The effect of price on the hypothetical bias in conjoint analysis
Exploring the relative effects of price on the consumers’ choice
By
Rob Valkenburg
The effect of price on the hypothetical bias in conjoint analysis
Exploring the relative effects of price on the consumers’ choice
By
Rob Valkenburg
University of Groningen
Faculty of economic and business
Master’s thesis MSc Marketing Intelligence
(Final version)
Completion date: January 14, 2019
Celebesstraat 26-B
9715JG Groningen
+31 6 152 279 87
PREFACE
More than two years ago, I enrolled for the Pre-MSc Marketing. I wanted to challenge myself, I was motivated to learn and improve my knowledge. However, I could not have imagined how I would improve myself as a person. The years at the University of Groningen have challenged me to perform on a high level, improved my knowledge, changed my attitude and helped me becoming the person I am today. This master’s thesis is my final project, I am going to miss my time here, but thanks to the University of Groningen, I am ready to take the next step.
For helping me writing this thesis, I want to thank my supervisor dr. Felix Eggers for providing me with very useful feedback and for challenging me to get the most out of this research. Secondly, I want to thank my fellow group member, Ji Xiao for his help and for providing a pleasant working environment during group meetings. Last, but definitely not least, I want to thank my friends and family for their help, motivation and support during my entire student life. Thank you.
Looking back at myself, I am proud of the achievements I have accomplished and the route I have taken. Starting almost ten years ago with a marketing study on intermediate vocational level, and by climbing the ladder, now finishing my master’s study. This master’s thesis is my final project of the study Marketing Intelligence at the University of Groningen. This is my final project, this is my accomplishment.
MANAGEMENT SUMMARY
Finding out the wishes and ideas of customers about products is an important reason for carrying out market research. Conjoint choice experiments are market techniques that enable researches to estimate these consumer preferences by analyzing which product choices consumers make. Within these conjoint choice experiments, the price attribute is an important factor. In combination with other added product utilities, the maximum willingness-to-pay can be assessed, or the maximum revenue can be calculated. However, conjoint choice experiments are hypothetical settings in which the outcomes are different from real purchase data. This deviation from real-life purchase data is referred to as the hypothetical bias. Research, however, shows that the WTP is commonly overstated due to this hypothetical bias with a factor of two or three, leaving the question of how relevant the inclusion of price really is. However, the hypothetical bias can be reduced by carrying out incentive-alignment with the study. With incentive-alignment, an experiment is created in a way to elicit “natural actions” from the respondents to help reducing the hypothetical bias. This thesis goes deeper into the effects of price on the hypothetical bias within conjoint analysis.
A choice-based conjoint analysis among 1,200 participants was conducted in December 2018 to provide evidence for the proposed research question. In this experiment, respondents were randomly allocated into four groups (price included vs – excluded and incentive-aligned vs hypothetical) and asked to choose their favorite book or eBook with both different attributes and levels.
TABLE OF CONTENTS
1. INTRODUCTION... 1
1.1 Conjoint analysis ... 1
1.2 Incentive Alignment ... 2
1.3 Problem statement and research question ... 3
2. THEORETICAL FRAMEWORK ... 5
2.1 Incentive-alignment as a moderating effect ... 5
2.2 The exclusion of price ... 7
2.3 The exclusion of price in incentive-aligned conditioning ... 9
2.3 Conceptual model ... 10
3. METHODOLOGY ... 11
3.1 Sample and design ... 11
3.2 Procedure and experimental design ... 12
3.3 Plan of analysis ... 14
4. RESULTS... 16
4.1 Model specification ... 16
4.2 Aggregate model main effects ... 18
4.3 Aggregate model moderating effects ... 19
4.4 Results of hypothesis testing ... 19
5. DISCUSSION ... 22
5.1 Summary of findings ... 22
5.2 Theoretical contributions ... 23
5.4 Limitations and further research directions ... 24
REFERENCES ... 26
1
1. INTRODUCTION
1.1 Conjoint analysis
An important reason for carrying out market research is to try to find out what the wishes and ideas of consumers are about products, i.e., to measure consumer preferences. Conjoint choice experiments are a market research technique that use stated choices to estimate consumer preference. Within conjoint choice experiments, consumers are not asked directly what they consider important in a product, but researchers infer the importance by analyzing which choices consumers make.
With the output of the research, the required analytics determine which product attribute specifications are most preferred and what the relative importance is for each attribute. However, the strength of the method is mainly in the prediction of the complete desirability of all possible product compositions. In the analysis phase, optimal product configurations can be sought; configurations that are preferred by the largest group (Sandor & Wedel, 2001). Such an option model provides insight into the preference structure of respondents and provides the marketing manager with information regarding new product development, pricing, branding, market segmentation or market scenarios. (Grewal & Liechty, 2005).
In a conjoint analysis, the inclusion of price as a specification is an important factor. In combination with other concrete product specifications, the optimum can then be determined between price (what is the end user willing to pay) and revenue (what is maximum feasible in the new situation). Assuming that the new offer is optimally adjusted to the wishes of the end user.
There is a growing interest in establishing the WTP and the utility of different attributes from a product from real market settings. There are many studies using these “revealed” preference data, e.g., using scanner data. However, the majority of tests are hypothetical and incentive-aligned experiments. The problem here is that real purchase data are different from hypothetical- and incentive-aligned data. This deviation from real-life purchase data is referred to in the literature as the hypothetical bias (Hensher, 2010).
2
The effect of the hypothetical bias on price sensitivity and, consequently, WTP is well-established. However, not much is known about the effect on other product attributes except price. The existing research regarding this topic is limited and although they did find some interesting results, they did not account for several factors that might influence the outcome of their results. This thesis goes deeper into this research gap which will be further explained in the theoretical framework.
The mentioned research above presents the impact and effect of the hypothetical bias on the WTP. However, since this thesis elaborates on the effect of price on other product attributes except price, focusing merely at the WTP subject of the hypothetical bias is less relevant. Hence, for the remaining of this thesis, the hypothetical bias is defined as any change in attribute importance between hypothetical and non-hypothetical, incentive-aligned questions.
1.2 Incentive Alignment
According to the induced value theory of Smith (1976), providing incentives for the respondents to act in accordance with one’s true preference is essential, if possible. His theory suggests that more realistic incentive-aligned studies results in a stronger predictive performance of actual purchase behavior than hypothetical studies. As stated in Smiths’ theory, salience is the most relevant condition within the case of conjoint analysis. This requires that the incentive must be directly related to the decisions the respondents make during the research. In many studies, respondents receive a fixed amount of money for participating to motivate and increase the number of respondents. However, this is not salient, since there is no relationship between the payed amount of money and the decisions the participant made. This will result in a significant difference in the respondents’ behavior during the research and his or her behavior during a similar, real-life situation (Ding, Grewal and Liechty, 2005).
3
respondents that correspond to their actions in the marketplace and with that reduce the hypothetical bias.
According to Ding (2007), incentive-aligned conditions will result in a much higher price heterogeneity than in a hypothetical setting. The results reported of a price heterogeneity of 5.60 in the incentive-aligned condition and .90 in the hypothetical conditioning. Participants were also more price sensitive in the incentive-aligned conditioning (-4.18) than in the hypothetical setting (-2.61).
These analyses mentioned above show that an incentive-aligned approach should be a more preferable choice for researchers and marketeers as long as it does not become too expensive to implement. However, it is not always possible to implement an incentive-aligned setting due to legal restrictions or availability of the products prototype (Miller, Hofstetter, Krohmer and Zhang, 2011).
1.3 Problem statement and research question
There are several studies which showed that the inclusion or exclusion of the price attribute, effects the utilities of the other attributes regarding risk seeking and willingness to try new things. (Bryan and Parry, 2002; Camerer and Hogarth, 1999; Carlsson, Frykblom and Lagerkvist, 2007).
The structural reliability of conjoint analysis has been a topic in many marketing studies, i.e. if and how the preferred utilities are affected by the design of the conjoint analysis. Many studies regarding conjoint analysis have found that the utilities for a particular attribute are not affected by the number of levels or attributes (Reibstein et al., 1988; Malhotra, 1982; McCullough and Best, 1979). A study by Hensher (2006) looked at the differences in WTP depending on the design dimensions themselves and research the effect of, among other things, the number of attributes. The overall conclusion of the study is that there was no systematic difference in the mean WTP among the different stated choice designs.
These findings are contradictory with the earlier mentioned findings regarding the effect of excluding or including the price attribute. Meaning the opinions are uncertain and there is no clear conformity regarding this research topic.
4
information about the price people are willing to spend for the product, the price sensitivity of different products, it creates an important tool for welfare analysis and even helps marketing managers how much a person is willing to spend for a single attribute of the product. However, research shows that WTP is commonly overstated due to the hypothetical bias with a factor of two or three (Similarly, List and Gallet, 2001). For management, expecting the outcome of the WTP to be true can be problematic, since chances are that the product or service that is offered will be overpriced by up to three times the actual WTP. Leaving the question of how relevant the inclusion of price really is, since it is constantly overrated. This thesis goes deeper into this research gap and aims to answer the following research question:
RQ: What is the effect of price and incentive alignment on consumers’ choices
Specifically, in this study, the Choice-Based Conjoint analysis (CBC) was used with four conditions. A hypothetical- or an incentive-aligned condition and two conditions where price is included or excluded (Louviere and Woodworth, 1983). Within the CBC, the respondent will repeatedly see a certain number of product designs side by side with different attributes and the respondent must select their most preferred choice every time. The power of a CBC is that it imitates a real purchase situation process, where consumers compare different products of the same category with each other and choose their most preferred option (Eggers and Sattler, 2011).
5
2. THEORETICAL FRAMEWORK
As shown in table 1, it is a common good to include price and an incentive alignment in studies regarding stated preference or conjoint analysis. Only Carlsson, Frykblom and Lagerkvist (2007) report of the effects of the price attribute on behavior in stated preference surveys, meaning that the topic has not yet been properly studied. Although Carlsson et al. went deeper into the effects of the price attribute on stated preferences and reported of significant differences in preferences with and without the price attribute included, they did not include the moderating effect of the hypothetical bias, nor did they compare an incentive-aligned- with a hypothetical setting to reduce the hypothetical bias.
This research gap is further explained in the theoretical framework and additional background information is provided regarding incentive-alignment and the effects of excluding price. This will ultimately result in several hypothesis. The theoretical framework presents a more detailed presentation of the concepts that were introduced in the introduction.
Table 1, studies regarding conjoint analysis with price included
Author Year Topic
* Method Comparing incentive-aligned with a hypothetical setting (y/n) Price included Effect of excluding price (y/n)
Carlsson, Frykblom & Lagerkvist
2007 Preferences with and without price - does the price attribute affect behavior in stated preference surveys?
CA No Yes Yes
Ding 2007 An incentive-aligned mechanism for Conjoint Analysis CA Yes Yes No
Ding, Grewal & Liechty 2005 Incentive-Aligned Conjoint Analysis CA Yes Yes No
Dong, Ding & Huber 2010 A simple Mechanism to incentive-align conjoint experiments CA Yes Yes No Miller, Hofstetter,
Krohmer & Zhang
2011 How should consumers' willingness to pay be measured? CA, RE, OE, BDM
Yes Yes No
Wlömert & Eggers 2016 Predicting new service adoption with conjoint analysis CA Yes Yes No Johansson-Stenman &
Svedsäter
2008 Measuring hypothetical bias in choice experiments OE Yes Yes No
Vecchio & Annunziata 2014 The effect of monetary versus nonmonetary endowment on WTP in BDM auctions
BDM No Yes No
Rob Valkenburg 2019 The effect of price on the hypothetical bias in conjoint analysis CA Yes Yes Yes
*CA = Conjoint Analysis; RE = Real purchase data; OE = Open-ended; BDM = Becker, DeGroot, and Marschak's
2.1 Incentive-alignment as a moderating effect
6
how it should be measured by comparing, among other things, the difference in an incentive-aligned- hypothetical setting and real purchase data. They observed steeper slopes in price parameter for incentive-aligned methods than for hypothetical methods. Indicating that respondents are more price sensitive in an incentive-aligned setting and tend to overstate their WTP in a hypothetical setting.
According to Harrison and Rutström (1999), respondents usually tend to overstate their true valuation in hypothetical settings. Since respondents do not genuinely have to pay for the product, they are less price sensitive so that they overestimate their WTP to increase the chance of hypothetically obtaining the product.
This overstatement is a consistent finding in many research topics regarding the hypothetical bias. Harrison and Rutström (1999) found that 34 of 39 stated preference observations had an average hypothetical bias of about 338 percent. Similarly, List and Gallet (2001) report that respondents in hypothetical situations tend to overstate their preferences by a factor near three and Ash, Murphy and Stevens (2004) report of the NOAA panel recommendations, suggesting the rule of thumb that hypothetical values should be divided by two. This is in line with earlier mentioned studies who found the same results (Ding, Grewal and Liechty, 2005).
Cao and Zhang (2018) describe effort as an explanation of why this consequent overstatement occurs. They report that consumes must make a costly effort to learn their true product valuation. I.e., consumers may spend time by researching about the different product attributes, investigating alternative products or thinking about how to use the product in real-life. Therefore, they explain that it must by likely for a respondent to obtain the product, otherwise they will have little incentive to uncover their true product valuation. Their results show that consumers are becoming more price sensitive when the probability of obtaining the product increases.
Concluding, respondents’ attitudes regarding the price attribute and their WTP are expected to change in an incentive-aligned setting compared to a hypothetical setting. Resulting in the following hypothesis:
H1a. In an incentive-aligned conditioning a consumer places more value on price and
7
Due to this predicted increase in relative measured price value, the expectations are that in an incentive-aligned conditioning, consumers place less value on the other attributes. Miller et al. (2011) noted in all hypothetical methods significantly different results regarding the product choices consumers made, compared to their incentive-aligned counterparts. Indicating a clear hypothetical bias for the measured preference for the product attributes. Moreover, as mentioned before, Ding (2007) reported that participants exhibit lower risk seeking and a lower willingness to try new things among the respondents in the hypothetical setting compared to the incentive-aligned setting. Camerer and Hogarth (1999) found that incentive-alignment help reducing overly socially desirable presentations of oneself and by that, create a more realistic situation in comparison to a hypothetical setting. However, since this study focusses on innovative features and not on socially desirability, there is no hypothesis about socially desirable aspects. These mentioned findings show that incentive-alignment creates different values, and as expected more trustful outcomes regarding the measured attributes. Resulting in the following hypothesis:
H1b. In an incentive-aligned conditioning a consumer places less value on innovative product
attributes. 2.2 The exclusion of price
Stated preference analysis and choice experiments are important tools to measure the WTP and elicit preferences for various characteristics of products or services. To be able to measure the WTP, a price attribute with varying levels is necessary. An assumption from the mainstream economy is that the preferences for individual attributes must be independent whether the price attribute is included or not. This behavior is also expressed by the three central assumptions of neoclassical economics; where people have rational preferences in the choice between outcomes, individuals maximize utility, act independently based on perfect information, and individuals should have clear and well-defined preferences that should not be influenced by the price of the product (List, 2002).
8
included and 3rd when a price variable is excluded). Unfortunately, there results do not give
an exclusion why these changes occur. They suggest that it might occur from strategical thinking from the respondents. Where they are expected to focus on the price attribute in order to avoid future price increases of the goods they are estimating. Another possibility they propose is a possible cognitive overload when the price attribute is introduced next to the other attributes. Lastly, they argue about another equally plausible explanation for the differences, namely that the differences between the price attribute and the other attributes were substantial enough. Likewise, a study by Harrison (2006) concerning private good experiments, showed that respondents’ may be influenced by the prices of similar, already familiar goods which influences their choices even when there is no price attribute included. Dodds, Willimam, Monroe and Grewal (1991), report in their paper of a positive effect of price on perceived quality, but a negative effect on the perceived value and willingness to buy. Their findings showed that, when the price level increases, the risk of an incorrect assessment increases. Meaning that the price attribute as a measurement of quality becomes less relevant compared to the assessment of the other attributes. Hence, for higher priced product that are purchased infrequently, the strength of the price attribute lowers, relative to the other, more well-known attributes.
The price attribute is also, by definition, different than the other attributes since it has no relation towards the design or the functional attributes of the product. Therefore, the behavior of the respondents is differently influenced by the price attribute. Some respondents can focus too much on the price of the product, expressing an attitude that they would not want to pay for a certain improvement and in that way act strategically with respect to the future price of the product. On the other hand, certain respondents might not pay that much attention to the price attribute, since they are more concerned about their preferences concerning the good itself and not the cost (Martilla and James, 1977).
The results of the Carlsson et al. (2007) study show that differences arise when price is excluded. However, it begs the question: which of the two preferences, if any, are valid? Their study did not provide an answer to this; however, the answer is important. Their results high-light a potential problem when preferences are extracted.
9
should not make a difference in the respondents’ choice is the broader accepted theory. Resulting in the following hypothesis:
H2. The exclusion of the price attribute does not change the respondents’ relative importance
of the remaining attributes.
2.3 The exclusion of price in incentive-aligned conditioning
Several studies confirm that price is positively related to quality and negatively related to the perception of value and willingness-to-buy (WTB) (Dodds and Monroe, 1985; Gardner, 1971; Beneke, Flynn, Greig and Mukaiwa, 2013). Hence, an increase in price reduces the perceived value of the product and the consumers’ WTB and a decrease in price will have the opposite effect.
As mentioned before, the Carlsson et al. (2007) study was the only study that measured the effects of the price attribute – and the exclusion of this attribute – on behavior in stated preference surveys. However, their study had some limitations. As according to the induced value theory and Ding et al. (2005), since the study did not provide the experiment with an incentive-alignment, their results will show significant difference in the respondents’ behavior during the research and his or her behavior during a similar, real-life situation. Ding et al. (2005) showed that data collected in hypothetical settings have weaker external validity than data collected within an incentive-aligned study. Hence, an incentive-alignment is necessary in an experiment to elicit “natural actions” from the respondents, to reduce the hypothetical bias, and to create improved valid results.
Concluding, as mentioned before, according to the mainstream economics, preferences for individual attributes should be independent of one another and price influences the perceived value and the consumers’ WTB. Combining these theories in an incentive-aligned conjoint analysis to reduce the hypothetical bias while leaving out the price attribute should result in the following hypothesis:
H3a. When leaving out the price attribute in an incentive-aligned conditioning, the
10
Miller, Hofstetter, Krohmer and Zhang (2011), reported of a much higher number of “no-choices” in the incentive-aligned setting (19%) compared to the hypothetical setting (5%). This indicates that within the hypothetical setting, participants behave as if they were interested in purchasing a particular product but behave differently than they would do in a real purchase decision in the incentive-aligned setting. Regarding the earlier mentioned theories confirming the negative relation of price with perceived value and WTB, a change in price would lead to a change in the consumers’ choice regarding the “no-choice” option. On the other side, when excluding the price attribute, there should not arise any effect since the WTB cannot be affected by price. Resulting in the following hypothesis:
H3b. When leaving out the price attribute in an incentive-aligned conditioning, the
consumers’ choice does not differ regarding the no-choice option.
2.3 Conceptual model
The mentioned hypothesis give more insights on the effect of price on the hypothetical bias in conjoint analysis. The following model (figure 1.) shows the relationship between the product- and price attributes (IV), and the respondents’ choice (DV). In this model, incentive-alignment functions as the moderating effect and the dotted outline around the price attribute indicates the situation where price is excluded as a variable.
Figure 1:
11
3. METHODOLOGY
This chapter explains the choice of the research and elaborates on the research procedure and how the data was collected. Furthermore, it presents the conjoint design and elaborates on the measurements and the data analysis.
3.1 Sample and design Sample
The sample consists of fictive data provided by Dr. Felix Eggers. The data was supposed to be generated through a survey and distributed among customers of MacMillan Publishers. Unfortunately, through a delay in the data collection, there was a risk of running into time-shortage. Hence, there is decided by Dr. Felix Eggers to continue this research with fictive data. For the remaining of this study, the data will be interpreted as real data.
The data consists of 1,200 respondents equally divided into groups of 300 over the four conditions; a hypothetical- or an incentive-aligned condition and two conditions where price is included or excluded.
Design
12
Table 2, Attributes and levels
Attributes: Levels: Innovative yes/no:
Format: (Printed) Book No
eBook No
Online supplements
Access to online resources, tests and quizzes: Excluded No
Included No Price €10,- No €20,- No €30,- No €40,- No
Table 3, Attributes and levels depending on the chosen format
Printed attributes: Levels: Innovative yes/no:
Combo: Excluded No
Combo: with additional eBook Yes
Condition: New No
Used, shows signs of wear No
Shipping time: 1 day No
3 days No
5 days No
eBook attributes: Levels: Innovative yes/no:
Interactive features: Excluded No
Videos in the eBook Yes
Quizzes in the eBook Yes
Homework/assignment questions in the eBook Yes
Notes and highlights: Excluded No
Add highlights and take notes Yes
Add highlights and take notes, see highlights and notes from other readers Yes
Add highlights and take notes, see highlights and notes from the teacher Yes
Read aloud function: Excluded No
Included Yes
Accessibility: On a computer No
On a smartphone, tablet, e-reader Yes
On all devices Yes
3.2 Procedure and experimental design
13
external validity. The motivational text described a situation in which they were attending a course of their academic field and the books they were presented with were recommended for the lecture.
The CBC is factional factorial, meaning the respondents are only presented with a subset of the full factorial in which they are presented with ten different choice-sets in which two alternatives were given with either a (printed) book or an eBook option in cue card format (visualized in figure 2), including each time different attributes and levels (presented in table 1 and 2).
In a real-life situation, when the different available products are not sufficiently preferred by the consumer, he or she buys nothing. This option is covered in the CBC as a dual-response “no-choice” option, where respondents are explicitly asked after choosing their most preferred choice whether they would actually buy the product if it was available or rather use other ways to get the book (e.g. from the library, pirated, borrowing it from a friend etc.). By using the dual response no-option, the predictive accuracy of the CBC significantly increases, and respondents cannot opt-out of making a choice, therefore they still provide data for the conjoint analysis (Wlömert and Eggers, 2016). Finally, the respondents were thanked for their time and effort.
14
Manipulation of incentive-alignment
Overall, the participants were allocated into one of four groups in a 2x2 between-subject design. Within these four groups, participants were randomly exposed to a motivational text where they were presented with either an incentive-alignment (N = 600) for completing the survey or allocated into the hypothetical setting (N = 600), in which there was no incentive provided. Within the incentive-aligned conditioning, participants were aware that the incentive they would receive was directly related to the choices they made. Therefore, one of the most important conditions for incentive-alignment, salience, is approved.
Manipulation of price
Participants were randomly allocated into two groups where they both would receive the same survey with the same attributes and levels. However, the only difference to manipulate the effect of the price attribute, was that one randomly selected group (N = 600) would not be provided with a price variable while the price included group (N = 600) would be provided with all the possible attributes and levels including the price levels.
3.3 Plan of analysis
15
Furthermore, three models were created for the None Option, the eBook- and the (printed) Book format including all possible variables and the three interaction effects of Incentive-Alignment, Price included and Incentive-Alignment * Price Included, resulting in the following three models:
16
Next, the models are imputed into the statistical software program R. Since the random error is assumed to be independent and normally distributed, a multinomial logit (MNL) model will be used. This will result in a MNL model with a S-shaped relationship between utility difference and choice probability (McFadden, 1981). The models will be systematically reduced in R where the least significant attributes will be deleted to improve the model fit. Every time the model is reduced, the model will be tested with the likelihood-ratio test, the Akaike’s Information Criteria (AIC), and the Bayesian Information Criteria (BIC) – which accounts for the number of variables - to measure if the model got significantly better than the previous model. Furthermore, the R2 and R2
adjusted will be calculated to assess the models as such. The
final model is complete when it only includes significant variables or when deleting more variables will no longer increase its predictive power. Finally, the likelihood-ratio of the models will be compared to the null model to see if the models predicts significantly better, and therefore better than chance.
4. RESULTS
In this chapter, all relevant results will be analyzed and described using the analysis described in the methodology part of this thesis. First, the right model will be created and specified, next preferences will be analyzed, and the hypothesis will be answered.
4.1 Model specification
As discussed, all attributes and levels will be added to the model for an aggregate estimation in which all the variables are assumed partworth except for Price and Shipping Time which are assumed linear. Next, the two least significant variables or interaction effects will be deleted, and a new model is generated measuring each time the Log-likelihood and the AIC, BIC, R2 and
R2
adjusted to assess the model and determine whether the model predicts better than the
previous one. The deleted variables and the accuracy metrics per model are presented below in table 5. At model 8, every included attribute contains of at least one significant level, these are deleted in model 9 to see whether deleting these would improve the predictive power. Although model 1 performs best regarding the log-likelihood, the AIC and BIC which are accounting for the number of parameters, show that model 8 is the best predictor. The R2 and
R2
17
acceptable. Since model 8 is overall the best model, this model is used for further analysis and is from this point on referred to as the final aggregate model.
Table 5, Log-likelihood and AIC of the generated models
Model: Deleted variables: Log-likelihood AIC BIC R2 R2Adjusted
#1 No Variables deleted -3337.4 6806.7 7295 0.23 0.21
#2 PI * IA * Notes and Highlights -3337.8 6799.5 7258 0.23 0.21 PI * IA * Online supplements #3 Innovative dummy -3339,4 6786.7 7186 0.23 0.21 IA * eBook PI * IA * eBook #4 PI * IA * Combo included -3341.1 6786.2 7171 0.23 0.21 PI * IA * Condition used #5 IA * Combo included -3342.5 6785.0 7155 0.23 0.22 IA * Condition used
#6 PI * IA * Read aloud function -3345.9 6785.7 7104 0.23 0.22 PI * Online supplements
#7 IA * Read aloud function -3349.9 6785.8 7104 0.23 0.22 IA * Accessibility
#8 PI * IA * Accessibility -3353.6 6783.3 7064 0.23 0.22 PI * Accessibility
#9
PI * notes and highlights readers
-3358.3 6790.6 7071 0.23 0.22 PI * interactive videos
IA * notes and highlights IA * Interactive video Notes and highlights readers
PI = Price included; IA = Incentive-aligned
18
Table 6, Deleted variables of the innovative dummy variable Model: Deleted variables: Log-Likelihood AIC BIC Model Without dummy - -3353.6 6783.3 7150 #1 Notes and highlights -3351.8 6787.6 7154 #2 Online supplements -3347.0 6777.9 7144 #3 Combo Included -3347.9 6779.7 7146 #4 Read aloud function -3348.3 6780.5 7147 #5 Accessibility -3348.3 6780.5 7148
In conclusion, the final aggregate model consists of all sixteen book variables (table 7), nine interactions with Incentive-Alignment, eleven interaction effects with Price Included and four effects where Incentive-Alignment is combined with Price Included (table 8).
4.2 Aggregate model main effects
Table 7 below shows the estimated utilities, ranges, relative importance and the WTP per attribute of the final aggregate model. All the added variables in the model are significant (p < .005) and most are highly significant (p < .001). The Price attribute has the highest relative importance (21%), followed by the amount of shipping days (19%). The attribute that has the highest utility (b = 1.878, p < .001) and adds the most to the WTP (€21.59) is the Combo Included variable.
Table 7, Estimates main effects aggregate model
Variables: Beta Range Importance WTP
None -0.369**
eBook -0.986*** -11.33
Book 0.986*** 1.972 16% 11.33
Combo included 1.878*** 1.878 15% 21.59
Shipping days -0.605*** 2.420 19% -6.95
IF: Videos in the eBook 0.239* 2.75
IF: Quizzes in the eBook -0.392*** -4.51
IF: Homework/assignments in the eBook 0.439*** 0.831 7% 5.05
NAH: AHaTN 0.248** 2.85
NAH: AHaTN, see highlights and notes from other readers -0.133** -1.53
NAH: AHaTN, see highlights and notes from teacher 0.861*** 0.994 8% 9.90
Read aloud function: included 0.475*** 0.475 4% 5.46
Accessibility: On a computer -0.406*** -4.67
Accessibility: On a smartphone, tablet, e-reader -0.178*** -2.05
Accessibility: On all devices 0.584*** 0.990 8% 6.71
Online supplements: Included 0.288*** 0.288 2% 3.31
Price value -0.087*** 2.610 21%
19
4.3 Aggregate model moderating effects
In table 8, the remaining significant moderating effects are presented between Incentive-Alignment, Price Included and Incentive-Alignment * Price Included including the marginal beta’s, and the cumulative beta’s, - range and – importance. The cumulative beta’s are calculated by adding the beta’s per variable in table 7, with the marginal betas in table 8.
Table 8, Estimates final aggregate model moderating effects
Variables: Marginal Beta Cumulative Beta Cumulative Range Cumulative Importance IA * None 0.413*** 0.044 IA * Shipping days 0.295*** -0.310 1,24 21%
IA * IF: Videos in the eBook -0.467*** -0.228
0,774 13%
IA * IF: Quizzes in the eBook 0.891*** 0.499
IA * IF: Homework/assignments in the eBook -0.471*** -0.032
IA * NAH: AHaTN -0.143* 0.105
0,494 8%
IA * NAH: AHaTN, see highlights and notes from other readers 0.094 -0.039
IA * NAH: AHaTN, see highlights and notes from teacher -0.262** 0.599
IA * Online supplements: Included -0.133* 0.155 0,155 3%
IA * Price value -0.023** -0.110 3,3 55% PI * None -1.909*** -2.278 PI * eBook 0.404** -0.582 1,164 20% PI * Book -0.404** 0.582 PI * Combo: Included -0.693* 1.185 1,185 21% PI * Shipping days 0.283** -0.322 1,288 22%
PI * IF: Videos in the eBook -0.084*** 0.155
0,757 13%
PI * IF: Quizzes in the eBook 1.068*** 0.676
PI * IF: Homework/assignments in the eBook -0.520*** -0.081
PI * NAH: AHaTN -0.267* -0.019
1,059 18%
PI * NAH: AHaTN, see highlights and notes from other readers 0.120 -0.013
PI * NAH: AHaTN, see highlights and notes from teacher 0.179* 1.040
PI * Read aloud function: Included -0.196*** 0.279 0,279 5%
PI * IA * None -1.161*** -1.530
PI * IA * Shipping days -0.358*** -0.963 3,852 62%
PI * IA * IF: Videos in the eBook -0.595** -0.356
2,381 38%
PI * IA * IF: Quizzes in the eBook -0.890*** -1.282
PI * IA * IF: Homework/assignments in the eBook 0.660** 1.099
IA = Incentive-Aligned; PI = Price included; NAH = Notes and highlights; AHaTN = Add highlights and take notes; IF = Interactive features; Significant codes: 0 ‘***’ ; 0.001 ‘**’; 0.01 ‘*’ ; 0.05 ‘.’ ; 0.1 ‘ ‘
4.4 Results of hypothesis testing
Effects of incentive-alignment on the consumers’ choice. With the interaction effect
20
The utilities provided in table 8 are visualized below in figure 2. As expected, respondents who where allocated into the incentive-alignment group place more value on price and are becoming more price sensitive (b = -0.110, p < .002) in comparison to the price sensitivity of the main effects (b = -0.087, p < .001). These results are in line with hypothesis 1a and suggest that we can support the claim than incentive-alignment influences a consumers’ price sensitivity. A consumer places more value on price and becomes more price sensitive in an incentive-aligned conditioning.
Figure 2, Utilities of Price and the interaction effect
Next, there was tested if in an incentive-aligned conditioning, a consumer would place less value on innovative product attributes.Since the added innovative dummy variable showed no signs of any significance, there was decided to analyze the significant innovative product attributes independently of each other. Leaving the variables which are presented and visualized in figure 3.
These results show that when both results were significant, in three out of five cases, the expected result occurred. Showing that hypothesis 1b, where was assumed that a consumer places less value on innovative product attributes in an incentive-aligned conditioning, is only partially accepted. Especially regarding Quizzes in the eBook, since when there is no interaction effect (b = -0.392, p < .001), a consumer places a lot less value on the innovative product attribute compared to when there is an interaction effect (b = 0.499, p < .001). -0,087 -0,11 -0,12 -0,1 -0,08 -0,06 -0,04 -0,02 0 U tility
21
Figure 3, Significant innovative product attributes
NAH = Notes and highlights; AHaTN = Add highlights and take notes; IF = Interactive features
Effects of excluding price. With the interaction effect of Price Included * Variable was
attempted to assess the influence of the price attribute on the remaining attributes. Specifically, whether the exclusion of the price attribute would not change the respondents’ relative importance of the remaining attributes
The marginal relative importance and the cumulative relative importance where price is included are presented in table 9. Here is clearly visible that the relative importance has changed for all the remaining product attributes. Hence, these results are not in line with hypothesis 2, so the exclusion of the price attribute does change the respondents’ relative importance of the remaining attributes. However, looking at ranked relative importance, we see that, while very close towards each other, only Format and Combo Included have switched places, while the other variables maintain their ranked order.
Table 9, Relative importance marginal and cumulative
Variable
Marginal importance
Ranked marginal importance
Price Included cumulative importance Ranked cumulative importance Format 16% #2 20% #3 Combo Included 15% #3 21% #2 Shipping days 19% #1 22% #1 Interactive Features 7% #5 13% #5
Notes and Highlights 8% #4 18% #4
Read aloud function included 4% #6 5% #6
22
Effects of excluding price and incentive-alignment. With the interaction effect of Price
Included * Incentive-Alignment * Variable are the effects measured if both situations occurred. There was first tested whether the consumers’ choice would not differ regarding the remaining product attributes, when leaving out price in an incentive-aligned conditioning. In almost all cases, the interaction including both effects were not significant. Only for Shipping days (b = -0.358, p < .001), Videos in the eBook (b = -0.595, p < .002), Quizzes in the eBook (b = -0.890, p < .001) and Homework assignments in the eBook (b = -0.660, p < .002). Meaning that as expected, for most cases the consumers’ choice does not differ regarding the remaining product attributes in an incentive-aligned condition where price is excluded. Meaning that the results are in line with hypothesis 3a.
Next, there was attempted to assess whether the consumers’ choice would not differ regarding the no-choice option when leaving out the price attribute in an incentive-aligned conditioning.
In contrary to what was expected, in table 10 can be seen that the interaction effect combined with the no-choice option was highly significant (b = -1.161, p < .001). Meaning that hypothesis 3b cannot be supported. When leaving out the price attribute in an incentive-aligned conditioning, the consumers choice does differ regarding the no-choice option.
5. DISCUSSION
The aim of this study was to explore the effect of price on the hypothetical bias in conjoint analysis, specifically focusing on investigating ways to reduce this hypothetical bias. In this chapter, all relevant findings will be summarized, followed by the practical implications and finally the limitations and further research directions.
5.1 Summary of findings
23
The results of this experiment only partially support the claim that a consumer places less value on innovative product attributes in an incentive-aligned conditioning. As mentioned before, participants were expected to exhibit lower risk seeking and a lower willingness to try new things in an incentive-aligned conditioning, indicating a clear hypothetical bias for the measured preference for the product attributes. Hence, expected was this to have a negative effect on the “innovative” product attributes since respondents were expected to interpret these attributes as “new” for which they would have a lower willingness to try. This study was only partially able to confirm hypothesis 1b.
Unfortunately, this study has not been able to support the prediction that an exclusion of the price attribute would not change the respondents’ relative importance of the remaining attributes. The assumption from the mainstream economy, that preferences for individual attributes must be independent of the in- or exclusion of a price attribute, or the assumption that individuals strife for utility maximization and have a clear and well-defined preference that should not be influenced by the price of the product, does not hold. Although the ranked relative importance has changed only slightly, the relative importance per attribute was in none of the variables the same. This study was not able to confirm hypothesis 2. This hypothesis is rejected.
However, results did show that for most cases there was no interaction effect regarding the remaining product attributes when leaving out the price attribute in an incentive-aligned conditioning. Though, there were still four variables that did occur to have a significant interaction effect. Hence, the assumption that preferences for individual attributes should be independent of one another in an incentive-aligned conditioning, regardless of the exclusion of a price attribute, does not completely hold. Hence, hypothesis 3a is partially supported.
Unfortunately, the assumption that when excluding the price attribute, there should not arise any effect since the WTB cannot be affected by price, does not hold. Therefore hypothesis 3b, that when excluding the price attribute in an incentive-aligned conditioning, the consumers’ choice would not differ regarding the no-choice option is rejected.
5.2 Theoretical contributions
24
is the addition of incentive-alignment towards the Carlsson et al. (2007) study, which shows that when including incentive-alignment, and by that improving the external validity, preferences are still affected by the inclusion or exclusion of the price attribute. Therefore, this study disproves again the assumption made by the mainstream economy and the neoclassical economics (List, 2002) who state that the preferences for individual attributes must be independent, individuals maximize utility and act independently whether the price attribute is included or not. Since in most cases of this study, excluding the price attribute caused a significant change in the preference of the remaining attributes
Furthermore, this study supported again that incentive-alignment has an effect on the value a consumer places on price and their price sensitivity. This statement is again confirmed and is becoming commonly known within conjoint analysis (Miller et. al., 2011; Ding, Grewal and Liechty, 2005; Dong, Ding and Huber, 2010).
This study found significant differences in the product choices consumers made when in an incentive-aligned conditioning. Although, the reported findings by Ding (2007), regarding participants exhibiting lower risk seeking and a lower willingness to try new things where only partially supported. Therefore, these results contribute to the literature of Ding (2007), showing that not all new things or in this case, innovative product attributes, are affected by incentive-alignment.
Lastly, this research contributes to the study of Miller, Hofstetter, Krohmer and Zhang (2011), who reported of a much higher number of “no-choices” in the incentive-aligned conditioning. This statement is confirmed and extended, since even when price is excluded as a variable, the WTB is still effected in a positive way.
5.4 Limitations and further research directions
25
decreases from 6793.0 (partworth) to 6783.3 (linear) which proves that a completely partworth model would be a better predictor.
Next, assuming all respondents to belong to the same segment would be hard to defend. Therefore in appendix B, a latent class analysis is created to assess whether the predictive power of the model would increase when assumed that the respondents belong to different segments. Comparing the accuracy metrics of both the latent class analyzes and the final aggregate model show that a latent class analyzes with 4 segments is significantly better at predicting than the final aggregate model. Hence, results may have a different outcome or show different levels of significance when a latent class analyses was used.
Furthermore, there were no descriptive statistics available to study the demographic information of the respondents. This also excluded the possibility to compare the demographics of the respondents with the actual population to assess the trustworthiness of the outcome.
Another limitation of this study was the unbalanced choice design since not all variables had the same number of levels, meaning the number-of-levels- effect could occur (Verlegh, Schifferstein and Wittink, 2002). This effect might lead to consumers assessing variables with more levels as more important in comparison to variables with less levels.
Finally, the results of this study might have had different outcomes if other types of products had been used since the attributes of books and eBooks might not have been innovative enough to, i.e., assess the willingness to try new things. Hence, using a product with more novel attributes would possibly have a higher effect in measuring the effect of price on innovative product attributes.
26
REFERENCES
Ash, M., Murphy, J. J., & Stevens, T. (2004). Hypothetical bias in dichotomous choice contingent valuation studies.
Beneke, J., Flynn, R., Greig, T., & Mukaiwa, M. (2013). The influence of perceived product quality, relative price and risk on customer value and willingness to buy: a study of private
label merchandise. Journal of Product & Brand Management, 22(3), 218-228.
Camerer, C. F., & Hogarth, R. M. (1999). The effects of financial incentives in experiments: A review and capital-labor-production framework. Journal of risk and uncertainty, 19(1-3),
7-42.
Cao, X., & Zhang, J. (2018). Prelaunch demand estimation. working paper presented at MIT Sloan School of Management, Cambridge, MA (September 7, 2017).
Carlsson, F., Frykblom, P., & Lagerkvist, C. J. (2007). Preferences with and without prices-does the price attribute affect behavior in stated preference surveys?. Environmental and
resource economics, 38(2), 155-164.
Ding, M. (2007). An incentive-aligned mechanism for conjoint analysis. Journal of Marketing
Research, 44(2), 214-223.
Ding, M., Grewal, R., & Liechty, J. (2005). Incentive-aligned conjoint analysis. Journal of
marketing research, 42(1), 67-82.
Dodds, W. B., & Monroe, K. B. (1985). The effect of brand and price information on subjective product evaluations. ACR North American Advances.
Dodds, W. B., Monroe, K. B., & Grewal, D. (1991). Effects of price, brand, and store information on buyers' product evaluations. Journal of marketing research, 307-319. Dong, S., Ding, M., & Huber, J. (2010). A simple mechanism to incentive-align conjoint
experiments. International Journal of Research in Marketing, 27(1), 25-32.
Eggers, F., & Sattler, H. (2011). Preference measurement with conjoint analysis. Overview of state-of-the-art approaches and recent developments. GfK Marketing Intelligence
27
Gardner, D. M. (1971). Is there a generalized price-quality relationship?. Journal of
Marketing Research, 8(2), 241-243.
Hair, J. F., Black, W. C., Babin, B. J. & Anderson, R.E. (2010). Multivariate data analysis: A global perspective.
Harrison, G. W. (2006). Experimental evidence on alternative environmental valuation methods. Environmental and Resource Economics, 34(1), 125-162.
Harrison, G. W., & Rutström, E. E. (2008). Experimental evidence on the existence of hypothetical bias in value elicitation methods. Handbook of experimental economics
results, 1, 752-767.
Hensher, D. A. (2006). Revealing differences in willingness to pay due to the dimensionality of stated choice designs: an initial assessment. Environmental and Resource
Economics, 34(1), 7-44.
Hensher, D. A. (2010). Hypothetical bias, choice experiments and willingness to pay. Transportation Research Part B: Methodological, 44(6), 735-752. List, J. A. (2002). Preference reversals of a different kind: The" More is less"
Phenomenon. American Economic Review, 92(5), 1636-1643.
Loomis, J. (2011). What's to know about hypothetical bias in stated preference valuation studies?. Journal of Economic Surveys, 25(2), 363-370.
Louviere, J. J., & Woodworth, G. (1983). Design and analysis of simulated consumer choice or allocation experiments: an approach based on aggregate data. Journal of marketing
research, 350-367.
Malhotra, N. K. (1982). Information load and consumer decision making. Journal of consumer
research, 8(4), 419-430.
Martilla, J. A., & James, J. C. (1977). Importance-performance analysis. The journal of
marketing, 77-79.
28
McFadden, D. (1981). Econometric models of probabilistic choice. Structural analysis of discrete data with econometric applications, 198272.
Miller, K. M., Hofstetter, R., Krohmer, H., & Zhang, Z. J. (2011). How should consumers' willingness to pay be measured? An empirical comparison of state-of-the-art
approaches. Journal of Marketing Research, 48(1), 172-184.
Murphy, J. J., Allen, P. G., Stevens, T. H., & Weatherhead, D. (2005). A meta-analysis of hypothetical bias in stated preference valuation. Environmental and Resource
Economics, 30(3), 313-325.
Reibstein, D., Bateson, J. E., & Boulding, W. (1988). Conjoint analysis reliability: Empirical findings. Marketing Science, 7(3), 271-286.
Sandor, Z., & Wedel, M. (2001). Designing conjoint choice experiments using managers’ prior beliefs. Journal of Marketing Research, 38(4), 430-444.
Smith, V. L. (1976). Experimental economics: Induced value theory. The American Economic
Review, 66(2), 274-279.
Srinivasan, V., & Shocker, A. D. (1973). Linear programming techniques for multidimensional analysis of preferences. Psychometrika, 38(3), 337-369.
Thaler, R. (1985). Mental accounting and consumer choice. Marketing science, 4(3), 199-214. Vecchio, R., & Annunziata, A. (2015). The effect of monetary versus nonmonetary endowment on WTP in BDM auctions. Applied Economics Letters, 22(2), 137-140. Verlegh, P. W., Schifferstein, H. N., & Wittink, D. R. (2002). Range and number-of-levels effects in derived and stated measures of attribute importance. Marketing Letters, 13(1),
41-52.
Wittink, D. R., Huber, J., Zandan, P., & Johnson, R. M. (1992, June). The number of levels effect in conjoint: Where does it come from, and can it be eliminated. In Sawtooth Software
Conference Proceedings (pp. 355-364). Ketchum, ID: Sawtooth Software
Wlömert, N., & Eggers, F. (2016). Predicting new service adoption with conjoint analysis: external validity of BDM-based incentive-aligned and dual-response choice
29
APPENDIX
Appendix A
To check whether changing the Price variable and the Shipping Days variable into a partworth model would increase the predictive power of the model, they were included in the model and the log-likelihood and the AIC were checked again to see if they would improve. These new figures are presented below in table 7. These outcomes show that in the aggregate model, changing both variables to a partworth model does not change the predictive power of the entire model. In the final model, the log-likelihood slightly increase. However, this is outbalanced with the AIC which shows a big drop. Finally, comparing the partworth model with the linearized model in a likelihood-ratio test of nested models shows they are highly significant [df = 38, p < .001], meaning there is no significant difference between both models.
Table 10, Partworth and Linear model ratio’s
Price Partworth Linear Shipping Days Partworth Linear
Log-Likelihood(1) -3337.4 -3337.4 Log-Likelihood(1) -3337.4 -3337.4
Log-Likelihood(8) -3354.5 -3353.6 Log-Likelihood(8) -3354.0 -3353.6
AIC(1) 6806.7 6806.7 AIC(1) 6806.7 6806.7
AIC(8) 6793.0 6783.3 AIC(8) 6784.1 6783.3
Appendix B
Table 11, metrics latent class analysis
Latent class LL AIC
Final Aggregate model -3353.6 6783.3
3 segments -3200 6632.8
30
Table 12, utilities per different class
Class Variable Beta P-value Sig.
Class 1
None 2,95E+03 0.2347469
eBook -3,68E+03 0.0954944 .
Combo included 8,78E+03 0.0917522 .
Condition used -1,31E+04 0.0380187 *
Shipping days -4,60E+03 0.0010546 **
Interactive features: Videos in the eBook 4,41E+02 0.7356999 Interactive features: Quizzes in the eBook -3,74E+03 0.0387383 * Interactive features: Homework/assignments in the eBook 6,79E+03 0.0003063 ***
NAH: AHaTN 1,51E+03 0.2785401
NAH: AHaTN, see highlights and notes from teacher 7,15E+03 6,04E-03 *** Read aloud function: included 5,87E+03 7,44E-09 *** Accessibility: On a computer -6,83E+03 5,51E-07 *** Accessibility: On a smartphone, tablet, e-reader -2,61E+03 0.0114533 * Online supplements: Included -2,43E+02 0.7851178
Price value -1,15E+03 1,13E-11 ***
IA * None -3,91E+03 0.1131329
IA * Shipping days 2,83E+03 0.0253875 * IA * Interactive features: Quizzes in the eBook 1,01E+04 0.0004504 *** IA * Interactive features: Homework/assignments in the eBook -1,10E+04 7,96E-02 *** IA * NAH: AHaTN, see highlights and notes from teacher -1,55E+02 0.9425716 IA * NAH: AHaTN, see highlights and notes from other readers 1,82E+03 0.3221962 IA * Online supplements: Included 4,26E+02 0.7530560
IA * Price value 1,36E+02 0.5330570
PI * None -1,50E+04 0.0041821 **
PI * eBook -7,18E+02 0.8505228
PI * Combo: Included 1,56E+04 0.1181091 PI * Condition: Used -1,79E+04 0.1288618 PI * Shipping days 1,35E+03 0.6011729 PI * Interactive features: Quizzes in the eBook 1,06E+04 0.0020479 ** PI * Interactive features: Homework/assignments in the eBook 8,74E+02 0.7913221
PI * NAH: AHaTN -2,80E+03 0.2309164
PI * NAH: AHaTN, see highlights and notes from teacher 5,18E+03 0.0324303 * PI * Read aloud function: Included -6,16E+03 4,08E-02 ***
PI * IA * None -4,37E+03 0.4678352
PI * IA * Shipping days -9,42E+03 0.0005339 *** PI * IA * Interactive features: Videos in the eBook 9,29E+01 0.9764990 PI * IA * Interactive features: Quizzes in the eBook -1,55E+04 0.0021478 ** PI * IA * Interactive features: Homework/assignments in the eBook 8,40E+03 0.0987753 .
31
Class Variable Beta P-value Sig.
Class 2
None -8,91E+02 0.7568808
eBook -1,99E+04 5,33E-12 ***
Combo included 3,00E+04 2,04E-02 ***
Condition used -2,64E+04 0.0001342 ***
Shipping days -9,68E+03 1,12E-04 ***
Interactive features: Videos in the eBook -3,59E+02 0.8115519 Interactive features: Quizzes in the eBook -4,77E+03 0.0753659 . Interactive features: Homework/assignments in the eBook 8,60E+02 0.7344681
NAH: AHaTN 3,51E+02 0.8669554
NAH: AHaTN, see highlights and notes from teacher 1,09E+04 2,11E-03 *** Read aloud function: included 8,61E+03 1,34E-07 *** Accessibility: On a computer -2,63E+03 0.0136827 * Accessibility: On a smartphone, tablet, e-reader -4,59E+03 3,90E-02 *** Online supplements: Included 3,51E+03 0.0001716 ***
Price value -1,06E+03 < 2.2e-16 ***
IA * None 2,57E+03 0.3275332
IA * Shipping days 4,89E+03 0.0029709 **
IA * Interactive features: Quizzes in the eBook 1,71E+04 4,16E-03 *** IA * Interactive features: Homework/assignments in the eBook -5,18E+03 0.2015937 IA * NAH: AHaTN, see highlights and notes from teacher -5,82E+03 0.0120543 * IA * NAH: AHaTN, see highlights and notes from other readers 1,61E+03 0.3783072 IA * Online supplements: Included -3,04E+03 0.0217701 *
IA * Price value 2,00E+01 0.9065186
PI * None -2,94E+04 2,52E-06 ***
PI * eBook 1,43E+04 8,87E-02 ***
PI * Combo: Included -1,80E+04 0.0430496 * PI * Condition: Used 1,95E+04 0.0496819 *
PI * Shipping days 6,17E+03 0.0088760 **
PI * Interactive features: Quizzes in the eBook 1,30E+04 0.0001081 *** PI * Interactive features: Homework/assignments in the eBook -7,58E+03 0.0204478 *
PI * NAH: AHaTN -7,84E+02 0.7622439
PI * NAH: AHaTN, see highlights and notes from teacher -2,37E+02 0.9234955 PI * Read aloud function: Included -5,22E+03 0.0014755 **
PI * IA * None 6,13E+03 0.2333082
PI * IA * Shipping days -5,90E+02 0.7820453 PI * IA * Interactive features: Videos in the eBook -7,49E+02 0.7974190 PI * IA * Interactive features: Quizzes in the eBook -1,69E+04 0.0004326 *** PI * IA * Interactive features: Homework/assignments in the eBook 9,39E+03 0.0770665 .
32
Class Variable Beta P-value Sig.
Class 3
None -3,64E+03 0.2841768
eBook -8,51E+03 0.0021664 **
Combo included 2,60E+04 0.0002090 ***
Condition used -3,02E+04 0.0001711 ***
Shipping days -5,41E+03 0.0028519 **
Interactive features: Videos in the eBook 1,19E+04 1,55E-07 *** Interactive features: Quizzes in the eBook -5,44E+03 0.0989696 . Interactive features: Homework/assignments in the eBook 6,20E+03 0.0649282 .
NAH: AHaTN 5,18E+02 0.7980489
NAH: AHaTN, see highlights and notes from teacher 1,70E+04 3,07E-06 *** Read aloud function: included 5,72E+03 5,04E-03 *** Accessibility: On a computer -3,69E+03 0.0119131 * Accessibility: On a smartphone, tablet, e-reader -2,61E+03 0.0675169 . Online supplements: Included 7,70E+03 1,34E-05 ***
Price value -4,40E+02 0.0153177 *
IA * None 1,09E+04 0.0004199 ***
IA * Shipping days 3,31E+03 0.0752649 . IA * Interactive features: Quizzes in the eBook 6,22E+03 0.1641256 IA * Interactive features: Homework/assignments in the eBook 8,33E+02 0.8461529 IA * NAH: AHaTN, see highlights and notes from teacher -5,91E+03 0.0696562 . IA * NAH: AHaTN, see highlights and notes from other readers -3,67E+02 0.8732174 IA * Online supplements: Included -2,27E+03 0.2095150
IA * Price value -2,36E+03 6,80E-04 ***
PI * None -1,18E+04 0.1351465
PI * eBook -1,21E+04 0.0262926 *
PI * Combo: Included 1,59E+04 0.4599554 PI * Condition: Used -1,50E+04 0.5164449 PI * Shipping days -1,98E+01 0.9954253 PI * Interactive features: Quizzes in the eBook 1,64E+04 0.0017596 ** PI * Interactive features: Homework/assignments in the eBook -1,19E+04 0.0228464 *
PI * NAH: AHaTN -1,11E+02 0.9740801
PI * NAH: AHaTN, see highlights and notes from teacher -5,40E+03 0.1412762 PI * Read aloud function: Included 3,87E+03 0.0992679 .
PI * IA * None -8,70E+04 3,61E-08 ***
PI * IA * Shipping days -1,59E+04 1,58E-03 *** PI * IA * Interactive features: Videos in the eBook -3,07E+04 1,33E-03 *** PI * IA * Interactive features: Quizzes in the eBook 4,39E+02 0.9576205 PI * IA * Interactive features: Homework/assignments in the eBook 8,17E+02 0.9161674
33
Class Variable Beta P-value Sig.
Class 4
None -2,12E+05 0.9962317
eBook -2,12E+05 0.9962315
Combo included 6,27E+05 0.9962794
Condition used -7,70E+05 0.9965692
Shipping days -8,93E+04 0.9953024
Interactive features: Videos in the eBook 7,60E+03 0.0055158 ** Interactive features: Quizzes in the eBook -8,22E+03 0.1476853 Interactive features: Homework/assignments in the eBook 5,04E+03 0.3042444
NAH: AHaTN 1,28E+04 0.0004039 ***
NAH: AHaTN, see highlights and notes from teacher 5,26E+03 0.2106445 Read aloud function: included -4,81E+03 0.0306307 * Accessibility: On a computer -6,08E+03 0.0010066 ** Accessibility: On a smartphone, tablet, e-reader 7,17E+03 0.0001081 *** Online supplements: Included 5,85E+03 0.0003939 ***
Price value -1,39E+03 1,85E-06 ***
IA * None 1,09E+04 0.0314161 *
IA * Shipping days -1,14E+05 0.9969546 IA * Interactive features: Quizzes in the eBook 1,81E+04 0.0155980 * IA * Interactive features: Homework/assignments in the eBook -1,19E+03 0.8640092 IA * NAH: AHaTN, see highlights and notes from teacher -7,33E+03 0.0930963 . IA * NAH: AHaTN, see highlights and notes from other readers 1,23E+03 0.6474334 IA * Online supplements: Included -2,69E+03 0.2640736
IA * Price value 7,71E+01 0.8060711
PI * None 1,62E+05 0.9971231
PI * eBook 2,14E+05 0.9961988
PI * Combo: Included -6,19E+05 0.9963236 PI * Condition: Used 7,73E+05 0.9965545 PI * Shipping days 8,56E+04 0.9955013 PI * Interactive features: Quizzes in the eBook 1,93E+04 0.0069165 ** PI * Interactive features: Homework/assignments in the eBook -3,25E+03 0.5831197 PI * NAH: AHaTN -1,80E+04 0.0001530 *** PI * NAH: AHaTN, see highlights and notes from teacher 1,88E+04 0.0001008 *** PI * Read aloud function: Included 8,84E+03 0.0016043 **
PI * IA * None -3,10E+03 0.7963981
PI * IA * Shipping days 1,17E+05 0.9968645 PI * IA * Interactive features: Videos in the eBook -3,77E+03 0.4814070 PI * IA * Interactive features: Quizzes in the eBook -1,87E+04 0.0484354 * PI * IA * Interactive features: Homework/assignments in the eBook -1,43E+03 0.8644199
