Project topics and materials
Get Complete Project Material File(s) Now! »
Population and sampling
In order to identify the respondents required for the research study, a sampling plan is required (Parasuraman, Grewal, & Krishnan, 2007). The sample selected needs to be representative of the population. A well-defined target population facilitates the accurate identification of the sample (Shiu et al., 2009). The population for this research consisted of respondents belonging to a loyalty programme in the financial services industry.
Respondents were sourced from a market research company named P-Cubed, which owns the largest established commercial consumer database in South Africa. P-Cubed collects data on consumers from credit bureaus, the deeds office, census information, CIPRO, and Home Affairs in order to provide insights. These insights relate to consumer credit history, to their overall wealth, where they live, whether they are company directors, and how many dependents they might have. At the time of the research, P-Cubed provided 2,700 fields of data on over 33 million economically active consumers in South Africa (P-Cubed, 2017). The data segment consumers in terms of their demographics, financial affluence, preferred credit consumption, digital enablement, entrepreneurial spirit, life-stage and neighbourhood trends. The P-Cubed database of individuals was used as the sample frame for this research due to its size and richness in terms of consumer data.
The sampling frame is defined based on the population, and can never be perfect (Shiu et al., 2009), as all sampling frames consist of an error margin which depict the deviation of the sample frame from the population. This error can be alleviated through appropriate screening questions (Malhotra et al., 2010). No database exists of consumers belonging to financial services loyalty programmes, and as such no suitable sample frame could be adopted for this research.
Whilst the sampling frame may not have included all consumers who belonged to loyalty programmes from financial services institutions in South Africa, and would have included consumers who did not belong to any loyalty programme in South Africa, only respondents that belonged to at least one loyalty programme were able to complete the questionnaire as the coding within the questionnaire enabled termination after this question was posed. The P-Cubed database could be seen as representative of the population due to its size and its spread across all consumer segments. This database is suited to research in the financial services industry as the services provided by P-Cubed target banking, micro-lending, credit, insurance, loyalty and rewards (P-Cubed, 2017). The survey introduced a bias towards English-speaking consumers as the survey was only presented in English. P-Cubed is one of the most dominant institutions with respect to customer insights, research and analytics in Africa, with the largest consumer database in South Africa and as such should have provided for maximum variation in terms of respondents targeted.
All sampling methods are either classified as probability or non-probability sampling (Churchill & Iacobucci, 2006). Probability sampling provides for equal probability for each respondent to be selected (Aaker et al., 2013; Leedy & Ormrod, 2010; Malhotra et al., 2010). Non-probability sampling entrusts the researcher to select the respondents for the study, and there exists no way to determine whether each element of the population is covered (Leedy & Ormrod, 2010; Malhotra et al., 2010). This research adopted a two-staged sampling approach where Stage 1 consisted of simple random sampling (the pilot) and Stage 2 of stratified random sampling. Only respondents with a) a high probability that their e-mail addresses are valid, b) who have not opted out to the P-cubed communication before, and c) have not opted out on the Direct Marketing Association database, were available for selection in stage 1. Stage 1 selected consumers randomly for the pilot. Stage 2 eliminated respondents who were not between 18 and 60 years old as well as those that were deceased. Stratified random sampling was then performed where male and female respondents were given an equal opportunity to be selected across six financial affluent segments. Each financial affluent segment was given equal weighting. The financial affluent segments indicate the consumer’s level of affluence and financial product utilisation and provide for six broad segments indicating wealth. These segments have been statistically obtained by P-cubed through the analysis of 140 variables, including profession, income class, marital status, household makeup, gender, age, race, property value, geographic location, preferred communications channels, retail store cards, account/credit/loan/vehicle finance status, and insurance policies. An equal number of respondents was selected between male and female per financial affluent segment. This ensured representativeness of the sample in terms of the population.
This sampling strategy ensured limiting wastage through mailing to incorrect email addresses and an equal distribution between male and female as well as across all six affluent segments. The strategy ensured realistic response rates and limited the harassment factor of unsolicited e-mail communication, but it introduced a bias towards respondents with email addresses. Due to the fact that respondents selected may not belong to a financial services loyalty programme, and as such not be able to complete the questionnaire, the sample size needed to accommodate this aspect. A sample of approximately 50 000 respondents was selected in line with the stratified random sampling strategy, to ensure a response of between 250 and 500 at a rate of between 0.5 and 1%. This response rate was acceptable due to the fact that e-mailers did not a) associate with a specific brand that the customer had an affinity or relationship with, but b) requested respondent to partake in a PHD research study that was not specific to any company they may have had an association with, c) was sent to the open market without knowledge whether the respondent had any affiliation with any loyalty program or not, d) a limited time period of three weeks were given to respond and e) no follow-up e-mailers were sent out. A response rate of 0.5% was thus justified, taking into consideration that this study covered all loyalty programmes in the financial services industry, and was not specific to a single programme where members could be targeted. This methodology aids to the strength and depth of the findings in this industry.
The number of respondents in recent empirical research on loyalty programme effectiveness ranged between 740 and 371, with an average of 500. Seminal research conducted in three countries by De Wulf et al. (2001) included 371 respondents in total, 120 per country, while that of Wang et al. (2014) included 750 respondents from an international airline loyalty programme in Taiwan. This research aimed to acquire 266 respondents as the questionnaire consisted of 38 questions, and sought to achieve 7/8 respondents per question for statistical relevance (Zikmund, Babin, Carr, & Griffin, 2012). This is supported by sample size requirements for SEM, with sample size requirements of more than 200 or 5 to 20 times the variables being measured (Lei & Wu, 2007). The samples size is further supported by Hair (2014) indicating a requirement of 5 respondents per item. As the model consists of 36 items, 180 respondents would be sufficient.
Reliability
Data need to be assessed in terms of both reliability and validity of the measurement scales contained in the questionnaire of the research study before the data can be utilised in further analysis (Malhotra et al., 2010). The reliability of the measurement scales of the questionnaire refers to the consistency of producing similar results should the study be repeated (Leedy & Ormrod, 2010). Internal consistency reliability measures the correlation between responses to items making up a measurement scale, to establish to what extent these items measure the construct to which they relate. The correlation between these items is used to calculate an average correlation measure which indicates the reliability of the measurement scale (Kent, 2007). The internal correlation measure calculated indicates the extent to which all the items in a measurement scale measure the same construct (Cooper & Schindler, 2011; Leedy & Ormrod, 2010). The Kuder-Richardson Formula 20 coefficient is used in this regard for dichotomous responses, and the Cronbach’s alpha (α) coefficients for rating scales (Leedy & Ormrod, 2010; Malhotra et al., 2010; Shiu et al., 2009). The internal consistency reliability measures used in this research were Cronbach alpha and composite reliability which are suitable for rating scales and mostly reported in SEM analysis (Hair, Black, Babin, & Anderson, 2014). A coefficient value of 1 for Cronbach’s alpha (α) or the composite reliability coefficient signifies perfect reliability, 0.80 to 0.96 signifies very good reliability, 0.70 to 0.80 indicates good reliability and values lower than 0.60 signify poor reliability (Shiu et al., 2009).
The reliability measure adopted for this research was internal consistency reliability, since questionnaires could only be presented to respondents once, and the measurement scales included multiple items. Cronbach’s alpha was used to confirm the items measuring every construct, due to the Likert-scale responses obtained. It confirmed uni-dimensionality of each construct with a Cronbach alpha (α) larger than 0.7 through a pilot test of the questionnaire. Composite reliability measures were also obtained to confirm the reliability of all measurement scales (Hair et al., 2014).
1 Introduction
1.1 Background
1.2 Problem statement
1.3 Purpose statement
1.4 Research questions
1.5 Research objectives
1.6 Scope and definitions
1.7 Importance and benefits of the proposed study
2 Literature review
2.1 Introduction
2.2 Loyalty programme effectiveness
2.3 Loyalty programme design
2.4 Customer relationship management
2.5 Customer loyalty
2.6 Moderating variables reward type and timing
2.7 Scope of the research
2.8 Theoretical model
3 Research design and methodology
3.1 Introduction
3.2 Research paradigm
3.3 Research design
3.4 Information type
3.5 Research type
3.6 Population and sampling
3.7 Units of analysis
3.8 Data collection methods
3.9 Data analysis methods
3.10 Quality and ethics
4 Results
4.1 Introduction
4.2 Questionnaire response
4.3 Sample profile of respondents
4.4 Participation in a financial services industry loyalty programme
4.5 Proposed theoretical model
4.6 Measurement model
4.7 Structural model
4.8 Moderation analyses
5 Discussion
5.1 Introduction
5.2 Review of the findings
5.3 Limitations of the research
5.4 Future research
5.5 Academic and managerial contribution
5.6 Conclusion
6 List of references
