PART I: POPULATION AND SAMPLING

The following paragraphs describe sources of data, purposes and the tests undertaken using each piece of data in Part I. The sampling frame is also explained, as well as the data collection instruments.

The first phase of Part I was taken as pilot research and used perceptions and experiences of people involved in commercial microfinance in the industry worldwide (Thiagarajan & Zairi, 1998). The target group was drawn from informed international players committed to promoting commercialisation and/ or those responsible for lending decisions.

The population was defined as proponents of commercialisation and interested commercial actors in microfinance. The panel of experts list was drawn largely from 117 participants of a conference on challenges to microfinance commercialisation convened at the World Bank in June 2001 and investors list (CGAP, 2002c). The objective was to get agreement (clarify importance) on the comprehensive set of factors and practices that are believed by the wider microfinance fraternity/academics and industry experts to have an impact on access to commercial funds. Simple stratified random sampling was applied on a broad-based group of industry 'experts' representative of proponents of commercial microfinance, from operational regions in Africa, Asia, the Americas and Europe. From each region a number of experts were randomly selected. The sampling process was augmented with the author's scan to ensure industry coverage. This group formed the respondents of the survey instrument. The respondents of the 53-item questionnaire included:

• Lenders or fund providers; • Microfinance technical advisors;

• Donors and national government agencies that provide funding to MFIs; • Advisors and consultants in microfinance;

• Social investors; • Rating agencies; and

• Bankers involved in lending to microfinance.

3.2.1 Part I: Survey design and success factor determination

The literature provided an applicable list of potential success factors in the context of the microfinance and money lending industry. A critical aspect in the evolution of a fundamental theory in any management concept is the development of good measures that enable the researcher to obtain valid and reliable estimates of the domain of interest (Sureshchandar et al., 2002). The development process began by first substantiating adequate representation of the constructs; with the aim of identifying relevant interventions (valid factors) that are vital for success in commercial lending. Based on a comprehensive study of economic literature, finance and banking theory, the factor items were assembled.

A pilot questionnaire was designed to measure the individual’s perceptions of the relative importance of a set of possible factor considerations for commercial lending. The initial questionnaire was presented to academics and other microfinance reviewers for refinement on construct and face validity (Kelsey & Bond, 2001; Goosen, 2002; Sureshchandar et al., 2002). This group was used as a control group to confirm validity of the content for the list of 53 potential success factors. This exploratory approach was intended to ensure a complete list of commercial lending practice criterion dimensions. A final list of 53 potential success factors of MFI access to

commercial funding was collectively identified. As expected, these factors are quantitative in nature although some are represented in the 33 quantitative variables used in Part II of the study. A sample of the questionnaire identified as MEP (Microfinance Experts Panel) can be found in Appendix E.

3.2.2 Part I: Survey framework and approach

The Microfinance Experts Panel (MEP) 2002 survey document consisted of three parts: i) Part one contained 53 potential success factors;

ii) Part two consisted of question number 54 meant to test the completeness of the dimensions of commercial microfinance.

iii) In recognition of the disparity of evaluation criteria, Part three (question 55) sought to find the respondents' experience of the five most important considerations in lending practice.

In the MEP 2002 questionnaire, the 'experts' were asked to indicate the importance of each of the 53 potential success items on a Likert scale of 1 to 4, ranging from ‘Not important’ to ‘Very important’ respectively. A rating of ‘0’ on the scale provided for non-response or ‘No Opinion’ which was also a measure of item inappropriateness/validity of the item. The Likert measurement examined the respondent's perception and experience of each item's importance rating to commercial lending decision.

The survey used a personal contact approach in collecting the views of informant respondents (Sureshchandar et al., 2002). That is, respondents were personally contacted and the survey explained to them in detail. An attached letter solicited and exalted the recipients to participate in the study. The internet was used as the method of gathering data, especially the email facility. This method was chosen because of the advantage of sending the survey document to a large number of respondents spread across the globe simultaneously and cheaply.

A total of 117 emails were sent to MEP experts in 17 countries that formed the operational base of the respondents. An attached official letter (see Appendix F) introduced and explained the purpose of the study (Chen, 1999). The respondents were asked to contact the author for any clarification, and indicate their consent for participation. From these 117 contacts, a total sample of 44 respondents committed to participate in the survey after periodic follow-ups. Securing agreement to participate was not easy. The MEP 2002 survey document was sent to the 44 experts in the sample, with clear instructions. A final usable sample of 36 replies was returned representing a 30.7 per cent response rate.

3.2.3 Part I: Factor analysis

Factor analysis method was employed to identify factors that contribute to success in commercialisation for MFIs in Africa (Hartungi, 2007; Zapalska et al., 2007). Factor analysis aims to summarise information requirements and unearth underlying factors that illustrate relationships among a set of interrelated items. This statistical approach was selected because of its ability to identify a small number of factors that are critically linked to the domain of interest and to group similar structures together. That is, it helps to understand interrelationships of factor items as represented by factor loadings (Zhang, Waszink & Wijngaard, 2000; Lekkos, 2001; Sureshchandar et al., 2002). Besides, it is also easy to use and interpret.

A key objective in undertaking factor analysis in this study was to reduce the set of variables to a smaller number by summarising the information contained in the number of original items/predictors with minimum loss of information (Child, 1970; Chen, 1999; Hopkinson & Pujari, 1999; Jain, 2001; Nunes, 2002; Liu & Lee, 1997). The basic assumption is that each variable can be expressed as a linear combination of hidden factors that affect the variable and possibly other variables (Jain, 2001). The other objective was to avoid both the problem of multi-collinearity among explanatory variables and the possibility of some variables masking others. The author had a feeling that the variables were too many in the analysis and therefore without finding a way of focusing on the critical ones, strategic fit in the model could be lost.

Given that the 53 success factors of commercial lending were pre-determined by the author it was necessary as a first step to, firstly, reduce the factors items to only those that are important for further investigation in subsequent tests and validation using other methods, and secondly, to use this analysis to identify suitable dimensions for commercial lending (Lekkos, 2001). In the former case, factor analysis was then used as the method of identifying the best proxies to be included as part of the 33 variables (used in Part II of the study) in developing a prediction model.

Factor analysis uses principal component extraction method on raw data (Hartungi, 2007; Antony, Leung & Knowles, 2002; Mazzarol, 1998; Goosen, 2002; Child, 1970; Chen, 1999). In the exploration process, factor analysis brings out the relationships between variables involved through a rotational process called varimax. A varimax matrix indicates if there are common factor structures by use of values called factor loadings that usually present variable relationships, strength of correlation between variables and provide basis for data interpretation. This method uses Eigenvalues to determine importance or suitability of data for factor extraction, the closer this value is to 1.00. However, only factors having Eigenvalues greater than 1 are considered significant for factoring (Harman, 1976; Lekkos, 2001; Mazzarol; 1998; Goosen, 2002; Child, 1970; Chen, 1999).

Factor items must relate to each other for an appropriate factor model. Where the correlation is too small (as shown by factor loadings <0.55), it is unlikely that the items have some property in common. Hence such items are not grouped together. The procedure is able to indentify suitable factor models that meet the criterion of more than one Eigenvalue, as per Kaiser's criterion (Antony et al., 2002; Mazzarol, 1998; Lekkos, 2001; Chen, 1999; Nunes, 2002). CSF approach helped to understand the importance attached to the set of evaluation criteria used by industry players. Under factor analysis method, the interpretation of factor loadings within a model is crucial and proceeds as follows:

• Absolute factor loadings greater than 0.3 are considered significant; • Loadings of 0.4 as important;

• If loadings are 0.5, 0.6 or greater, they are considered very good and significant (Antony et al., 2002; Goosen, 2002; Zhang et al., 2000; Hopkinson & Pujari, 1999).

High factor loadings suggest that the variables or items are critical and indeed such variables are best choice representatives of the corresponding factor (Antony et al., 2002; Mazzarol, 1998; Lekkos, 2001; Chen, 1999; Nunes, 2002). The higher the value of the loading the better, and indeed these items provide the flavour of the factor and in naming of the factor dimension in the selection process. Results of the analysis are presented in Chapter four.