CHAPTER THREE – METHODOLOGY
In this chapter we critically look at the
theoretical framework and the methodology adopted for the study. It starts with
the area of the study and proceeds to look at the strategy employed for
sampling. The design of the Discrete Choice Experiment (DCE); determination of
attributes and their levels, the questionnaire, pilot questionnaire, collection
of data, data entry and analysis are also discussed and finally presents a
detailed discussion of the DCE rooted in random utility theory taking critical
look into the logit model.
3.1 The Study Area
Comprising of 27 administrative
districts/municipalities the Brong Ahafo region of Ghana is the second largest
region with a land area of 39,558 km2 , located within longitude 00 15′
E-30 W and Latitude 80 45′ N-70 30′ S in the
west central part of Ghana. It shares boundaries with five other regions
namely, Ashanti and Western Regions to the south, the Volta Region to the east,
Northern Region to the north, and the Eastern Region to the south-east. It
shares also an international boundary to the west with the Republic of Cote
region currently has a total population of 2,282,128 (2010) with an estimated
growth rate of 2.2% (against 2.4% national average). Proportion of rural/urban
is 62.6: 37.4 (CSIR, 2015) with majority of the indigenes being farmers and
3.5 Sample Size and Data Collection
The study used simple random sampling technique.
Vehicle drivers in the Sunyani Municipality were targeted since the study
sought to analyze the hypothetical choice of an insurance company by these
people. A sample size of 100 respondents was selected for the study. With a
total sample of fifty (50) individuals, each presented with at least 16 choice
sets and entirely generic parameter description for design attributes and
covariates might just be sufficient for choice experiment (Hensheret al.,2005).
Design of the Discrete Choice Experiment
Respondents were asked to choose between
pairs of hypothetical companies resulting from the grouping of both attributes
and corresponding levels. However, this technique necessitates respondents to
trade-off the different aspects of their choice of a company thereby identifying important
attributes/levels in the study.
3.3.1 Choice of Attributes and Levels
In every DCE studies, the second step
after the problem refinement stage is the choice of the attributes and their
corresponding levels (Hensher et al., 2005). The method implemented in this
research is to use data generated from a discrete choice stated preference (SP)
survey. The design stages of such surveys generally include the following
1) Determination of attributes, levels and
2) Elicitation of
3) Data analysis and interpretation
3.6.1 Determination of attributes and
When creating an SP survey, the attributes
(characteristics) of importance need to be defined and levels (values) assigned
to them respectively. In order to make sure that an SP survey is genuine, it is
suggested that attribute levels should be reasonable and capable of being
traded. Once the attributes
and their respective levels have
been determined, they
are joined into profiles to present to survey
participants for evaluation.
The authors identified four insurance
company attributes and associated levels to be the most vital on the basis of a
broad preliminary qualitative research. The prior list of possible attributes
and their levels were got through comprehensive discussions with experts in the
insurance sector and drivers in the Sunyani Municipality. Adamowvic etal.
(1998) pronounced that attributes are frequently known from previous experience
and either primary or secondary research. Nonetheless, focus group discussions
were conducted to decrease the initial list of potential attributes to four
(i.e., Premium Cost, Claims Settlement, Customer Service and Proximity). The
company-choice attributes and their levels are defined in Table 1.
3.3.2 Experimental Design
Experimental design aids in organizing and
running our experiment. According to Louviere et al., (2000), it is a way of
manipulating attributes and their levels to enable rigorous testing of certain
hypothesis of interest.
The next stage in DCEs is the experimental
design process to simulate the choice sets to be offered to the individuals.
The choice sets for the DCE questionnaire were produced using well-established
statistical procedures DCE macros in the statistical programme SPSS was used to
create ideal orthogonal design with eight profiles (Kuhfeld 2010; Hensher et al., 2005). This
technique takes into consideration orthogonality (attribute levels are
independent of each other), level stability (attribute levels appear with the
same rate), and marginal overlap (attributes do not take the same level within
a choice set) (Kuhfeld, 2010).
\The profiles generated from the orthogonal design
facility in SPSS are presented in Table 3.2.
The profiles were put together to produce
28 choice sets, which is within the acceptable range for DCE studies and were
randomly blocked into two sets in order to reduce respondent’s exhaustion or
wear-out. In the process of the main
survey individuals were asked to assess the 14 choice sets and point out the
type of company they would prefer to work with.
\Table2 reveals a
choice set presented in the stated preference survey.
The questionnaire is preceded by an
introduction, indicating the subject
The questionnaire is preceded by an
introduction, indicating the focus of the study, the carrier of the survey, and
how the results will be processed.
It contains also the demographic
information on the respondents and their literacy level. According to Adamowicz
et al., (1998), a choice task (questionnaire) is preceded by a set of
standardized instructions to respondents regarding the task, its objective, its
context and how to respond to the scenarios.
3.4.1 Data Collection
The data used for
this study come from a cohort survey of 100 vehicle drivers in the Brong Ahafo
region of the Republic of Ghana. The data were collected using paper and
pencil/self-administered questionnaire in the various vehicle stops and
showed thoroughly how to answer the choice sets in the questionnaire to the
respondents before they started.
were not remunerated for the time spent in completing the questionnaire.
3.4.2 Data Entry
In stated or discrete choice experiment
datasets, respondents answer more than one choice question, which brings about multiple
opinions for each respondent. The questionnaires given out were crisscrossed
prior to entering the data. The data were entered into Microsoft excel before exporting
to the SPSS statistical software for analysis.
The data produced from the survey were
analyzed using the logit estimator in SPSS. The logit model estimates the
probability of choosing a company given the differences in company attribute
levels from the choice set.
3.2 Discrete Choice Experiment (DCE)
Discrete choice models derive from random
utility theory of choice behavior which is a well-tested theory of choice
behavior that can take inter-linked behaviors into account. It was proposed by
Thurstone (1927), and is called random utility theory (RUT).
Currently works in DCE theory and
methodology rely strongly on work by McFadden, did extend Thurstone’s original theory of
paired comparisons (pairs
of choice alternatives)
to multiple comparisons (e.g., McFadden 1986; McFadden
and Train 2000; McFadden 1974; Thurstone 1927).
3.3 Random Utility Theory
Random utility theory gives a thorough explanation
of the behavior of by humans. Precisely, RUT suggests that there is a concealed
component called “utility” present in a choice maker’s mindset that is
difficult to observe by researchers. That is to say that, an individual has a
certain “utility” for each choice alternative, but on the other hand these
“utilities” cannot be observed by the researcher that is why they are characterized
3.3 Discrete Choice Experiments rooted in
Random Utility Theory
RUT takes it that the concealed utilities can
be grouped into two components. An explainable (systematic) component and an
unexplainable (random) component. Systematic
components consist of attributes clarifying discrepancies in choice
alternatives and covariates explaining differences in individuals’
Random components consist of all unrevealed
factors that influence choices. It is further assumed that humans are unsatisfactory
measurement instruments; so, random constituents also can include factors
reflecting inconsistency and variances in choices related with individuals and
not choice options as such.
3.4 Stated Choice/Revealed Choice Design
The parameters of a discrete choice model,
can be estimated using data
relating to observed
choice behavior (revealed
preference data), simulated choice
behavior (stated preference
data) or both combined.
Stated preference (SP)
to Kroes and
Sheldon (1988) SP procedures refer
to ‘a group
of techniques which uses the statements
acquired from individual
respondents about their
preferences’ in a
set of alternatives to
estimate utility functions.
SP data are gathered by means of
experimental circumstances or surveys where the individual respondents are
presented with hypothetical choice problems. For example, the respondent is
asked to choose between three insurance companies.
In this imaginary circumstance only these three
companies exist. The response is the stated choice. Due to this approach, data
collected does no describe the actual behavior but however fully describes how
the decision maker chooses the state they would behave alike.
The main shortcoming of SP data appears to
be clear: the way
themselves to act and how respondents say they will act,
the way they in reality
will act. This occurrence possibly
will rise as a result of the fact that the respondent in reality has no idea of
how they would answer or because he or she feels it is anticipated of them to reply
in an exact manner.
In comparison to SP data, RP data relates
to real behavior. It is referred to as RP because the decision makers actually
reveal their preferences through their own choices. In the example used in the
SP section above and considering the RP method, the respondent would be
questioned as to which insurance company he or she worked with last, instead of
making a choice from a set of hypothetical companies. Hence we can vividly
state that the choosing habits reveal preference. Here utility functions are
represented by observing behavior. With the SP method referred to as direct
approach the RP on the other hand is the indirect approach.
RP approach real
behavior is observed,
rather than confronting the respondent
with a hypothetical
The major benefit of RP data is that it
presents the actual representation of choices. The disadvantage of RP data
representing actual choices is that it’s not suitable for situations that do
not exist currently. Behavior is observed thus there is too much uncertainty in
stating behavior. Estimates can be made, but SP data is just suitable for these
3.7 Econometric Specification
Discrete choice experiment modeling is
rooted in Random Utility Theory (RUT). However, the utility can be modeled as;
=Vci + Eci
Where where Uci
concealed, unobservable utility
that individual i associates
with choice alternative c , Vci
is the deterministic term of the
utility and Eci is the random term,
taking care of the uncertainty.
The deterministic term Vci of each alternative is a function of
the attributes of the alternative itself and the characteristics of the
commuter. McFadden (1974) opined that a utility can be characterized by a
The primary objective of this study is to
assess the various factors that influences individuals company choice
(Insurance company attributes) by vehicle drivers.
The specific objectives are to:
the properties of discrete choice experiment.
a discrete choice experiment to identify factors that Influences an
Individual’s Choice of choosing a specific policy.
the attribute which contributes most in determining the choice of an insurance
the trade-offs between company attributes.
5. Implement subgroup
analysis on demographic
features of respondents to
see how preferences differ by
i. Design a discrete choice experiment on vehicle
drivers and assess the effects of attributes on insurance company choice.
ii. Assess the attribute which contributes
most in determining the choice of an insurance company.
iii. Evaluate the trade-offs between company
v. Implement subgroup
analysis on demographic
features of respondents to
see how preferences differ by company
Relevance of the Study
This study will help insurance companies
in adopting expedient strategies to brand and market their products to maximize
The study is necessary and its outcome
shall benefit research and academia by adding up to the body of knowledge on
factors that influences individuals’ choice of insurance companies.
Finally the outcome of the study will lay
down the foundation for additional research for both policymakers/insurers and
Limitations of the Study
The analysis result might be affected as
only stated preference data was collected.
Only drivers in the main Sunyani
municipality in Ghana were covered in the study. This might not be fully useful
when the population or sample area is extended as it relates to a certain group
of drivers only.
Organization of the Study
This research consist of five chapters
preceded by a summary of the full research (The abstract), table of contents,
list of figures and tables, all abbreviations used, dedications and
The first chapter consists of the
background of the study, problem statement, research objectives, limitations
and the organization of the study.
Chapter two reviews relevant literature
and chapter three expounds on the methods of discrete choice experiments used
for the study.
The analyses and discussions of the
findings from the study are presented in chapter four and finally summary,
conclusions and recommendations in chapter five.
References and appendix can also be found
after the last chapter.