling in the survey. Also a special thanks to my thesis group, for providing me with insights in the group meetings, but also via the App group.
A special thanks to my mother for having a final check on minor errors. I also want to thank my father and brother for supporting me and of course my girlfriend, for both supporting me and accepting the fact that she did not get all the attention she would have wanted over the past weeks.
You were all very important to me over the past two years, and I cannot thank you all enough for it.
6
Table of contents
1. Introduction ... 8
1.2 eWOM ... 8
1.2 OCRs ... 9
1.3 The effect of eWOM & OCRs ... 10
1.4 Negative OCRs ... 11
1.5 Purchase intentions ... 12
1.6 Problem statement and research questions ... 12
1.7 Thesis structure ... 13
2. Theoretical Framework ... 14
2.1 The impact of negative OCRs ... 14
2.2 Cognitive dissonance and information avoidance ... 16
2.3 Regulatory Focus ... 17
2.4 The impact of price promotions ... 17
2.5 Dealing with negative OCRs ... 18
2.6 Conceptual model ... 19
3. Methodology ... 21
3.1 Methods ... 21
3.1.1 Attributes and levels ... 22
3.1.2 Moderators ... 25
3.1.3 Control variables ... 25
3.2 Choice Design ... 26
3.3 Data Collection ... 27
3.4 Plan of analysis ... 28
3.4.1 Population and sampling method ... 28
7
3.4.3 Model specification ... 31
3.4.4 Model fit and validation ... 31
4. Results. ... 32
4.1 Model specification ... 32
4.2 Aggregate model main effects ... 34
4.3 Aggregate model moderating effects ... 36
4.4 Latent Class segmentation ... 40
4.4.1 Moderation segment 1 ... 41 4.4.2 Moderation segment 2 ... 43 4.5 Purchase intention ... 44 4.6 Hypothesis Overview ... 46 5. Discussion ... 48 5.1 Main effects ... 48 5.2 Regulatory focus ... 50 5.3 conclusive ... 50 5.4 managerial implications ... 51
5.5 further research and limitations ... 51
Reference list ... 54
Appendix A Pre-test results ... 61
Appendix B SPSS output Factor analysis + Reliability analysis ... 63
Appendix C The Survey ... 66
8
1.
Introduction
Word-of-mouth marketing (WOM) has long been recognized for having a strong influence on product attitudes, consumer behavior and sales, and has been the subject of many studies (e.g. Arndt, 1967; Brown & Reignen, 1987; Herr, Kardes & Kim, 1991). Misner (1994) takes it one step further by calling it the “most effective, yet least understood form of marketing”. He made clear that the impact of WOM is big, but incorporating WOM marketing into the marketing process has proven to be difficult for many companies.
Dichter (1966) says that a WOM recommendation is the strongest ally of any product, having more impact on purchase intentions than any other form of advertising. Moreover,
advertising, which is passively receiving information, is always outweighed by personal recommendations (Dichter, 1966). In the past, Word of mouth marketing was merely centered as giving a face-to-face recommendation, meaning that in many cases people were only able to provide people they know with their opinions and experiences.
1.2 eWOM
However, the emergence of the internet has provided consumers with multiple new options to share their opinions and experiences with other consumers (Trusov, Bucklin & Pauwels, 2009). Electronic Word Of Mouth (eWOM), using the internet as a venue to share opinions, experiences and advice on products and services was one of the fastest growing internet phenomena at, and just before, the turn of the century (Armstrong & Hagel, 1996). The internet has therefore extended the possibility for consumers to gather and search for information, written and shared by other consumers (Henning-Thurau et al., 2004). This means consumers no longer have to rely on the information that is provided by the company, which could be biased, or information from and to the few people they know. Consumers now have access to a vast amount of product information provided by a diverse group of other consumers that have experience with the product or service, meaning recommendations are also less biased by the opinion of few (Ratchford et al., 2001).
Moreover, the emergence of the internet and eWOM has entirely changed the way consumers gather information and how consumers rely on the information source (Cheung, Lee & Rabjohn., 2008; Park, Lee & Han, 2007). Facilitating consumers with the option to post their opinions, eWOM is a marketing tool that can be used to compete for the attention of
9 Henning-Thurau et al. (2004) define eWOM as follows: “Electronic word‐of‐mouth (eWOM) communication refers to any positive or negative statement made by potential, actual, or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet”. Due to the internet, mass communication has become bidirectional: individuals can now easily access information from others but also provide other internet users all over the globe with own experiences (Dellarocas, 2003).
A great amount of research has been done in the field of electronic word of mouth, covering many aspects within the domain. As compared to traditional WOM, satisfied consumers are more likely to share their experiences when shopping online (Cheung & Lee, 2012). On the downside for companies however, consumers were also very likely to participate in eWOM when dissatisfied (Lee & Cude, 2012). This means companies are forced to deal with negative eWOM more often.
1.2 OCRs
Online Customer Reviews (OCRs) make up a large part of eWOM. OCRs are customer reviews that are provided for a wide arrange of products and services. OCRs form a personalized advice or product descriptions and are mainly generated by automated recommendation systems. OCRs provide helpful information for both consumers and companies (Mudami & Schuff, 2010). OCRs have become one of the main sources of
information for consumers to evaluate products before actually buying them (Cui, Lui & Guo, 2012).
There is strong support that OCRs have a strong impact on the product choices that consumers make: recommended products are sometimes twice as likely to be chosen by consumers (Senecal & Nantel, 2004; Park & Park, 2008).
10 Both findings on OCRs and the overarching subject eWOM are used for this paper since the two are closely related to each other. OCRs are the most used form of eWOM, meaning most findings on eWOM also apply to OCRs (Purnawirawan, Dens & de Pelsmacker , 2012).
1.3 The effect of eWOM & OCRs
Several studies address how eWOM and OCRs impact sales and sales elasticity. Floyd et al. (2014) state that online product reviews have a significant effect on sales, with reviews by critics, third party sources and valence; either a positive or negative opinion, increasing this effect. Moreover, they find that OCRs have greater impact on sales elasticities for high-involvement products. Interesting is their finding that valence has a greater impact on sales than review volume, meaning the negativity or positivity of OCRs have a bigger impact on sales than the sheer number of OCRs. They also point out that sales are a good indicator on eWOM performance. You, Vadakkepatt & Joshi (2015) draw the same conclusion, stating that eWOM has among the highest short term elasticities, with only price having more impact on sales. In the presence of OCRs, consumers do no longer value brand strength when making purchasing decisions (Kostyra, et al. 2016). This gives an indication on how
important the role of OCRs is in the online consumption process. Park & Park (2008) found that OCRs do not only influence sales, but also that intention and the attitude on the product or service are significantly influenced by the valence of OCRs. This means negative OCRs could inflict negative attitudes towards the product or company.
Floyd et al., (2014) and You, Vadakkepatt & Joshi (2015) add to the existing field of research by proposing that negative ratings drastically affect valence elasticity compared to neutral or positive OCRs. The research also suggests that the effect of eWOM on sales elasticities is dependent on various factors like product, industry and platform. Most important finding is that of all marketing mix variables, eWOM has among the highest elasticities in the short term after price elasticity, stretching the importance of eWOM for marketing managers (You, Vadakkepatt & Joshi 2015).
Online reviews have a larger impact on sales for lesser known products than for well-known products, reason for this is the scarcity of product information. Downside to this is that one single negative review could have a great impact. For marketing managers that are
11 purchasing behavior of consumers. An improvement in book reviews led to an increase in sales (Chevalier & Mayzlin, 2006). The same effect was measured for experience goods, an early positive review on a movie can enhances the number of visits (=sales) at the expense of other movies (=competition) (Reinstein and Snyder 2005).
Forman, Ghose and Wiesenfeld (2008) also conclude that OCRs impact sales, they find that reviews have stronger effect on sales when consumers can identify with the source of the review; a heuristic kicks in. Finally, Chen, Wy & Yoo (2004) add to the field of research by stating that more recommendations are related to more sales, the number of reviews is positively related to sales.
1.4 Negative OCRs
So, eWOM elasticity has one of the strongest short-term elasticities as compared to other marketing mix variables (You Vadakkepatt & Joshi 2015). This also means that negative OCRs could seriously harm a company: a negative review could result in lost sales. On top of that, previous research on offline WOM as well as eWOM points out that there is a negativity bias regarding reviews, meaning negative OCRs weigh heavier than positive OCRs (You, Vadakkepatt & Joshi, 2015; Lee, Park & Han, 2008).
The mere fact that negative OCRs weigh heavier than positive OCRs and the fact that eWOM has among the highest short-term elasticities clarifies that OCRs are indeed a strong
marketing tool for increasing sales (You, Vadakkepatt & Joshi, 2015). The downside of course is that negative OCRs could decrease sales for a company. Not dealing with negative OCRs could impose great risk for marketing managers.
As a result, dealing with these negative OCRs has become an important task for marketing managers. Dealing with negative OCRs is the problem at hand in this paper. Current literature on how to deal with negative OCRs is very limited, and merely pointed towards complaint management (e.g. van Noort and Willemsen, 2012; Kerkhof et al., 2010).
12 considered when writing an OCR. Adapting the price after an OCR however is possible, in the form of price promotions. Price promotions are one of the most popular and most frequently used means to stimulate sales in the short run (e.g. Inman & McAllister, 1993). Therefore this paper proposes that price promotions could be used to diminish, or outbalance, the negative effect of negative OCRs on sales.
1.5 Purchase intentions
This large amount of research conducted towards the relationship between eWOM and/or OCRs and sales shows there is a significant positive relationship between the two. Moreover, they point out that sales are a good indicator on eWOM performance (You, Vadakkepatt & Joshi, 2015; Floyd et al., 2014) . However, sales and eWOM data are needed to use product sales as a dependent variable. Various studies have used purchase intentions as a proxy for sales, or to forecast sales since there is a positive relationship between the two (Juster, 1966;
Morwitz & Schmittlein, 1992; Morwitz, Steckel & Gupta 2007; Kalwani & Silk, 1982). In this study, given the lack of sales- and eWOM data that is available to the author, purchase intention will be used as a proxy for sales.
1.6 Problem statement and research questions
So far, it has been described how the impact of OCRs has grown in the purchase process of consumers due to the evolution of the internet. After price, OCRs have the strongest
elasticities out of all marketing mix variables. For marketing managers this means negative OCRs can seriously harm sales, purchase intentions and attitudes. Dealing with negative OCRs has become an important part of the job responsibilities of many marketing managers. This thesis paper aims to investigate whether the negative effect of OCRs could be
outbalanced using price promotions, but also which level of price promotion is needed. The following research question was formed:
Which level of price promotion is needed to overcome the negative effect of negative OCRs?
13
1.7 Thesis structure
14
2. Theoretical Framework
In the theoretical framework all mechanisms and additional background information that can be found in academic literature on the independent variables and the dependent variable is given. Moreover, a more detailed presentation of all concepts that were introduced in the previous chapter will be provided.
2.1 The impact of negative OCRs
As mentioned before, there is quite some evidence on the effect of eWOM and OCRs on sales, and the strength of it. eWOM elasticity has one of the strongest short-term elasticities of all marketing mix variables (You Vadakkepatt & Joshi 2015). Meaning negative OCRs could seriously harm a company, product or service: a negative review could result lost sales. In literature, the negativity or positivity of reviews is referred to as valence. Valence is the average rating of OCRs and represents the satisfaction level of consumers (Chintagunta, Gopinath, Venkataraman, 2010). OCR valence is generally based on star-ratings that accomplish the OCRs. Ratings of 1 to 3 can be considered negative, in which 3 is
neutral/slightly negative and 1 is very negative. (Kostyra et al., 2016; Moe & Trusov, 2011). An OCR accompanied by a 3 star rating usually address both positive and negative aspects of a product. Since it was found that negative information has a larger impact on sales, intentions and attitudes (Chevalier & Mayzlin, 2006) an OCR accompanied by a three star rating is seen as slightly negative in the remainder of this thesis.
Minor dissatisfactions often lead to minor consumer responses, in most cases these do not lead to negative word of mouth. When a dissatisfaction is big enough, consumers will complain, this does however not necessarily mean consumers will spread negative WOM. When a company is dealing with complaints this works as a cure to prevent from giving negative WOM, regardless of whether the complaint was settled or not. Only when it is discouraged by companies to complain about satisfaction, chances are consumers will participate in facilitating negative WOM (Richins, 1983).
15 Meyers-Levy, 1990; Skowronski &. Carlston 1989); “Bad is simply stronger than good” (Baumeister et al., 2001).
The usage of the internet to spread WOM has however changed a lot of things, it has become relatively easy to spread a negative experience with a big audience, and consumers might be tempted to provide negative eWOM when dissatisfied regardless of the complaint procedure of a company (Henning-Thurau et al, 2004). Even though there are various differences
between WOM and eWOM, negative eWOM also weighs heavier than positive eWOM (You, Vadakkepatt & Joshi, 2015; Lee, Park & Han, 2008).
Negative information has more value to consumers than positive information, therefore negative information is weighed heavier by consumers than positive in the decision making process. For eWOM this means that negative reviews will be evaluated and used more than positive reviews (Sen, Lerman, 2007). This negativity bias translates into the following behavior; consumers will assess negative eWOM over positive and negative information is trusted more than positive information. However, this negativity bias only exists for utilitarian products and not for hedonic products (Sen and Lerman 2007). Chevalier and Mayzlin (2006) find that the impact of 1-star OCRs is greater than that of 5-star ratings and state that negative online product ratings are more powerful in preventing sales than positive online product ratings are in increasing sales. These findings were replicated by Sun (2012) and
Wangenheim (2005).
There is however also a body of research that states the opposite; negative eWOM does not have a stronger effect, and moreover negative eWOM does not have to translate into a
negative effect on sales. Doh and Hwang (2009) state that when there are only a few negative OCRs, this might actually have a positive influence on sales. Reasoning behind this statement is that consumers question the credibility of a certain message or page in the absence of negative reviews, or when there are only few negative reviews. They state that a single negative message could be harmful indeed on product sales, but one negative OCR in the presence of ten positive reviews can be more beneficial than there being just positive reviews. But even when the negative OCR is not accompanied by multiple positive OCRs, the effect of a negative review can be positive, for example in the movie industry (Liu, 2006, Sandes & Urdan, 2013).
16 eWOM has. The studies that contradict this claim all state that the occurrence of negative reviews not having negative effects on sales only happens in certain conditions. For example; when consumers have prior knowledge or when reviews are extremely negative, the negative effects on sales disappear (Herr, Kardes & Kim, 1991). There for the first hypothesis is:
H1: The more negative an OCR is, the more negative the effect on purchase intentions becomes.
2.2 Cognitive dissonance and information avoidance
The fact that negative reviews have a bigger impact on sales than positive reviews derives from the loss aversion principle in the prospect theory (Kahneman & Tversky, 1979). The loss aversion principle states that losses loom larger than gains. Negative OCRs point out the negative aspects of a product: the risks, resulting in greater perceived losses (Sandes & Urban, 2013). Experiences of other consumers provide information on product quality, which is used to reduce risk when shopping online (Cui, Lui &Guo, 2012; Zhu & Zhang, 2010). But as soon as perceived gains become more attractive than losses, consumers tend to ignore the information that is available and negative, a price promotion could facilitate this increase in attractiveness of gains.
The underlaying theory behind this phenomenon is cognitive dissonance (Festinger, 1957). According to Festinger (1957) consumers are inclined not to use information that makes them uncomfortable. So, once committed to the choice of a product or service people prefer to use consonant information versus dissonant information; i.e. selective exposure to information or so-called information avoidance (Sweeney et al., 2010). In the light of this study this means the perceived gains induced by a price promotion become the consonant information, while the negative OCRs are information that makes consumers uncomfortable.
Golman, Hagmann & Loewenstein (2017) confirm that information avoidance results from loss aversion. Information is used to guard the consumer against possible disappointment. Existing- or prior beliefs about a product or service is one of the main sources of dissonance: more commitment to a product or service means more cognitive dissonance towards
conflicting information (Jonas et al., 2001; Frey, 1986). The idea behind this form of
17 are therefore the mechanism that would explain how both negative OCRs and price
promotions impact the online purchase process.
2.3 Regulatory Focus
Not all consumers are equally susceptive to losses and gains. What some consumers might consider a considerable risk, might not be for another consumer. Some consumers focus more on maximizing gains whereas others are merely trying to minimize the risk of losses. Higgins (1997) identified two types of consumers relating hedonic regulation: those motivated by promotion and those motivated by prevention: the regulatory focus of consumers. Whereas for the group motivated by promotion the emphasis lies in approaching positives, the emphasis for the latter group, whom are merely motivated by prevention, lies in avoiding negatives. Promotion focused individuals use a positive reference point while prevention focused consumers use a negative reference point. People seek pleasure by either avoiding pain or promoting pleasure. Whether consumers aim at avoiding losses or maximizing gains impacts how consumers obtain and interpret information (Sandes & Urdan, 2013). The regulatory focus of consumers might therefore influence how they perceive Negative OCRs, but also price promotions. It could be argued that consumers who aim at avoiding losses are more susceptive to negative OCRs, while consumers aiming at maximizing gains are more susceptive to price promotions.
H2a The effect of a negative OCR on purchase intentions becomes more negative (positive) when consumers are prevention (promotion) focused.
H2b The relative importance of OCRs increases (decreases) when consumers are prevention (promotion) focused.
H3a The effect of a price promotion on purchase intentions becomes less (more) positive when consumers are prevention (promotion) focused.
H3b The relative importance of price promotions increases (decreases) when consumers are promotion (prevention) focused.
2.4 The impact of price promotions
18 short-term. Price promotions however only work for a short amount of time. The short term elasticity of price promotions is among the highest (Hoek & Roelants, 1991). Price savings in particular are the most common type of price promotions. They benefit consumers with both hedonic and utilitarian benefits, not just monetary savings (Chandon, Wansink & Laurent, 2000) which in turn can result in a short term increase of sales (Aaker, 1992).
Selvakumar & Vikkraman (2011) found that price promotions do not only have positive outcomes on short term. They found evidence that short term monetary price promotions negatively impact loyalty and perceived quality.
H4: The higher a price promotion becomes, the more positive the effect on purchase intentions is.
2.5 Dealing with negative OCRs
Existing literature on dealing with negative OCRs is merely pointed towards the field of complaint management. Literature specifically pointed at responding to negative OCRs is scarce however. Marketing managers should actively search for reviews that are placed online, moreover, consumers expect companies to actively engage in online conversations (Kietzman & Kahoto, 2013).
In the hotel industry responding to a negative OCR resulted in a more positive evaluation of the company as compared to not responding at all. The speed of reacting is also positively related to a positive evaluation (Sparks, So & Brandly, 2016). These results are consistent with the findings of van Noort and Willemsen (2012) on online damage control. Proactive actions like apologizing or compensating in many cases help rebuilding the trust of
consumers, however there is also the risk that the company appears to be accepting
responsibility for failure (Lee & Cranage, 2014; Lee & Song, 2010). This means there is some confusion for marketing managers regarding how to deal with negative OCRs given these statements contradict each other. According to Kerkhof et al. (2010) a financial compensation is a good way to raise the valence of OCRs.
This paper will examine an entirely different way of battling the negative effect that a
19 It would make sense that the negative effect of a slightly negative OCR could be outbalanced with a price promotion. For example, in various review sentences like: “some problems, but at this price I can’t complain” show that price promotions have quite an important impact on the evaluation of the product. Both quality and price play an important role in consumer decision making. Quality is uncertain in most cases before the purchase. OCRs partially resolve this uncertainty. Price promotions can bias these reviews on quality; “can’t expect higher quality at this price”. This gives companies a new opportunity when quality is not top notch; using price promotions might lead to benefits on both sales, and future OCRs (Li & Hitt, 2010).
H5 A price promotion can be used to outbalance the negative of a negative OCR.
2.6 Conceptual model
Figure 1: conceptual model.
20 negative OCRs on purchase intentions: H2 and H3. The finale hypothesis: H5 proposes that the positive effect of price promotions will be high enough to overcome the negative effect of negative OCRs. Table 1 provides a final overview of all hypotheses.
Table 1: Hypotheses.
H1 The more negative an OCR is, the more negative the effect on purchase intentions becomes.
H2a H2 The effect of a negative OCR on purchase intentions becomes more negative (positive) when consumers are prevention (promotion) focused.
H2b The relative importance of OCRs increases (decreases) when consumers are prevention (promotion) focused.
H3a H3 The effect of a price promotion on purchase intentions becomes less (more) positive when consumers are prevention (promotion) focused.
H3b The relative importance of price promotions increases (decreases) when consumers are promotion (prevention) focused.
H4 H4: The higher a price promotion becomes, the more positive the effect on purchase intentions is.
21
3. Methodology
This chapter elaborates on the research method, data collection and data analysis. Moreover, a plan of analysis is included on the tests that will be used for the analyzation part.
3.1 Methods
The goal of this research is to determine whether the negative effect of a negative OCR could be balanced out with a price promotion. For this purpose, a choice-based conjoint experiment was used. The conjoint analysis is an experimental method that is mainly used to find
preferences of consumers for certain products or services. Products or services are perceived as “attribute bundles”, treatments consist of a combination of experimental factors that are “CONsidered JOINTly” (Green & Wind, 1975; Green & Srinivasan, 1978).
The valuation of consumers for a certain product or service creates a utility function, that can be used for prediction. The utility of a product or service is equal to the sum of the utilities of all attribute levels. These utilities are statistically decomposed from the overall product or service evaluation. The conjoint analysis also reveals the underlaying motives consumers have for their actions. So, the conjoint analysis helps identifying which attributes and
characteristics have which impact on choice processes of consumers. It could also be used for measuring price elasticity or for segmentation (Eggers & Sattler, 2011).
For this specific study the conjoint analysis will provide utilities based on preferences of the respondents on attributes and attribute levels: price promotions and negative OCRs in this case. This provides insights in whether the utility of price promotions can outbalance the negative utility resulting from a negative OCR. Moreover, the conjoint analysis gives segmentation possibilities.
In this research, the Choice Based conjoint analysis (CBC) was used (Louviere &
Woodworth, 1983). In CBC consumers repeatedly choose their preferred option from a set of alternatives. The advantage of using the CBC instead of a rating based conjoint analysis lies in the fact that choosing between various alternatives is more natural than rating products. Consumers are relatively often faced with tradeoffs in purchase situations, it is part of the daily life for many consumers (Eggers & Sattler, 2011). For the estimation procedure an aggregate logit will be used, followed by a latent class conjoint analysis. With the aggregate logit consumer preferences are measured on an aggregate level, therefore assuming
22 class analysis will be used to group consumers with similar preferences together in segments (classes).
3.1.1 Attributes and levels
The initial step in designing a CBC is choosing the right attributes and according levels. The attributes should represent the product or service as good as possible. In general, no more than six attributes should be added (Eggers & Sattler, 2011). In this study only two attributes are considered interesting; i.e. the level of price promotions and the negativity level of OCRs. However, to increase reality three more “filler attributes” are added: these attributes have no role in the analysis though. Offering a realistic setting, affects both accuracy and precision of the model (Eggers, Hauser & Selove, 2016). Before deciding on all attributes and attribute levels a product has to be selected. OCRs have more impact on search goods than on experience goods (Cui, Lui & Guo, 2012). Therefore a pre-test (n=16) was conducted in which consumers had to choose one out of various “popular” search goods. The respondents then had to choose the product for which they depend most heavily on OCRs. The pre-test resulted in people depending most on reviews when searching for a new camera. For the results of the pre-test see appendix A.
When selecting the attribute levels a few things have to be considered (Green & Srinivasan, 1978; Eggers & Sattler, 2011) :
- Levels should span a range that is larger than it would be in reality, to cover any possible scenario
- Levels should have concrete meaning
- The number of levels should be kept low, 3 to 5 levels maximum.
- The number-off-levels-effect should be avoided, meaning the number of levels should be balanced across all attributes (Verlegh, Schifferstein & Wittink, 2002).
- Attributes should be mutually exclusive.
23 Klapper, 2016). The input for the pre-test was determined by Amazon reviews accompanied with 1-to-3 star ratings on various search goods. The reviews were than combined and generalized, since the product-specific content depends on the outcomes of the pretest. The seven reviews that were shown to the respondents can be found in Appendix A.
Table 2: Negativity levels OCRs
Level Dutch English
1 (slightly negative)
Redelijk product maar verwacht geen indrukwekkende prestaties, het design is mooi maar de beeld kwaliteit valt enigszins tegen.
It is not a bad product, however, don’t expect to be impressed by its performance. The design is nice, but the quality is somewhat lacking.
2 Geen verschrikkelijk product maar het voldoet niet aan de verwachtingen, de camera ziet er mooi uit maar hij is absoluut niet gebruiksvriendelijk: Tegenvallend!
It is not a horrible product, but it does not meet the expectations. The camera looks nice but is not user-friendly at all: Disappointing!
3 (Very negative)
Ik kocht dit product voor persoonlijk gebruik en het is een van de slechtste aankopen die ik ooit heb gedaan, niet gebruiksvriendelijk, daarnaast heeft het te weinig ruimte en werken bepaalde
functies niet als aangegeven. Als je overweegt dit product te kopen raad ik aan om verder te zoeken.
I bought the camera for personal use and it is one of my worst purchases ever. The camera is not user-friendly, has limited capacity and certain specifications do not function as promised. If you are considering this product I advise you to look further.
24 the acceptance price of consumers. This means consumers no longer trust the brand, product or service due to the drop in price (Morwitz, Greenleaf & Johnson, 1998). The 50% price promotion will therefore be the maximum. The other levels will be set at none, 10% and 25% since it is assumed that the effect of price promotions increases linear: see hypothesis 4. As mentioned before, creating a conjoint analysis that is similar to a realistic purchase decision situation improves valence (Eggers, Hauser, and Selove, 2016). Three “filler attributes” are added, since in a realistic setting consumers focus on more than just price and reviews. The filler attributes are based on previous research on features that consumers value most when buying a new camera. Decker and Trusov (2010) conducted a conjoint analysis on the purchase of a camera, basing their attributes and attribute levels on consumer reviews. Reliability and manufacturing quality came forward on being the most important attributes, with good to moderate being the levels. In this study however, these are not valuable
attributes: they are too abstract. A more interesting research, that also made use of a conjoint analysis concluded that a wide angle lens, (on levels 28-35) and whether Wi-Fi
communication was possible (yes/no) were the most important attributes, or at least they had the highest utilities (Wang, 2015). When shopping online these are however not attributes that are prominently shown, and therefore using these would not reflect a realistic online shopping situation.
Better fitting filler attributes are brand, pixels and zoom with attribute levels Canon, Sony and Nikon, which are the most popular brands. 10 and 16 megapixels for the pixel attribute and 4x and 10x as levels for the optical zoom (Allenby et al., 2014; Kim et al. 2016). Looking on Amazon tells you that megapixels have improved therefore 18 and 20 will be the attribute levels. To improve valence, pictures of similar cameras of each brand are added (Eggers, Hauser, and Selove, 2016). OCRs have higher impact for high involvement products (Floyd et al., 2014), therefore the selected cameras are priced around €200,-.Table 3 provides on
25 Table 3, all attributes and levels
Attribute Attribute level 1 Attribute level 2 Attribute level 3 Attribute level 4 Price Promotions None (€200,-) -10% (from €200,- for €180,-) -25% (from €200,- for €150,-) -50% (from €200,- for €100,-) Negative OCRs (see table 1) Slightly negative OCR
Negative OCR Very negative OCR
Brand Canon Nikon Sony
Zoom 4x 10x
Megapixels 18 20
A possible drawback of this design is the fact that the number-of-levels effect might occur since not all attribute levels are balanced, meaning respondents could regard attributes with more levels as being more important (Eggers & Sattler, 2011; Verlegh, Schifferstein & Wittink, 2002).
3.1.2 Moderators
The moderating role of regulatory focus (Higgins 1997), which can be split up into promotion focus and prevention focus, is measured using the Regulatory Focus Questionnaire (RFQ) (Higgins et al., 2001). There are multiple scales to measure regulatory focus, but the RFQ is regarded as the best measurement (Haws, Dholakia & Bearden, 2010). The RFQ is an 11-item 5-point-likert scale, with six 11-items measuring promotion focus and the other five measuring prevention focus. For this research a 7-point-likert-scale was used ranging from totally disagree to totally agree.
3.1.3 Control variables
Three different control variables were measured, all demographic: age, gender and
26 measured with an open question while gender was nominal and educational level was
measured with an eight-point ordinal question.
3.2 Choice Design
The choice design, which entails the number of choice sets and the attribute level
combinations that have to be included, is a critical step in the CBC. Finding the ideal choice design continues to receive a great amount of attention in scientific research, but so far there is no ideal choice design (Eggers & Sattler, 2011). There are however, some guidelines regarding the ideal choice design.
The CBC is quite a complex survey to fill in for respondents, since they are desired to repeatedly choose the preferred option out of various similar looking attributes and levels. Respondent fatigue resulting from the task being too long could seriously impact choice behavior, respondents might only focus on a certain aspect instead of processing all attribute levels (Eggers & Sattler, 2011). Meissner, Oppewal, & Huber, (2016) suggest to use two to five alternatives per choice. Most CBC’s contain eight to sixteen choice sets, with consumers needing motivation in the form of an incentive when being faced with more than twelve choice sets (Eggers et al., 2018). The ideal number of choice sets however depends on Fatigue effects.
Hair et al. (2010) developed an equation to calculate the minimal number of stimuli: The minimum nr. of stimuli = total nr. of levels across all attributes – nr. of attributes + 1. Which in this case means the minimum number of stimuli that has to be presented is ten. For the study, a choice design was constructed containing six choice sets, with three alternatives per choice. This means a total number of 18 stimuli were presented to the respondents. The number of choice sets is below the usual number, but still acceptable (Hair et al., 2010) to reduce fatigue effects that might arise from the complexity of the task.
Preferencelab was used to construct the CBC, the attributes and levels were randomized to create a design with minimal overlap. So the respondents were confronted six times with a tradeoff between three different choices. To increase realism a no-choice option was included, in a realistic setting consumers always have the option to opt-out (Eggers, Hauser, and Selove, 2016).
27 purchase intention an actual purchase situation has to be mimicked as good as possible. In a real life purchase situation consumers always have the option not to pick any of the options. Therefore the none-option is added to the research. However, the problem of the classical none-option is that consumers might choose this option often which, in the worst-case scenario, leads to no preferences at al. Therefore the dual response option is added to the design (Brazell et al., 2006; Wlömert & Eggers, 2016).
The dual response no-choice option first lets respondents select their preferred option. After that it is asked whether they would actually buy the product. This way respondents cannot opt-out of making a choice, and still provide data for the conjoint, even though they would not actually buy the product (Brazell et al., 2006; Wlömert & Eggers, 2016). When using a
traditional no-option, respondents might not only select the none-option when none of the choices is attractive, they might also select it when decision difficulty occurs (Brazell et al., 2006). According to Wlömert and Eggers (2016) using a Dual-response no-option
significantly increases the predictive accuracy of a CBC. Moreover, since a realistic purchase situation is created, choice becomes a better predictor of purchase intention. Appendix
3.3 Data Collection
The pre-test was constructed in Qualtrics and filled in by sixteen respondents. The pre-test gave input for the negative OCR levels and the product that was used for the actual survey. The actual survey was constructed on www.my.preferencelab.com, a sample of the questions, the scenario and the live link can be found in appendix C. The survey was constructed in Dutch to maximize the number of respondents, since the larger part of the author’s network is Dutch. The survey was distributed via multiple social media channels, meaning it was
28
3.4 Plan of analysis
3.4.1 Population and sampling method
After filtering out respondents that did not satisfy the attention check, and those that did not complete the survey, 183 respondents remained in the sample, n=183. Of this sample 80 (43,7%) were male, while 103 (56,3%) were female. The mean age of the respondents was 27,64, the youngest participant was 17 years old while the oldest was 72 years old. In graph 1 the distribution of age is visualized, the majority of respondents was aged 20 - 28. Since the survey was shared with peers and by peers, it makes sense that many students participated.
Figure 2: distribution of age.
The fact that many students participated in the study is also reflected in the education level of the respondents: 85,3% have a HBO degree or higher.
29
3.4.2 Validation moderators
The concept regulatory focus consists of multiple scales, which need to be tested on construct validity (Malhotra, 2010). The concept exists of 11 items, of which six form the concept promotion focus and the other five form prevention focus. First, to see whether it is appropriate to perform the factor analysis the Keyser-Myer-Olkin measure of sampling Adequacy (KMO) and Bartlett's test of sphericity are consulted. The Bartlett's test of sphericity is significant (P <0.001*), meaning some variables in the analysis are correlated. The KMO is 0.749, while it should have a minimum of 0.5. Finally all communalities should be >.4; this condition is satisfied except for item 11which has a communality of 0.350. However, deleting this item would mean a decrease in the KMO therefore it is left in. This means that the correlation between those variables can be explained by other variables. The principal component analysis was performed. The SPSS outputs can be found in Appendix B. All items load on the expected components, moreover all loadings should be >0.5; this
condition is satisfied as can be seen in table 4. The factor solution in the data complies with the assumed factor structure. Two factor scores, of which one summarizing the five
prevention survey items and one summarizing the six promotion survey items, will be used as the moderators.
To determine whether the factor is reliable Cronbach's alpha has to be inspected, this should be >0.6. For the factor measuring promotion focus the Cronbach's alpha is: 0.107. After consecutively deleting items 10, 3 and 7 a Cronbach's alpha of 0.633 is reached meaning both factors are now internally constant and reliable.
For the factor measuring prevention focus this condition is not satisfied (Cronbach's alpha = 0.471). The "Cronbach's alpha if item deleted" table, shows that deleting item 5 would lead to a Cronbach's alpha of 0.794, so this item is deleted from the factor.
30 Table 4. Factor + reliability analysis
Variables Item Loadings α α if item
deleted Promotion Compared to most people I am typically
unable to get what I want out of life.
I1= -.630 .107
I have often accomplished things that got me psyched to work even harder.
I3= .663 .412
I often do well at the different things I try. I7= .631 .633
When it comes to achieving things, I find that I don’t perform as well as I ideally would like to.
I9= -.674
I feel like I have made progress towards being successful in my life.
I10= .732 .236
I have found very few hobbies and activities in my life that capture my interest or motivate me to put effort into them.
I11= -.590
Prevention Growing up I would sometimes “cross the line” by doing things my parents would not tolerate.
I2 = .855 .471
When growing up I got on my parent nerves often.
I4 = .854
I always obeyed the rules and regulations that were established by my parents.
I5 = -.674 .794
Growing up, I sometimes acted in ways that my parents thought were objectional.
31 Not being careful enough has gotten me
into trouble at times.
I8 = .637
3.4.3 Model specification
The CBC will result in estimates for all attribute levels. The utility function assumes that goods or services are combinations of attributes, to which consumers attach path-worth utilities (Jun & Park, 1999). All attribute levels were effect coded, the reference level is coded as -1. LatentGold was used for modelling the data from the CBC.
Then, for each attribute a model is specified: since brand is nominal a part worth model is chosen. For all other attributes a part-worth model will also be chosen, when they are almost linear or quadratic the model is changed and the model fit will be compared. However, since the focus is on in all levels of Price promotions and OCRs a part-worth model will always be used for these attributes.
3.4.4 Model fit and validation
For the assessment of the model fit, the Loglikelihood will be calculated and compared to the NULL-model. To see if the estimated model performs better than the NULL-model, and therefore better than chance, the likelihood ratio test is performed, in which the Chi-square is the test statistic, h0 should be rejected, meaning the estimated model performs significantly better than zero. The fit will be assesed using the Likelihood Ratio test, the pseudo R² and the adjusted pseudo R². Moreover, the BIC and CAIC (Yang, 2006) will be used to compare the models to each other, since these measures account for the number of parameters in a model. Afther an aggregate choice model is estimated, all moderators will be added, and checked on significancey. If not all moderating effects are significant, the most insignificant variable will be deleted from the model and a new model will be estimated. This process is repeated untill al moderating variables are signifcant.
32
4. Results.
In this chapter all relevant results are analyzed and described using the analyzation methods described in the previous chapter. First the right model will be specified, thereafter the preferences on aggregate level will be analyzed, followed by the preferences resulting from the latent class analysis.
4.1 Model specification
First, the aggregate model fit will be determined. All attributes and levels will be added for an aggregate estimation, hereafter the model will be specified since attributes might have a linear, quadratic or part-worth function. First a nominal (part-worth) distribution is assumed for all attributes. When looking at figure 4a and 4bm it becomes clear that attributes Price Promotions and OCRs might be linear, the same holds for zoom and megapixels. However, the goal of this study is to find the point where utility of a negative OCR at a certain level of negativity could be outbalanced by a price promotion of a certain level. When using linear models for OCRs and Price promotions only one utility is given for each attribute. Therefore, both OCRs and price promotions will be estimated using a part-worth model. The same holds for brand, which is always a part worth model. For the results however, the linearity of price promotions is a good thing, because this way a specific level of price promotions can be calculated to outbalance the negative effect of negativity from a negative OCR.
Figure 4a Price promotion distribution & Figure 4b distribution of OCRs
Two models are specified in which Brand, Price promotions and OCR's are always assumed to be nominal. For the other attributes a model in which they are assumed to be nominal is compared with a model in which they are assumed to be linear. The models are compared to
33 determine the best model fit. Model 1: all attributes are nominal, model 2: Zoom and
Megapixels are linear, the other variables remain nominal. Table 5, aggregate model specification
Model 1 Model 2 LL Model -942.26 -942.26 LL Null* -1206.27 -1206.27 Npar 11 9 N 183 183 LR test** 528.03 528.03 P(chisq) LR test <0.01 <0.01 Pseudo R²*** 0.2188 0.2188 Adj. Pseudo R²**** 0.2097 0.2097 BIC 1941.81 1931.39 CAIC***** 1952.82 1940.45 Hit rate 61.38% 61.38% *LL Null = n*c*ln(1/m) **LR test = -2*(LL(0)-LL(ß) ***Pseudo R² = 1 - LL(ß) / LL(0)
****Adj. Pseudo R² = 1 - (LL(ß) - npar) / LL(0)
34 Table 5 shows that both models perform significantly better than the NULL-model with Chi-square distributions of 528,03 with 11 and 9 degrees of freedom (p=<0.001), meaning both models predict better than chance. The models perform pretty similar, with model 1 and 2 not differing on any of the criteria. A pseudo R² between 0,2 and 0,4 is generally considered acceptable, this is the case for both models. The same can be said about the hit rate, which indicates a good degree of predictive validity (Papies, Eggers & Wlömert, 2011). For this reason the model with the lowest BIC score is chosen, which is model 2 .
4.2 Aggregate model main effects
Table 6 shows the estimated utilities of the aggregate model. The p-values show that all main effects are significant, except for brands Canon and Nikon. The model had a hit rate of 61,38% with a prediction error of 0,38; this is considered acceptable (Papies, Eggers & Wlömert, 2011). The main effect of brands Canon and Nikon are not significant, meaning brand does not significantly change the utility/preference of consumers. Moreover, the relative importance column shows that negative OCRs have the highest relative importance: on aggregate this means that negative OCRs, instead of price promotions have most impact on the preferences of consumers. The difference however, is very small, and might not be
significant. However, since the choice design was not balanced, it is not possible to perform a t-test to confirm whether the difference between the Relative importance of Price promotions and OCRs is significant.
Table 6 aggregate preferences
35 OCRs 1.348 34.06% Slightly negative 0.6222 202.14 <0.001 Negative 0.1041 Very negative -0.7263 Canon 0.0885 3.35 0.067 Nikon 0.0840 3.05 0.080 Sony -0.1725 0.261 6.6% Megapixel 18 -0.1642 19.56 <0.001 0.3284 8.3% 20 0.1642 Zoom -0.3387 81.81 <0.001 0.6774 17.11% 0.3387 Sum: 3.958
36
4.3 Aggregate model moderating effects
To test the moderating effects of the regulatory focus scales, interaction effects between the created prevention and promotion variables and all attribute levels for price promotions and the OCRs are added to the model. First, a model is estimated which entails the before determined model including all moderators. After that, when there are any insignificant estimates, the least significant estimate will be taken out. This only entails moderating
variables as the main effects should always remain in any model. This backwards elimination will continue on until the right model fit is found and all effects are significant.
Table 7 model fit when moderators are added
37 Model 6 7 8 9 10 LL -931.66 -931.74 -932.37 -933.62 -934.35 LL null -1206.27 -1206.27 -1206.27 -1206.27 -1206.27 Npar 15 14 13 12 11 LR test 549.22 p<0.01 549.06 p<0.01 547.8 p<0.01 545.3 p<0.01 525.84 p<0.01 Pseudo R² 0.2276 0.2275 0.2270 0.2260 0.2254 Adj. Pseudo R² 0.2152 0.2159 0.2162 0.2160 0.2163 BIC 1941.46 1936.41 1932.47 1928.62 1926 CAIC 1956.46 1950.41 1945.46 1941.75 1937 Hit rate 61.84% 61.84% 61.84% 62.48% 62.02%
Model 1 is the aggregate model that has been used for the aggregate estimations. Model 2
contains all moderators, since all but one were insignificant, more models had to be estimated. To least significant estimate is deleted, and a new model is estimated. The models can be assessed in a similar way to how this was done in chapter 4.3. All models performed significantly better than the null model (P<0.01) meaning they predict better than chance. Both the R², the adjusted R² and the hit rate are acceptable for all models too. So once more it comes down to the BIC. Model 1 and model 10 have the lowest BIC scores, as expected. Out of the ten moderating variables, 2 are found to be significant.
38 Table 8 Estimates model 10
Class 1 Wald p-value Range Relative
39 (8.4%)** 0.3293 19.4471 <0.001 Zoom 0.6959 17.31% (17.79%)** 0.6959 84.8751 <0.01 Sum: 4.02 (3.91)** Moderator Prevention x negative* -0.1497 9.7331 0.0018 Moderator Promotion x slightly negative** -0.1026 5.4485 0.02
Regulatory focus determines the preferences and relative importance of attributes and levels. Therefore utilities are provided on three levels, when there is no moderation, when consumers are prevention focused* and when consumers are promotion focused**.
All main effects remain significant, with OCRs being slightly more important than price promotions: 34.32% versus 33.75%. Moreover, the following can be said about the main effects: when there is no price promotion, consumers will not choose any product: the negativity of OCRs are too high. The negativity of a slightly negative OCR can be balanced out by the positivity of a -10% price promotion. A 25% price promotion is sufficient to outbalance the negativity of a negative review, however a price promotion of -50% is not sufficient to even out the negativity of a very negative review.
-40 0.1497 = -0.035. This would indicate that a consumer that is prevention focused would need a higher price promotion to balance out the negativity of a negative OCR.
The other moderator that was found to be significant states that promotion focus moderates slightly negative reviews. This means consumers with a promotion focus the effect of a slightly negative review becomes more negative in this case: 0.6336 – 0.1026 = 0.5310. This decrease does however mean that the relative importance of OCRs from changes 34.32% to 32.60% while the relative importance of price promotions increases from 33.75 to 34.63%, making it more important than OCRs.
4.4 Latent Class segmentation
The aggregate choice model estimated the preferences for the average consumer that is shopping online for a digital camera. This model gave great insights, but does not account for heterogeneity among consumers. A latent class analysis does account for consumer
heterogeneity. The aggregate choice model in which no moderator effects were included will serve as the marker. First, the optimal number of segments has to be found, posteriori
segmentation will be used to determine the number of segments or "classes". Six different models have been estimated: the BIC and the CAIC are critical measures in determining the number of classes and are therefore added (Andrews & Currim, 2003).
Table 9 Latent class model fit
Model 1 (1-class) Model2 (2-class) Model 3 (3-class) Model 4 (4-class) Model 5 (5-class) Model 6 (6-class) BIC 1931.39 1903.87 1910.43 1924.24 1936.77 1958.37 CAIC 1940.39 1922.87 1939.43 1963.24 1985.77 2017.37 Class. Error 0 0.0668 0.1379 0.1492 0.1558 0.1350
BIC and CAIC are the best indicators on which model performs best, when looking at table 9
41 different segments. The segments differ significantly from each other in their preference for price promotions (Wald = 16,65, p<0,001), OCRs (Wald = 44,83, p<0,001). The difference in preference is only significant at the 10% confidence level, while differences on preferences for megapixels and zoom are insignificant.
4.4.1 Moderation segment 1
The moderating roles of prevention focus and promotion focus are measured in the same way as in the aggregate model. First adding all moderator interactions, then deleting the least significant one. This time however, this process has to be repeated twice, first focusing on segment 1 and thereafter on segment 2.
Table 11: moderation segment 1
Model 1 2 3 4 5 LL -868.50 -868.63 -869.07 -870.01 -871.81 Npar 39 37 35 33 31 BIC 1940.16 1930.01 1920.48 1911.95 1905.13 CAIC 1979.17 1967.01 1955.48 1944.94 1936.13 Hitrate 69.22% 69.58% 69.76% 69.49% 69.58%
After deleting four moderating variables, both the main effects and the moderating effects are significant. Moreover, all models perform significantly better than the null model, and the BIC decreases each time an insignificant moderator variable was deleted from the model. The preferences for segment 1 when all significant moderators are added can be found in Appendix D, table D2. A detailed elaboration on the outcomes, and what they mean is given here.
42 Not taking the moderators into account yet, the following can be concluded from the model. First, a -25% price promotion is necessary to balance out the negativity of a slightly negative or negative OCR. Moreover, a -50% price promotion was sufficient to outbalance the
negativity of a very negative OCR.
There were seven significant moderating effects, of which, there were four interactions between promotion focus, price promotions and OCRs. Also, there three interactions between prevention focus, price promotions and OCRs.
For prevention focused consumers, the utility of no price promotion became more negative, which has impact on the relative importance of OCRs. The utility of the -10% price
promotion becomes more positive while effect of a negative OCR also becomes less negative. A -10% price promotion is still not sufficient to balance out the slightly negative OCR, but the utilities are very close to each other, which would indicate a price promotion of slightly more than -10% could be enough. The relative importance is also affected by the moderating effect of prevention focus: price promotions become more important: 46.06%, while OCRs also become a little more important: 21.2%.
For the promotion focused consumers in segment 1 the utility of a -50% price promotion becomes more positive. The utilities of both slightly negative and negative OCRs become more negative, while the opposite holds for a very negative OCR: the utility becomes more positive. This means a -25% price promotion is sufficient to outbalance the negativity of OCRs on all levels. The relative importance of OCRs is lower for promotion focused consumers.
43
4.4.2 Moderation segment 2
The same procedure was used to determine the best model, taking the main effects + all moderating effects as the starting point. After that, the most insignificant moderator variable is deleted, this process is repeated until all moderators are significant.
Table 12: moderation segment 2
Model 6 7 8 9 10 11 LL -868.71 -869.63 -876.41 -876.76 -878.48 -880.54 Npar 37 35 33 31 29 27 BIC 1930.18 1921.60 1924.74 1915.02 1908.03 1901.74 CAIC 1967.18 1956.60 1957.74 1946.02 1937.03 1928.74 Hitrate 69.22% 69.13% 69.13% 69.31% 68.4% 68.03%
After deleting six insignificant moderating variables, model 11 shows the best fit, with the lowest BIC and CAIC, all hit rates are acceptable, so are the R²'s. The hit rates are higher for models with more parameters, the BIC and CAIC penalize the number of parameters, giving a better indication on what model to use.
Table D3 in Appendix D depicts the utilities of model 11 along with the significant moderating effects. In this paragraph the main outcomes will be discussed.
For segment 2, the main effects of brand and megapixel are insignificant, meaning they do not significantly influence the preferences of consumers. The other main effects are significant, as are four different interaction effects between regulatory focus and price promotions and negative OCR’s. First, the main effects for segment 2 will be discussed. For this segment the relative importance of OCRs is very high; 49.05 percent, meaning the preference for a product is determined for almost 50% by OCRs. The relative importance for price promotions is 20.52%.
44 negative price promotion however, is to negative. A 50% price promotion is not sufficient, consumers will not choose this product.
Prevention focus interacts with two different price promotions and one OCR. First, for consumers that are focused on prevention the utility of no price promotion becomes less negative, resulting in a decrease in the relative importance of price promotions to 13,96%. This also means that a negative OCR can be outbalanced without using a price promotion, which is remarkable. Furthermore, the utility of a -25% price promotion becomes less positive.
For consumers that are promotion focused within segment 2 there is only one significant interaction effect: the utility of a slightly negative price promotions becomes more negative meaning the negativity can no longer be outbalance without using a price promotion, a -10% price promotion is need for consumers that are promotion focused. The relative importance of price promotions increases: price promotions are more important for consumers that are promotion focused.
64% of segment 2 is female while 36% is male. This segment is one year younger on average as compared to other segment. 90% of the segment has an education of HBO, or higher. With a population of only 45 respondents this segment is a lot smaller as compared to segment 1.
4.5 Purchase intention
45 Therefore, the dual-response none-option was added (Brazell et al., 2006). Respondents were forced to select their preferred data, and then indicated whether they would actually buy the product or not. The software that was used for estimation (Latent Gold) however, does not offer the possibility to analyze the dual-response none-option.
To still get an indication on the utility of the none option the data file that was used for the initial estimations had to be transformed. Originally, each choice set had three options to choose from. A fourth option was added which would serve as the none choice. After that, the entire data set had to be transformed: effect and dummy coded attributes were set to “missing” whenever the none choice was chosen. For the aggregate model that was used for the prior estimations, the none option had a utility of 1.863 which could only be reached with a slightly negative or negative OCR in combination with a 50% price promotion. This does not mean a lower price promotion is never sufficient. The aggregate model does not take consumer heterogeneity into account. Investigating the data set tells that the high utility was to expect since almost 81% of all respondents chose no when asked whether they would buy the product. A remarkable finding when looking into the data set is that of the consumers that choose “yes”: the highness of a price promotions did not really matter:
Table 13: none-option Yes No None 4.5% 81% -10% 4.8% -25% 4.7% -50% 5%
46
4.6 Hypothesis Overview
Table 14, hypothesis overview
Hypothesis Supported
by
Aggregate model
Segment 1 Segment 2
H1 The more negative an OCR is, the more negative the effect on purchase
intentions becomes. Accepted (Choice as the DV) Accepted (Choice as the DV) Accepted (Choice as the DV) H2a H2 The effect of a negative OCR on
purchase intentions becomes more negative (positive) when consumers are prevention (promotion) focused.
Rejected (only 1/7 sig. + right direction) Rejected (4/7 sig. , one in the right direction) Rejected (only 1/7 sig. right direction)
H2b The relative importance of OCRs increases (decreases) when consumers are prevention (promotion) focused.
Partially Accepted
Accepted Accepted
H3a The effect of a price promotion on purchase intentions becomes less (more) positive when consumers are prevention (promotion) focused. Rejected, not significant Rejected (only 2/7 sig. + right direction) Rejected (only 1/7 sig. + right direction) H3b The relative importance of price
promotions increases (decreases) when consumers are promotion (prevention) focused. Partially Accepted Partially Accepted Accepted
47
H5 A price promotion can be used to outbalance the negative of a negative OCR.
48
5. Discussion
The aim of this study was to investigate to what extent price promotions can be used to outbalance the negative effect that a negative OCR has on purchase intentions. To investigate differences between consumers, the regulatory focus that consumers have; either promotion focus or prevention focus was used. A Choice-based-Conjoint analysis was performed to measure the utilities of consumers when shopping online. This way, both the utilities of price promotions (on various levels) and negative reviews (differing in valence) were obtained and could be compared. First, an aggregate logit model was estimated to investigate the
preferences of consumers on an aggregate level. After that a Latent class analysis was performed, resulting in two different segments of consumers, based on similarity in preferences. This chapter discusses the main findings.
5.1 Main effects
The findings in this study only partially match with conclusions that were drawn in previous research. Conclusions on short term elasticities of price, price promotions and OCRs (e.g. Floyd et al., 2014; You, Vadakkepatt & Joshi 2015; Bijmolt, van Heerde & Pieters, 2005 Aaker, 1992) were the baseline to perform this analysis, indicating that price was the main reason for consumers to choose a product. The aggregate model and the model for segment 1 confirm this, with price promotions having the highest relative importance. For the second segment however, this is not the case; these consumers find OCRs to have a higher relative importance. The negativity bias (e.g. Maheswaran & Meyers-Levy, 1990; Skowronski &. Carlston 1989; Lee, Park & Han, 2008) indicating that the more negative an OCR becomes, the more negative the effect becomes, is supported by all three models.
The main objective of this study was to investigate whether negative OCRs, on different levels of negativity, could be outbalanced with a price promotion. In other words, finding a point at which the negativity of the risks associated with negative OCRs, are balanced out by the positivity of the potential gains that price promotions offer (Sandes & Urdan, 2013). If the potential gain that is indicated by the price promotion is high enough, the potential risk that is indicated by the negative OCR will be ignored (e.g. Sweeney et al., 2010) All three models confirm that the height of a price promotion is positively related to choice.
