Methodology & Method
The third chapter gives an overview of the methodology and the method used and applied in the research – the research design. Firstly, the philosophy and the approach of the research is presented and argued for. Furthermore, conjoint techniques as base for the research strategy and the method of data collection are presented. The questionnaire design, the data analysis process and lastly the research quality are assessed.
Research philosophy is defined by Saunders, Lewis, Thornhill and Wilson (2009) as the way the world is perceived by the researchers and thus indicates what techniques, methods and strategies will be used in the research. By that, the philosophy a research commits to is an important component during the research itself. It impacts how the subject of interest will be interpreted throughout the research (Johnson & Clark, 2006).
Several research philosophies could be elaborated on, but the most frequent ones are pragmatism, interpretivism, realism and positivism (Saunders et al., 2009). Positivism refers to a philosophy where the social world exists independent and objectively from the researcher’s viewpoint, i.e. the investigation is thought to be made without any external factors influencing. By that, a positivistic standpoint implies that the topic of interest, the research problem, can be answered or discussed from an objective position and thus, an objective conclusion can be drawn (Blumberg, Cooper, & Schindler, 2011).
Interpretivism however, is located on the other side on the spectrum of research philosophies and refers to a standpoint where the social world is a subjective matter. An interpretivist thus accepts the reality as being a vision of human imagination (Collis, & Hussey, 2014). Easterby-Smith, Thorpe and Jackson (2015) refer to this stand as a constructionist approach and argue that a constructionist stance has its interest in people and how they sense the reality.
Hinkelmann and Witschel (2013) describe positivistic research being a great fit if the main goal is to describe a given phenomenon from an objective standpoint and understand and describe individuals’ attitudes. By this, the philosophy behind this study is of positivistic nature since we believe it to be applicable in regard to our goal to collect and process data from an objective standpoint.
Two primary approaches could be applied when a research is conducted, it could either be inductive or deductive (sometimes one with attributes of both – abductive). However, both refer to the relation between theory and conducted research during the study (Bryman & Bell, 2007). An inductive approach is based on the premises that data will be the foundation for the construction of new theory or knowledge, based on the gathered data and furthermore be linked to existing literature (Saunders et al., 2009). Additionally, inductive reasoning is generally said to be a good fit for qualitative methods, even though it can be used in quantitative research as well (Saunders et al., 2009).
A deductive approach however, could be said to be the opposite of induction, by being a method where data is facilitated to test a theory i.e., the research uses existing theories to empirically test these against collected data, often by confirming or denying pre-constructed hypotheses (Saunders et al., 2009). According to Malhotra, Birks and Wills (2012), raising issues from existing research by conducting a literature review and construct hypotheses based on this review, is called deductive reasoning. Deductive reasoning, in contrast to inductive reasoning, tends to be used in quantitative research even though it can be used for qualitative research as well.
Although we follow a quantitative approach, our study is based on inductive reasoning, primarily because of three reasons: First, while there is some theory, that could be used as groundwork to tackle the phenomenon of CC, develop hypotheses and subsequently test them, we simply do not believe that this would be expedient in this case and, as being stated by Patton (2002), approaches can be combined or mixed creatively if that is believed to facilitate the research. As there is very little to no evidence on what attributes could define a user’s willingness to embrace CC, we decided to choose a more exploratory approach. Moreover, as we are trying to actively contribute our humble share to the further development of CC-technology and in relation to Guenther’s (2013) and Boy’s (2012) HCD models, our gathered data could be seen as a small extract in an iterative and ongoing process, and not as definitive static truth. Furthermore, our approach is in line with Saunders et al.’s (2009) statement of generating new knowledge from emerging data, that subsequently can be linked to existing theory
Data Collection Method
The method of data collection describes how and what actions that are taken during the research, in order to reach the goals and the purpose, and by extension, how researchers plan on answering the formulated research question(s) (Saunders et al., 2009). Different philosophical stances or research approaches call for different methods. However, there are three alternatives when conducting a research. It could either be qualitative, quantitative or a combination – a mixed methods approach (Easterby-Smith et al., 2015).
In qualitative studies, case study, action research and grounded theory are common strategies (Saunders et al., 2009). A qualitative study is defined as a non-numeric way of collecting data (Easterby-Smith et al., 2015) and instead has its focus on the phenomenon that do have an impact on individual’s reality, both in an individual or an organizational context (Mills & Birks, 2014). A qualitative study is often performed through interviews and observations, which provide researchers with primary data, that later can be examined in-depth (Bluhm, Harman, Lee, & Mitchell, 2011). A quantitative study aims at understanding and examining relationships between variables from an objective standpoint by using mathematical tools as the way of analyzing data (Muijs, 2010).
In quantitative studies, hypothesis testing and exploratory techniques are more common (Saunders et al., 2009).
As mentioned in previous chapters, this study will be quantitative in its nature since our purpose is to quantitatively measure variables that reflect people’s approval/rejection of CC’s possible characteristics in a numerical way and understand the given phenomenon from an objective perspective. Also, our intention is to answer the question with higher reliability, something that quantitative research enables.
Furthermore, the gathered data could either be primary or secondary. Primary data refers to data that is collected by the researchers first-hand and secondary data thus refers to data collected from other sources, e.g. published articles, books and reports etc.
(Easerby-Smith et al., 2015). This research will be based on primary data collected through a survey. Malhotra et al. (2012) state that the use of surveys is a powerful tool in research since it lets researchers objectively examine possible dependabilities between variables. Secondary data was also facilitated in the theoretical framework and is therefore building a foundation of this study
Ethical considerations are an important factor for research itself that every researcher should internalize since it establishes and increases the overall quality of the research, while at the same time also working as a comforting factor for participants of the study (Easterby-Smith et al., 2015). Easterby-Smith et al. (2015) mention 10 key principles that help to protect participants and the research integrity:
- Ensure that no harm comes to participants.
- Respect the dignity of research participants.
- Ensure a fully informed consent of research participants.
- Protect the privacy of research participants.
- Ensure the confidentiality of research data.
- Protect the anonymity of individuals or organizations.
- Avoiding deception about the nature or aims of the research.
- Declare affiliations, funding sources and conflicts of interest.
- Honesty and transparency in communication about the research.
- Avoid of any misleading or false reporting of research findings.
In terms of ethical issues regarding participants, both anonymity and consent were ensured and offered. Individuals who participated in our research were offered full anonymity, no name was mentioned in the report, nor asked for in the survey. Also, the purpose of the study was explicitly described so that participants would have no doubt what the study was about or what it would be used for. The participation was completely voluntary. Additionally, gathered data will be protected and only the two authors will have access to all information, so that confidentiality can be ensured.
Moreover, our research is proven to be justified by our literature review and the identified gaps make the subject novel. We did not plagiarize and every content that was taken from other sources than ourselves is referenced and credited to our best knowledge and believe
To enable us to identify the significance of attributes and their characteristics that users wish for when using CC applications, we decided to perform a conjoint analysis. In respect to our research question, we believe that this technique provides promising possibilities in the field of user research, respectively allows us to classify and weight potential customers’ needs and attitudes towards CC and subsequently rank them in a hierarchical way. In the following sections, conjoint analysis as a theory will be explained. Further on, different techniques and tools within the theory of conjoint analysis, that were used within this research, will be reviewed and explained
The technique of conjoint analysis (also referred to as conjoint measurement) is a survey-based research tool, established in the 1970s through a number of articles by Paul E. Green and colleagues (e.g. Green & Rao, 1971; Green, Wind & Carroll, 1973; Green, 1974; Green & Srinivasan, 1978). Green transferred procedures, originating in mathematical psychology to marketing, respectively especially market research, to create a multivariate method, that allows to measure (potential) consumers valuation of certain product attributes and their characteristics, and furthermore enables the creation of predictive models to forecast consumers (purchase) behavior, act accordingly to their preferences, or even predict and estimate accepted prices.
Conjoint analysis as one of the presumably most significant tools used in marketing research (Rao, 2014; Toubia, 2018), decomposing products or services into smaller units, and enabling the individual measurement of these units (Toubia, 2018). According to various authors (among others: Malhotra et al., 2012; Rao, 2014; Toubia, 2018), this measurement technique is usually done by identifying the defining attributes (e.g. price, color, size) of a product/service, followed by a differentiation of its different levels (e.g. for price: 100 SEK, 200 SEK, 300 SEK). And while some sources, for instance the survey platform Qualtrics substitute the term ‘attributes’ with ‘features’ and then use ‘attribute’ to describe a feature and all its levels at once (Qualtrics, n.d.), we decided to keep it as simple as possible and reduce redundancies by proceeding with ‘attributes’ and ‘levels’ only.
After attributes and levels are defined, the researcher composes different profiles, each consisting of a set of attribute levels, which then are evaluated by the participants and subjectively assessed against each other – the participant is hereby forced to make a (or multiple) decision(s) in favor/against certain profiles, so called trade-offs (Rao, 2014). This individual (and somewhat forced) decision process between different product profiles, mimics real-life purchasing choices or usage preferences way better than a simple rating of single attributes ever could and is therefore what makes conjoint analysis so interesting for market research. As Aaker, Kumar and Day (2004) postulated, the information obtained with this trade-off method is most certainly of greater value and more precise than data gathered by only asking what attributes the consumer deems important as possible product characteristics, as consumers tend to answer that all of them are important. Therefore, by making consumers sacrifice certain profiles, and thereby attributes plus subsets, in favor of others, the joint value effects of attributes’ levels can be identified and subsequently decomposed back into individual evaluations of single attribute levels, which further enables the computing of numerical scores for each level of all attributes in all possible combinations (Rao, 2014). With the generation of so-called utility values, based on the participants’ choices, a numerical score for every single attribute and as well for each and every level of all attributes, is computed. These utility values can also be called partworths and are usually calculated with appropriate software solutions (e.g. IBM SPSS). Subsequently, it is possible to simulate various models, using all kinds of attribute-level combinations, even if they were not rated upon as full profile by the participants. With this kind of modelling it is even possible to predict market shares in percentage for every profile possible or price sensitivity, respectively what users are willing to pay for what (if the attribute price with different levels was part of the tested set) (Rao, 2014)
Types of Conjoint Analysis
According to Rao (2014), there are mainly four distinguishable types of conjoint methods:
The traditional conjoint analysis (CA) is the one Green established as pioneer in this field (e.g. Green & Rao, 1971; Green, Wind & Carroll, 1973; Green, 1974; Green & Srinivasan, 1978). His full-profile method lets the participant rate or state preferences for profiles constructed from the whole set of attributes and levels in the test. Therefore, theoretically all possible profiles are rated by the participant. However, as the number of full profiles increases exponentially with the number of attributes and levels (e.g. 4 attributes with 3 levels each, equals 3*3*3*3 = 81 different profiles), it would be utopic to believe that a participant would or even could judge them all. To cope with this problem, usually a significantly smaller set of profiles is selected for the conjoint analysis, based on the experimental design. In the data analysis the stated preferences of all individual participants are transferred and decomposed (usually with regression-based methods) to separate utility values for each level corresponding to one attribute – the outcome are attribute-specific partworth functions, showing the consumers preferences in terms of the attributes’ different levels.
Choice-based conjoint analysis (CBCA) is nowadays the presumably most popular type of CA. These methods draw on stated direct choices between product or service profiles instead of the theoretical usage of full profiles in traditional CA, and therefore are closer to the reality of a participant’s real purchase decisions. Usually, the participant has to choose several times between multiple profiles, to measure his/her preferences and trade-offs. In CBCA the partworth functions are typically computed using multinomial logit methods.
Adaptive conjoint analysis (ACA) is an approach to handle extensive attribute/level numbers with a two-folded procedure: Firstly, the participant fills a self-explicated questionnaire, concerning attribute importance and level preferences; Secondly, adapted to the preceding questionnaire, a number of partial profiles (only two at a time) is to rate by preference. As the shown partial profiles are instantly customized for each participant, and therefore need complex and various calculations. ACAs needs to be computer-aided to unfold their potential.
Self-explicated conjoint analysis, in contrast to the other three mentioned methods, that are all decompositional, is a compositional approach and works bottom-up in comparison to the others. The participants are asked to rate all levels of all attributes individually in terms of their desirability, plus the relative importance of each attribute. From this rating then, the importance and desirability of each possible profile can be extrapolated, based on the single levels’ and attributes’ ratings.
We did choose a full profile method since it could be said to be a realistic one from a consumer perspective (Toubia, 2018); consumers would not be faced with individual attributes of conversational commerce, but rather experience conversational commerce as a whole – with all the attributes working simultaneously. Moreover, we have chosen the traditional method within this research since we believe this to be the best choice for our purpose and our limitations timewise as well as financially (Toubia, 2018).
Application of Conjoint Analysis
To perform and administer a conjoint analysis in a sufficient way, several authors suggest general frameworks on how to design the research process in order to obtain valid and robust results. In this chapter we will discuss recommendations, and subsequently explain and motivate the construction of our own path.
When examining the theoretical models of Gustafsson, Herrmann and Huber (2000) in fig. 3-1 and Malhotra et al. (2012) in fig. 3-2 below, it is evident that different authors use different wordings and emphasize or prioritize in different ways
1.2 Problem Statement
1.3 Purpose and Scope
1.5 Thesis Disposition
2 Frame of Reference
2.1 Choice of Method
2.2 Application of Method
2.3 The Definition of Conversational Commerce
2.4 The Technology Acceptance Model
2.5 Human Centered Design
3 Methodology & Method
3.1 Research Philosophy
3.2 Research Approach
3.3 Data Collection Method
3.4 Research Ethics
3.5 Conjoint Analysis
4 Empirical Findings .
4.2 Cumulative score
5.1 Overall Statistical Results and Analysis
5.2 Cross Country Comparison
5.3 Generation and Analysis of Simulations
7.1 Recommendations for Further Research
7.2 Managerial Implications
7.3 Ethical Implications
GET THE COMPLETE PROJECT