Research Method: Data Collection

As mentioned above, two main data collection methods are utilised in this study, namely questionnaire and interviews. The following sections explore the details of both methods of data collection.

The survey questionnaire   A questionnaire is a research instrument which consists of a series of questions and other prompts for the purpose of gathering information from respondents. It is very popular, since many different types of primary data can be collected including attitudinal, motivational, behavioural and perceptive aspects of the subject being studied (De Vaus, 2002).

In designing the questionnaire, it is important that the questions address the aims of the study. The questionnaire is considered to be effective if it suits the research objectives and questions. A good questionnaire has to be clear and unambiguous, and encourage respondent participation (Creswell, 1994).

If properly designed and implemented, surveys can be an efficient and accurate means of determining information about a given population. Results can be provided relatively quickly and, depending on the sample size and methodology chosen, they are relatively inexpensive. Survey questionnaires have many advantages over other methods of data collection (De Vaus, 2002).

Some advantages of questionnaires can be listed as follows:

1. The responses are gathered in a standardised way, so questionnaires are more objective, certainly more so than interviews;
2. It can be completed at the convenience of the respondents;
3. Generally it is relatively quick to collect information using a questionnaire;
4. Low cost of data collection and processing;
5. As the questionnaire can be anonymous, it gives the respondents freedom and encouragement to answer questions honestly, especially sensitive questions; and
6. It can cover a large sample of respondents at the same time.

On the other hand, the questionnaire method has some disadvantages which have to be taken into consideration. Oppenheim (2001) highlights the following problems:

1. Some respondents may not be willing to answer the questions;
2. Respondents may answer superficially especially if the questionnaire takes a long time to complete. The common mistake of asking too many questions should be avoided;
3. The validity of the responses may be compromised by a biased view of the respondent;
4. No opportunity to correct misunderstandings or to probe, or to offer explanations or help;
5. Questionnaires are standardised so it is not possible to explain any points in the questions that participants might misinterpret; and
6. Respondents' inability to answer a question might affect the response rate and reliability.

Despite the disadvantages of the questionnaires, they are rather useful and efficient in aiming to collect data related to the perceptions and opinions of individuals on a particular subject. This study utilised a questionnaire survey in collecting primary data from bankers and financiers in the form of their opinions and perceptions in mapping out the risks and aspects of risk management in IBF. Thus, a questionnaire survey is considered one of the main methods of primary data collection for this study.

Open-ended versus closed questions   In terms of questionnaire design, the questions included may be divided into those which are ‘open-ended’ and those which are ‘closed’. In open-ended or free-response questions, respondents are free to reply to the questions in any way they wish and the answers have to be recorded in full. In ‘closed’ questions, respondents are offered a choice of alterative replies and they must reply in one of a predetermined number of ways, such as ‘yes’, ‘no’ or ‘don't know' (De Vaus, 2002). The advantages of using closed-ended questionnaires are that this technique is easier and quicker for the respondents to answer; they require no writing. In addition, closed-ended questionnaires are easier to code and statistically analyse as quantification is straightforward. Disadvantages of closed questions are loss of spontaneity and expressiveness, and perhaps the introduction of bias by ‘forcing’ respondents to choose between given alternatives or by making them focus on alternatives that might have not occurred to them (Oppenheim, 2001). On the other hand, open-ended questions have many advantages, stemming from the fact that respondents are encouraged to structure the answer as they wish. This provides a means for obtaining information which cannot be obtained adequately by the use of a closed question (Creswell, 1994). Another advantage of the open-ended question is the information which the respondents indicate with respect to their level of knowledge or degree of expertise. The disadvantage of open-ended questions is that they produce a mass of different words meaning the same thing, or a number of similar words meaning different things. It can therefore be stated that open-ended questions are easy to ask, difficult to answer, and more difficult to analyse. Oppenheim (2001) explains that these free-response questions require drawing up some system of categories known as coding. The design of such coding framework and the actual coding operation require trained staff and are extremely time-consuming; for this reason researchers have to curb their desire to have too many open-ended questions (Oppenheim, 2001).

Level and characteristics of measurements   The level of scales measurement of a variable in statistics is a classification that is used to describe the nature of data contained within numbers assigned to objects and therefore within the variable. According to De Vaus (2002), there are three main levels of measurement scales. These are:

1. Nominal scale, in which a distinction between categories of a variable can be made, but one cannot rank the categories in any order. The nominal scale is used to measure qualitative variables and yields frequency data that may be subjected to non-parametric statistical tests, such as gender.
2. Ordinal scale, in which it is meaningful to rank the answers by categories, but it is not possible to quantify precisely how much difference there is between categories (such as more than, less than, equal to).
3. Interval/ratio scale, in which ranking of categories can be made and it is also possible to quantify the differences between the categories precisely. Likert scales are very commonly used with interval procedures.

Sampling in the questionnaire   A sample is a small selected portion of the whole population. According to Bryman (2008:85), “a sample is the segment of population that is selected to be investigated”. The size of the sample must be sufficient in order to represent the population which the study is intended to investigate. The sample size depends on the homogeneity of the population. If the pilot study indicates that there is a considerable heterogeneity of the population, then it is important to choose a larger sample. As Robson (2011:164) contends, if the population is heterogeneous and the main interest of the study is to generalise the findings to the population from which the sample was drawn, then a larger sample is needed. In addition, a larger sample size will decrease the probability of sampling error.

According to Bryman (2008), sample sizes smaller than 500 cases and larger than 30 cases tend to be suitable for most studies. As far as the survey sample in this study is concerned, there were obviously some real cost and time constraints which limited the sample size. The target population is the wider group of banking professionals worldwide, both Islamic and conventional banking practitioners, whose perceptions about risk management in Islamic banking could shape the outcome of this study. The significant diversity and dispersion of the population meant the time and cost constraints would be unusually high due to the inherent extra complications associated with such a target population. Caught between these challenges and the strong desire to make the sample size as large as possible, the researcher completed 77 questionnaires out of which five were not fit for purpose. The sample size for this study, upon which both descriptive and inferential statistical analysis will be performed, is therefore 72 questionnaires.

There are different sampling strategies such as simple random sampling, systematic sampling, stratified sampling, cluster sampling, panel sampling and others, each with their own advantages and disadvantages.

According to Robson (2011), a variety of sampling methods can be employed, individually or in combination. Factors commonly influencing the choice between these designs include:

1. Nature and quality of the research;
2. Availability of auxiliary information about units on the research;
3. Accuracy requirements, and the need to measure accuracy;
4. Whether detailed analysis of the sample is expected; and
5. Cost/operational concerns.

It should be noted that the snowball sampling method is used for this research, which is a sampling method used to obtain research and knowledge from extended associations through previous acquaintances; it uses recommendations to find people with the specific range of skills that has been determined as being useful. It is referred to metaphorically as snowball sampling because as more relationships are built through mutual association, more connections can be made through those new relationships and a plethora of information can be shared and collected, much like a snowball that rolls and increases in size as it collects more snow. Snowball sampling is a useful tool for building networks and increasing the number of participants. However, the success of this technique depends greatly on the initial contacts and connections made (Babbie, 2010).

Snowball sampling has a number of advantages over other sampling methods. It is possible for the surveyors to include people in the survey that they would not have known. It is also useful for locating respondents of a specific population if they are difficult to locate. The advantage of this is that the researcher can quickly find respondents who are experts in their fields. This leads to having the most well-known experts for the sampling group, and also can help the researcher find lead users more simply (Babbie, 2010). The method is, however, heavily reliant on the skill of the researcher in conducting the actual sampling, and that individual's ability to vertically network and find an appropriate sample. To be successful requires previous contacts within the target areas, and the ability to keep the information flow going throughout the target group. Identifying the appropriate person to conduct the sampling, as well as locating the correct targets is a time-consuming process which renders benefits that only slightly outweighing the costs. Another disadvantage of snowball sampling is the lack of definite knowledge as to whether or not the sample is an accurate reading of the target population. By targeting only a few select people, it is not always indicative of the actual trends within the result group (Babbie, 2010).

The experience in conducting this research indicates that due to the nature of the research as well as the subject matter, the snowball sampling strategy proved to be a very successful strategy.

Pilot study   A pilot study is a small-scale preliminary study conducted before the main research in order to check the feasibility or to improve the design of the research. The questionnaire must be evaluated rigorously before final administration (De Vaus, 2002). A pilot test is important as it highlights any shortcomings before the document is fully launched. The objective is to check the overall presentation, clarity and reasonableness in terms of the length of the questions and the depth of the information sought (Bryman, 2008). Also, piloting is important to check the uniformity of interpretation of each respondent, and whether respondents are answering the questions correctly (Dillman, 2000).

The drafted questionnaires for this study were first pilot tested on a group of 10 bankers working in London. The respondents were asked the following questions:

1. Is the questionnaire too long?
2. Were the instructions clear?
3. Were any of the statements ambiguous?
4. Did they find any of the questions sensitive?
5. Whether they had any comments and suggestions.

The feedback of this piloting provided the following observations:

1. Q7: Some respondents did not understand what is meant by Displaced Commercial Risk.
2. Q7: There was some ambiguity concerning which risks fall under market risk.
3. Q11: Some sub-statements are unclear.

This feedback was then incorporated into the questionnaire and a further random sample of seven bankers was selected for second piloting. This time the results from the pilot test resulted in no noticeable difference to the original questionnaire. This produced the final version used for this research.

Interviews   An interview is a qualitative research technique that allows face-to-face interaction. It involves asking questions and receiving answers from respondents in an identified research area. As compared to questionnaires, it can lead to increased insight into respondents' thoughts, feelings and behaviour rather than simple answers.

Robson (2011) classifies interviews into structured, semi-structured and unstructured interviews. The different types can, to some extent, be linked to the depth of response sought. Oppenheim (2001) classifies interviews into essentially two kinds:

1. Exploratory interviews, depth interviews or free-style interviews; and
2. Standardised interviews such as used, for example, in public opinion polls, market research and government surveys.

Oppenheim (2001) provides some advantages of interviews compared to questionnaires:

1. Improved response rate;
2. Interviews can give a prepared explanation of the purpose of the study more convincingly than a cover letter;
3. Flexibility, as questions that are inappropriate to a particular interviewee can be omitted or additional ones included;
4. Gives the interviewer the opportunity to probe further into a subject to extract more details from the interviewee;
5. In interviews it is easier to keep the attention of the respondent; and
6. Enhancing data validity: due to human interaction, interview results have less chance of being biased and unreliable.

However, interviews have some common disadvantages. Creswell and Clark (2007) argue that the disadvantages of using interviews are to some extent a reflection of their advantages. Obviously, interviews are much more expensive than questionnaires. The larger or the more dispersed the sample, the greater the cost. Travel costs and call-backs add to this. The cost factor also enters the data processing stage: since interviews are used particularly where many open-ended questions have to be asked, there will be a major and costly coding operation allied to the use of interviews. Analysis of interview data may be challenging as the data collected usually contains non-standard responses. Moreover, interviews tend to be time-consuming, as they require lots of preparation and coordination with the interviewees.

Operationalising interviews   This study collected qualitative data by conducting exploratory semi-structured interviews because exploratory interview is essentially heuristic, which helps to develop ideas and research hypotheses rather than to gather facts and statistics (Oppenheim, 2001). In addition, Bryman (2008) describes the in-depth interview as an engaged conversation between two people. In the interview, the researcher puts themselves in the participant's situation to try to understand that person's point of view (De Vaus, 2002). The researcher needs to listen and pay constant attention to the participants as they are responding, repeatedly attempting to understand the meaning of what is being said and how the person has shaped their perspective. In this way, interviewing is more than ‘collecting data’. Furthermore, interviewing allows the researcher and the participant to connect in a profound way, reducing the distance between them (Creswell and Clark, 2007). This type of interview is often unstructured and therefore permits the interviewer to encourage a respondent to talk at length about the topic of interest in a flexible approach (Robson, 2011).

(i) Interview structure   The interview script was developed within the context of the original research questions and hypotheses. The script helps to guide the interview sessions. The interview is divided into six main parts corresponding to the six research parts introduced under the structured approach in the Interview Analysis section. This facilitated data collection and analysis. The interview script covers the same topics as the questionnaire, as the main purpose of the semi-structured interview is to prove or disprove the conclusions derived from the questionnaire data analysis.

(ii) Administration of the interview and sampling   The snowball sampling method, as discussed before, was used for the interview sampling to obtain perceptions and knowledge from an extended network of respondents, through previous acquaintances. The researcher utilised his network of initial contacts and connections to find participants with valuable experience and knowledge in Islamic banking and risk management. From June 2010 to January 2011, in-depth semi-structured interviews were completed with 37 Islamic banking professionals. The interviewees included a mix of senior banking executives and heads of business units who work at either Islamic banks or conventional banks with Islamic activities/windows, researchers, academics, Shari'ah scholars, consultants and specialised analysts at rating agencies. Five of the respondents in the interview were also included in the sample for questionnaires.

Out of the 37 interviews conducted, four interviews were discarded because the researcher felt that the interviewees were biased in their replies or were not well informed about the issues discussed. Therefore, the final sample included only 33 respondents (five of whom were included in the questionnaire sample) who were knowledgeable about the topic and contemporary developments, and whose replies could be taken, with a high level of confidence, as bias-free.

Out of the 33 interviews that comprise the final sample, 21 interviews were conducted face-to-face, either in London or in the participants' cities. The researcher utilised his numerous business trips to arrange for these face-to-face interviews. Seven interviews were conducted via teleconference, and five interviews were conducted via video conference facilities. Both the teleconferences and the video conferences were dialled from the researcher's primary location in London. The final sample was diverse both geographically and by participants' roles.

Interviews were conducted to gather primary data for this research to support the primary data generated through quantitative method, namely the questionnaire. It should be noted that the preparation for the interview was carefully planned and professionally conducted. When possible, these interviews were audio recorded with the permission of the interviewee. When recording was not possible because of the spontaneous arrangement of the interview, notes were taken in shorthand by the interviewer. Even when an interview was being recorded, shorthand notes were also kept.

However, the interview sample was not as big as the questionnaire sample for a number of reasons:

1. In-depth interviews take much more time than structured questionnaires;
2. Interviews require one-on-one interaction;
3. Travelling to meet interviewees was difficult due to time and cost constraints;
4. Despite exploiting video conference and teleconference facilities to conduct some interviews, several potential interviewees did not have the time or the wish to participate; and
5. Some potential interviewees were located in remote time zones (like America and Southeast Asia), which added to the difficultly of arranging suitable interview times.

Despite these difficulties, the sample size and distribution are within acceptable limits and therefore allow for reliable data.

Table 6.2 combines the geographic distribution with the position of interviewees. It is obvious that interviewees represent a wide range of expertise and roles.

Validity and reliability of the data   Validity refers to whether the questionnaire or interviews measure what they intend to measure, which is crucial, regardless of the method used to collect such data, as invalidity makes the results worthless. Validity depends largely on how honest and accurate the responses given by the respondents are, which is a difficult factor to measure.

Reliability refers to the consistency of the questions. This means that if the research were to be carried out by other independent researchers employing the same methodology and strategy, they would arrive at a similar conclusion, all other things being equal (Creswell, 1994). If a method of collecting evidence is reliable it means that anybody using this method, or the same person using it at another time, would come up with the same results. According to Oppenheim (2001), the key components of data reliability include consistency, precision and explicability of results, which suggests that the researcher should be consistent when collecting the data and should aim for a high degree of precision and accuracy, which of course will be subject to many factors outside the control of the researcher. However, the researcher should try to minimise bias in the data collection process.

In this study, the validity and reliability of the data were proved acceptable for a number of reasons:

1. The use of multiple methods of data collection;
2. The use of a cover letter explaining the purpose of the research and assuring the confidentiality of responses;
3. The questionnaire was subjected to a sequence of pilot tests which involved every question being scrutinised and edited when necessary;
4. Collected raw data was screened and filtered for errors;
5. Personal close follow-up with the questionnaire respondents via telephone calls and emails to ensure elimination of any confusion or lack of clarity that might arise;
6. Checking consistency of answers in questionnaires through multiple questions asking about the same point;
7. Five questionnaires were excluded from inclusion in the final sample as the researcher felt the inconsistency in the answers might spoil the data;
8. Four interviews were discarded for the same reasons;
9. Personally assuring the interviewees involved in the semi-structured interviews of the anonymity of both their identity and personal responses; and
10. Sending the draft questionnaire and interview script to a number of PhD students and academics to seek their opinions on the proposed drafts and their potential effect of the data validity and reliability.

Cronbach's alpha test   Cronbach's alpha is the most common form of internal consistency reliability coefficients; it ranges in value from 0 (when the true score is not measured at all and there is only an error component) to 1 (when all items measure only the true score and there is no error component). The higher the value of alpha, the more reliable the scale is. As a rule of thumb, alpha should be at least 0.7 (De Vaus, 2002).

Table 6.3 reveals that the Cronbach's alpha coefficient for respondent groups for the 105 items that used the scale was 0.912 (>0.70), which should be taken as confirming the reliability of the contents of the questionnaire used in this study.

Research Method: Data Analysis

Data analysis is one of the most difficult parts of the research process. Having chosen the appropriate method of analysis, the choice of statistics is affected by the method of analysis itself, the level of measurement of variables and the complexity of research questions (De Vaus, 2002). This section provides a detailed description of the methods used to analyse the assembled qualitative and quantitative data.

Quantitative data analysis   An initial screening of the questionnaire was carried out regarding the completeness and eligibility of the responses. As a result of this initial screening, only 72 out of the 77 returned questionnaires were included in the final sample.

The questionnaires were numbered and data was checked for errors. Cross-tabulations were carried out to check the inconsistency of the data (skip errors, missed answers, values outside the range). In addition, the frequency distribution for all question items was checked and corrected if required. Once all the errors had been corrected, raw data was coded and saved as a new master file for statistical analysis. The final complete sample was then entered directly into the Statistical Package for Social Science (SPSS) programme. Initially, all the variables were created, and then the actual questionnaires were entered. This enables the data to be created in statistical tables in order to facilitate inferential statistical analysis.

Most questions in the questionnaire are designed along a five-point Likert scale in order to measure the respondents' opinions about sets of statements, which make codification and analysis of the data easier and more efficient.

The following statistical techniques are utilised:

(i) Descriptive analysis methods   Descriptive statistics are summaries of data, which can be tabular, numerical or graphical. Different types of descriptive statistics such as the mean, the mode, the median, the frequency distribution, the minimum, the maximum and percentages are calculated, and these are presented in Chapters 7, 8, 9 and 10.

(ii) Non-parametric tests   The main objective of statistical analysis applied in this research is to test whether there are significant differences in perceptions of respondents through various control variables at the overall sample level and among various groups of respondents. Significance testing is usually concerned with accepting or rejecting hypotheses or propositions, and can be conducted by parametric and non-parametric tests. Non-parametric tests were considered to be appropriate for this study because the data collected was mainly nominal and ordinal; the responses were not normally distributed; and the sample size was relatively small. Parametric tests usually suit samples which are drawn from a normally distributed population and data collected on an interval or ratio scale (Hebel, 2002).

The following non-parametric techniques were used:

(a) Chi-Square test   The Chi-Square test is used to measure the association between dependent variables and independent variables (Saunders et al., 2007). The test is appropriate for testing the goodness-of-fit variables because the test can be applied to determine whether or not an observed set of frequencies matches some expected set of frequencies. The Chi-Square test was used to verify the existence of any significant differences in the responses regarding the degree of response for each statement. A significance level of 5% is used for this study as justification for rejecting the null hypothesis.

(b) Kruskal-Wallis and Mann-Whitney U test   The Kruskal-Wallis (K-W) test is a non-parametric method for testing equality of population medians among groups. It is identical to a one-way analysis of variance with the data replaced by their ranks. It is an extension of the Mann-Whitney U test to three or more groups.

The K-W test allows researchers to measure the possible differences between two or more groups in relation to particular control variables. In this study, the K-W test of significance was intensively used for the inferential statistical analysis to test the impact of control variables like region, country, respondent's position, nature of financial institution, nature of activities and accounting standards on the perception of survey participants. The significance level used for this Kruskal-Wallis test is 5%.

(c) Spearman's Rank Correlation Coefficient   Spearman's Rank Correlation Coefficient is a measure of correlation, which shows how closely two sets of data are linked. It can be done only on data that can be put in order, highest to lowest (Bryman, 2008). In this study, the Spearman's Rank Correlation Coefficient is used to test whether there is correlation between different groups of respondents.

(iii) Factor analysis   Factor analysis as an inferential statistical analysis tool is used as a data-reduction method to reduce a large number of variables to a small number of factors to facilitate the process of summarising the data which has been collected. Pallant (2007) states that in order to conduct factor analysis the Kaiser-Meyer-Olkin (KMO) test and Bartlett's test need to be conducted. For factor analysis to be considered appropriate, the Bartlett's test of sphericity value should be significant (p < 0.05), while for the KMO test, the suggested minimum outcome must be at least 0.6 (KMO score ranging from 0 to 1). The KMO test's benchmarks are as follows: if the KMO measure is in the 0.90s, the sampling is considered marvellous. If the outcome is in the 0.80s, then the sampling is considered meritorious; if it is in the 0.70s, then the sample is middling; if it is in the 0.60s, then the sample is mediocre; if it is in 0.50s, then the sample is deemed miserable; and lastly, if it is below 0.50, then the sample is unacceptable (Pallant, 2007).

In this study, factor analysis was used for Questions 11 and 16 to test whether the observed variables can be explained largely or entirely in terms of a much smaller number of components. This also helps to organise large numbers of factors into components generated by the study.

(iv) MANOVA   Multivariate analysis of variance (MANOVA) is a generalised form of univariate analysis of variance (ANOVA). It is used when there are two or more dependent variables (Tabachnick and Fidell, 2006). MANOVA tests aim to establish whether mean difference among groups on a combination of dependent variables is likely to occur by chance (Pallant, 2007). In this study, after conducting factor analysis, the MANOVA test was computed for Questions 11 and 16 in order to investigate if there is any significant difference between the component groups identified by factor analysis in relation to some control variables. This helps to locate the impact or significance of each control variable on the generated distribution and components.

(v) Friedman test   The Friedman test is used to find a tendency for some variables to receive higher ranks than others, for example, assigning the ranks of 1 to 10 to the most preferred and least preferred variables respectively (Creswell, 1994). The Friedman test ranks the scores for each of the cases and then calculates the mean score for each sample. If there is no significant difference between the samples, their mean score ranks should be similar (Bryman, 2008). The Friedman test determines whether the rank totals for each condition or variable differ significantly from the values which would be expected by chance (Bryman, 2008).

In this study, the Friedman test was used in Questions 9 and 10 to examine whether there is a significant difference between the respondents' perceptions in ranking the given options. The significance level used for this Friedman test is 1%.

(vi) Interpretative analysis   In addition to these quantitative methods, an interpretative approach was employed to provide further meaning to the results of the questionnaires and in-depth understanding of the issues in an integrated manner. This interpretative approach introduces interaction between the primary data findings and the literature review in order to provide better understanding of the findings of the questionnaire analysis.

Qualitative data analysis   Analysis of qualitative data collected through semi-structured interviews is more complex and demanding than analysis of quantitative data. This means that unless great care is taken in analysis, it may cause real harm to the research itself (Robson, 2011).

Since the interviews were based on open-ended questions, the researcher transcribed all recorded interviews and read the interview notes and then transferred them into segments representing complete thoughts on a single question or topic, in line with the original research questions. All transcribed interviews were broken into coded segments representing complete thought statements. Answers were codified according to the most common responses provided by interviewees. After coding, the interview segments were transferred from word-processing format to a spreadsheet for further analysis.

As Charmaz (1983:114) states “Codes serve to summarise, synthesise, and sort many observations made of the data…coding becomes the fundamental means of developing the analysis…Researchers use codes to pull together and categorise a series of otherwise discrete events, statements, and observations which they identify in the data…At first the data may appear to be a mass of confusing, unrelated, accounts. But by studying and coding, the researcher begins to create order.” Thus, the qualitative data collected through interview was analysed by the use of coding analysis, which was conducted manually rather than with the help of software such as n-vivo.

The findings generated through coding analysis were also subjected to interpretative analysis with the objective of developing a better understanding of the data and findings. This social constructivist-oriented method helps to develop an integrative approach to the data to render a rich qualitative analysis.