CHAPTER 6 Research methodology 6.1 INTRODUCTION I n order to determine how to utilise CPD of ethics amongst prospective CAs at an undergraduate level, one has to be able to develop a ‘model’ or ‘method’ of approaching this ‘learning process’. Moreover, this model needs to be implemented during a student’s undergraduate learning experience. Before one can actually commence with the development of this so-called ‘integrated teaching and learning model’, one has to be certain that students grasp the issues of CPD and ethics within a holistic process. Chapter 7 therefore provides an assessment of whether CPD of ethics is understood and possible to communicate to prospective CAs at an undergraduate level by focusing on the attitudes and perceptions of undergraduate students toward the CPD of ethics. Chapter 6, however, first provides the background to the research methodology to be used in the empirical analysis in Chapter 7. This chapter will initially provide the reader with some thoughts on research design, incorporating issues such as qualitative and quantitative research. The quality of measurement is also discussed looking at the issues of reliability and validity specific to this study. 217 Research methodology Next, the chapter provides an overview of the design of a questionnaire as an instrument to be used in an empirical analysis, which is then also investigated within the ambit of this empirical study. Discussions surrounding issues such as the questionnaire layout, the issue of pre-testing, sampling approach, administration, data analysis, research ethics and confidentiality are also provided. 6.2 RESEARCH DESIGN Planning an empirical analysis for research purposes requires the researcher to investigate various issues around the research design. For the purpose of this thesis, a mixed research paradigm employing a combination of both quantitative and qualitative research methodologies will be used. 6.2.1 Qualitative research Grounded theory, as a phenomenological or qualitative research paradigm, will be used in this thesis to describe the world of the prospective CA. In order to define it one may state that grounded theory “... uses a systematic set of procedures to develop an inductively derived theory about a phenomenon” (Collis & Hussey, 2003: 73). Data will therefore be gathered and analysed in order to generate hypotheses. 6.2.2 Quantitative research A quantitative, descriptive survey will be used to investigate opinions, attitudes and perceptions of third year B Com Accounting students at RAU on the utilisation of CPD of ethics amongst prospective CAs. Quantitative research or a positivistic paradigm (Collis & Hussey, 2003: 47) is a formal, objective, systematic process to obtain information and describe variables and their relationships (Burns & Grove, 1993: 26). Quantitative research is also a conclusive research format involving large samples and fairly structured data-collection procedures (Struwig & Stead, 2001: 4). The design will be quantitative in nature in that the strategies the researcher used to CHAPTER 6 218 Research methodology collect and measure data were in a numerical format. Measuring or measurement is simply the allocation of numbers to objects or events according to predetermined rules. Huysamen (1998: 7) defines it as “... the assignment of numbers or numerals, according to fixed rules, to persons or objects in order to reflect differences between them in the attribute or characteristic of interest.” The reason for the use of numbers, as is evident from Huysamen’s definition, is that numbers can show distinction between different mutually exclusive categories; numbers can indicate rank; and numbers can indicate equal intervals between successive numbers (Uys, 2003: 118) – i.e. measurement can describe characteristics of variables (Eiselen; Uys & Potgieter, 2005: 11). The definition furthermore encapsulates the issue of “... differences in attributes ...” This implies that the level of measurement for each variable may differ. Based on work done by Stevens (1946), the literature describes basically four levels (or scales) of measurement – namely, nominal, ordinal, interval and ratio measurement scales (Uys, 2003: 119). 6.2.2.1 Nominal scale Within nominal measurement (sometimes called categorical scales), the issue of mutual exclusiveness stands out. This means that when completing a questionnaire, the respondent has to choose one aspect relating to his/her circumstances. Examples of nominal measurements are gender, language or marital status (Bryman & Cramer, 1997: 56). 6.2.2.2 Ordinal scale As is the case with nominal measurement levels, ordinal levels are also distinguishable, with one difference: categories can now be sequenced in terms of ‘more’ or ‘less’. The respondent can, for example, provide additional characteristics about his/her career status or educational background (Bryman & Cramer, 1997: 56). CHAPTER 6 219 Research methodology 6.2.2.3 Interval scale Within this measurement level, equal intervals are placed between different categories. Numbers now correspond to differences in degree, and ranking of characteristics is also now possible (Bryman & Cramer, 1997: 56). This scale is very often used in surveys and questionnaires and will be predominantly applied to the empirical analysis in Chapter 7. 6.2.2.4 Ratio scale This measurement level is an extension of the interval measurement level with an addition: absolute zero values are now included. Examples are income, family size and age (Bryman & Cramer, 1997: 56/57). 6.2.3 Quality of measurement Various conditions for good measurement are found in the literature. Neuman (1997: 152-153) argues that good measurement must satisfy three conditions, namely: categories must be mutually exclusive; possibilities must be exhaustive; and measurement must be one-dimensional in order to measure a single construct. Further conditions for quality of measurement are reliability and validity. 6.2.3.1 Reliability Reliability demonstrates the consistency of measurement (Bryman & Cramer, 1997: 63; Uys, 2003: 122). This means that with repeated measurements, equivalent results must be found. Neuman (1997: 138-141) states that reliability is mainly tested in three ways. These are: stability reliability: i.e. whether the instrument will provide the same results over time; representative reliability: i.e. whether the instrument will provide the same results when applied to different sub-populations e.g. different age groups or CHAPTER 6 220 Research methodology gender groups; and equivalence reliability: i.e. whether the instrument will provide the same result when multiple indicators are used to measure a construct. For this study, the issue of equivalence reliability is of great concern, as multiple indicators were used. Equivalence reliability can be measured by using item analysis. With this test, the response to each item versus the response to all the items in the questionnaire is measured. The coefficient of reliability (Cronbach’s alpha or α) must be at least 0.7 or higher to be reliable (Bryman & Cramer, 1997: 63/64). 6.2.3.2 Validity Validity is the degree to which the measuring instrument actually measures what it is supposed to measure (Uys, 2003: 123). Validity can firstly be divided into internal and external validity. a) Internal validity In a quantitative study, the test for internal validity is how confident the researcher is that the independent variable is at least partly responsible for the variation found in the dependent variable (Smallbone & Quinton, 2004: 154). Cooper and Schindler (2003: 231-236) sub-divide internal validity into three ‘types’ of validity. These are: content validity – is the definition of what is to be researched represented in the measuring instrument?; criterion validity – how does a measuring instrument weigh up to other instruments if results are compared?; and construct validity – to what degree do different measurement indicators correspond to results from these indicators? Very often a process called factor analysis is used to determine construct validity (Uys, 2003: 124). Factor analysis was invented nearly 100 years ago by psychologist Charles CHAPTER 6 221 Research methodology Spearman, who hypothesised that the enormous variety of tests of mental ability – measures of mathematical skill, vocabulary, other verbal skills, artistic skills, logical reasoning ability, etc. – could all be explained by one underlying ‘factor’ of general intelligence that he called g (Williams, Zimmerman, Zumbo & Ross, 2003: 114). Neuman (1997: 142) and Bryman and Cramer (1997: 66-67) also distinguish between Cooper and Schindler types of validity but include a fourth type: face validity, i.e. whether the measuring instrument is measuring what it is supposed to. b) External validity External validity concerns whether the results could be applied to other contexts or situations and to what extent this might be possible. In quantitative studies the representativeness of the sample is the key issue in generalising about a larger population (Smallbone & Quinton, 2004: 154). The validity of a sample depends on two considerations – namely, accuracy and precision. Accuracy is the degree to which bias is absent from the sample (Cooper & Schindler, 2003: 181). Even though precision is a criterion of a good sample design, no sample will be fully representative of the population in all aspects (Cooper & Schindler, 2003: 181). 6.3 6.3.1 QUESTIONNAIRE DESIGN Introduction Polit and Hungler (1991: 189) defined a survey as “... designating any research activity in which data is obtained from a specific population for the purpose of examining characteristics, opinions or intentions of that population.” The survey method of data collection requires (Struwig & Stead, 2001: 41): the application of questionnaires for data gathering; the population (or universum) being studied must be accurately described; CHAPTER 6 222 Research methodology the sample should be representative of the population; the scientific character of the data should not be adversely affected or influenced by bias; and data gathered should be systematically organised in order to make valid and accurate interpretations. Questionnaires (or surveys) have formed and will continue to form the bases of many different research projects. Questionnaires are considered an appropriate method of research if the ‘individual’ is the unit of analysis (Booysen, 2003: 129). A questionnaire often provides unique ways of ascertaining attitudes, opinions, perceptions and reports of individual behaviour (Booysen, 2003: 129). Using questionnaires as a measuring instrument often compromises the validity of the results as ‘direct intervention’ is not possible in many instances, for example mail surveys. However, when the survey is done in a direct, personal format (as was done in this study), the validity of the results tends to increase. This is done by, for example, administering the questionnaire to a group of people where respondent behaviour can be controlled (Booysen, 2003: 129). 6.3.2 General guidelines in constructing questionnaires Various aspects to be considered in questionnaire design can be found in the literature (for example Booysen, 2003; Struwig & Stead, 2001; Cooper & Schindler, 2003; Eiselen et al., 2005). Some of these are: ensuring that the research problem fits the questionnaire contents – it does not leave space for respondent assumptions; taking note of style and paying attention to language and formulation – language must be unambiguous and clear; looking at the order of the questions – that it is logical and does not allow for bias in answers; and paying attention to the format of the questionnaire – think ahead to the way data will have to be analysed. CHAPTER 6 223 Research methodology 6.3.3 Types of questions Questions (or items) used in questionnaires are basically classified as either unstructured with an open-ended response or structured with a closed response (Cooper & Schindler, 2003: 373). 6.3.3.1 Open-ended questions As open-ended questions have no pre-coded ‘answers’ and respondents are therefore open to undue influence, they should be used with caution and only when necessary (for example in a qualitative research design) (Eiselen et al., 2005: 21). 6.3.3.2 Closed questions Closed questions in a questionnaire have response categories attached and respondents should therefore only choose an applicable answer or answers (Eiselen et al., 2005: 21). These responses can then be coded and analysed statistically. Examples of closed questions include (Struwig & Stead, 2001: 94-95): dichotomous questions with nominal scales e.g. yes/no or male/female; scaled response questions which include Likert scales and semantic differential scales; and ranking questions such as ranked order scales. For the purposes of this study a series of closed questions or items were used. 6.3.3.3 Likert scales Likert scales, named after the ‘inventor’, Renis Likert, are by far the most common type of survey item used in conjunction with questionnaires. Likert proposed a summated scale for the assessment of respondents’ attitudes in questionnaires in 1931/32 (Gliem & Gliem, 2003: 82). Individual items in Likert’s sample scale had five response alternatives: strongly approve; approve; undecided; disapprove; and strongly disapprove. Academics and scholars use a ‘rule-of-thumb’ that there must be a certain minimum number of classes CHAPTER 6 224 Research methodology or responses. In general, five would be acceptable (as Likert suggested) but Berry (1993: 47) states five or fewer is “… clearly inappropriate …”; others have insisted on 7 or more. The use of 5-point Likert scales with interval procedures is extremely common in the literature, however. a) Advantages of using Likert scales Cooper and Schindler (2003: 253-256) provide the following advantages of using the Likert scale in a questionnaire: responses are gathered in a standardised way; it is a relatively quick method of collecting information; questionnaires can be relatively easy to construct; responses can be collected from a large portion of a sample; it is relatively easy to use; and it gives participants a wide range of choices, which may make them feel more comfortable in responding to questions. b) Disadvantages of using Likert scales Cooper and Schindler (2003: 253-256) also provide some disadvantages of employing the Likert scale in a questionnaire. These are: participants may not be completely honest, intentionally or unintentionally; bias may be introduced as participants may base their responses on feelings with regard to the assessor or the subject area; participants may respond according to what they feel is expected of them as participants; the scale used in the questionnaire requires a great deal of decision-making; and c) it can take a long time to analyse and interpret the data. Criticism of the use of the Likert scale Although the Likert-scale is a useful method to determine judgements, it also has its CHAPTER 6 225 Research methodology fair share of assumptions that may be regarded as negative and should be taken into consideration when utilising it. Clason and Dormody (2004: 31-33) noted: Likert’s original work assumed an attitude scale would first be pilot-tested to assess the reliability of the individual items before it would be used. Items that would not correlate would be discarded; Likert scaling presumes the existence of an underlying (or latent) continuous variable whose value characterises the respondents’ attitudes and opinions; Likert Scales are considered a misnomer by some, as their ordinality refers only to an ordinal relationship of values within a single item. Many psychometricians would argue that they are interval scales because, when well constructed, there is equal distance between each value; Likert items on the same subject are referred to as scales even though no test of intercorrelation is employed to ensure a common meaning of items comprising a latent variable; and d) ignoring the discrete nature of the response can lead to inferential errors. Likert scale employed in this study Clason and Dormody (2004: 31) noted: “We see contemporary work using many classifications besides the traditional five-point classifications; some researchers use an even number of categories, deleting the neutral response.” Because Likert implied that the number of alternatives were open to manipulation, this study employed a manipulated Likert scale by only allowing for a 4-scale response and thereby eliminating a ‘default’ median-response. 6.3.3.4 Ambiguity and bias Some inherent ambiguity and bias is present in the research field of ethics (Rossouw, 2004: 30-31). This ambiguity stems from two sources, namely the nature of ethics and the stage of development in this field of research. Rossouw (2004: 30-31) states that this inherent ambiguity should be accounted for in the research strategy by using open-ended questions, personal interviews or focus groups. CHAPTER 6 226 Research methodology This notion was taken into consideration in the research design of this study but for practical and logistical reasons a questionnaire using closed questions was used, as unstructured, open-ended questions would have meant the use of a qualitative research design that required a different emphasis in its analysis. Even though it was stated in section 6.2.1 that data gathered within a qualitative research design may be used to generate hypotheses, this study used factor analysis for this purpose. Even though open-ended type of questions cannot adequately deal with “... complex attitudinal issues ...” (Booysen, 2003; 135), they were taken into consideration in the ultimate design of the survey instrument used in this study. 6.4 6.4.1 THE SURVEY INSTRUMENT Introduction In order to address the research problem of the study, the researcher decided on the use of a research questionnaire as a measurement instrument. Various questionnaires used for other very similar studies were investigated to determine their usefulness to this study (e.g. Diamond & Reidpath, 1992; Lee, 1997; Felicetti & Stewart, 1998; Chambliss, 2003; Balotsky & Steingard, 2005; Hoffman, 2005). Some features of these questionnaires were applicable but the majority of them were not as they used a qualitative research design with the aid of vignettes. A detailed and comprehensive questionnaire utilising a quantitative research design was therefore developed in collaboration with the Statistical Consultation Service at the UJ. 6.4.2 Questionnaire layout The questionnaire that students were asked to complete consisted of two sections. The first section (Section A) dealt with the biographical background of the respondents. Although the biographical background may not be central to this study, it did assist the researcher in analysing the data in a holistic manner. Section B of the questionnaire focused on six areas pertaining to the prospective CA. CHAPTER 6 227 Research methodology These were judgement items related to: the CA and his/her profession (Section B1 – 25 items labelled as U1 – U25); the CPD of CAs (Section B2 – 14 items labelled as V1 – V14); aspects of LLL of CAs (Section B3 – 9 items labelled as W1 – W9); aspects of ethics education related to CAs (Section B4 – 11 items labelled as X1 – X11); the core values CAs should demonstrate within their profession (Section B5 – 8 items labelled as Y1 – Y8); and the education of CAs at RAU (Section B6 – 10 items labelled as Z1 – Z10). FIGURE 6.1 indicates the various independent variables focused on in the study. The CA profession CPD and CAs LLL and CAs Ethics and CAs CPD of ethics amongst prospective CAs Core values of CAs Education of CAs FIGURE 6.1: Independent variables focused on in this study SOURCE: Own deductions One inherent limitation in the questionnaire design is the large number of items it CHAPTER 6 228 Research methodology included. As this study wished to describe something with a wide focus, the main issue was estimating the effect of every possible variable that is likely to affect the dependent variable (i.e. the CPD of ethics amongst prospective CAs) and therefore to include as many independent variables – e.g. the CA profession; CPD and CAs; LLL and CAs; Ethics and CAs; Core values of CAs; and Education of CAs (also see FIGURE 6.1). Descriptive research, as was undertaken in this study, also provides an accurate account of characteristics of a particular individual, event or group in real-life situations (Hlongwa, 2003: 33). 6.4.3 Testing of the instrument Testing the instrument (i.e. questionnaire) before it is used invariably brings to light item ambiguities and other sources of bias and error. In fact, Converse and Presser (1986: 65) argue persuasively that a minimum of two pre-tests are necessary, with pre-test sizes of 25 – 75 administered to respondents similar to those who will be in the final sample. In the case of this study, five junior lecturers in the Department of Accounting at RAU, who had just completed and passed their professional CA qualification examination, completed the questionnaire and commented on it. Experts in the field of statistics were also consulted on the probable effectiveness of the instrument. Due to the length of the questionnaire, the ‘pre-testers’ were asked to see if respondents’ attention could be maintained and if they felt that the survey had a natural flow. As RAU is a bilingual higher education institution, the translation of the questionnaire from English into Afrikaans was also an issue. After consideration and discussion with statisticians at STATKON, it was a decided not to translate it into Afrikaans due to the problem of validity. This is also the opinion of Behling and Law (2000), who suggested that the simple direct translation of validated scale items into another language often will not necessarily create a valid scale in that language. After receiving feedback and comments, the researcher then made adjustments and CHAPTER 6 229 Research methodology came up with the final questionnaire that was given to students to complete. An example of this questionnaire can be found in APPENDIX B of this study. It is important to note at this point that the role of STATKON in this study was only that of a soundboard for ideas and the entering of data in a spreadsheet format. All calculations, analysis and interpretation in respect of the data were performed solely by the writer/researcher himself (also see section 6.4.6). 6.4.4 Sampling approach The sampling approach is a process of selecting a portion of the population to represent the entire population (Cooper & Schindler, 2003: 181). The sample must therefore be valid. The non-probability sampling approach used in this research was purposive (judgement) sampling. Judgement sampling occurs when a researcher selects sample members to conform to some criterion (Cooper & Schindler, 2003: 201). For the purposes of the empirical analysis in Chapter 7, a group of third-year students at a residential university were used as respondents with the following criteria: they all had to have the prescribed four major subjects (i.e. Accounting, Auditing, Financial Management and Taxation) at a third-year level; no students repeating any of the subjects were considered; and they had to have the intention of continuing with an honours degree in accounting the following year. 6.4.5 Administration As this questionnaire was aimed at final-year accounting students at RAU, all the third-year B Com Accounting students were visited at a combined lecture on the first day of the second semester of their third year. This meant that a large percentage of the population of accounting students were then targeted as the sample for this study. Before commencing with the completion of the questionnaire, the aim of the study and its various administrative aspects were explained to the students. Students were then CHAPTER 6 230 Research methodology given 30 minutes to complete this questionnaire under controlled conditions. After the students completed the questionnaire, it was collected by the writer/researcher and three student assistants. 6.4.6 Data analysis The completed questionnaires were then handed to a data capturer for coding and capturing. The data capturer was fully briefed on all categories of data as well as on the pre-determined code rules to be used. After coding and capturing on a spreadsheet, the data was forwarded to the writer/researcher by STATKON for analysis and interpretation using the SPSS13® software package. 6.5 ETHICAL CONSIDERATIONS In order to comply with the ethical constraints underlying the undertaking of a research project such as this, attention was given to the following aspects: 6.5.1 Permission to collect data Permission to conduct the survey amongst the students was sought and given by the Head of the Department of Accounting, RAU. 6.5.2 Informed consent Prior to the handing out of questionnaires (see section 6.4.5), consent was asked from each participant. Although no written consent was sought, participants were informed that if they did not wish to participate, they were free to hand back their uncompleted questionnaires. 6.5.3 Confidentiality and anonymity To ensure confidentiality, respondents were reassured verbally and in writing (on the cover of the questionnaire) that the information would be treated with the utmost confidence. Although various research reports would be published, it would contain figures, percentages and deductions based on the analysis and interpretation of the CHAPTER 6 231 Research methodology data provided without identifying any respondent personally. An example of the questionnaire can be found in APPENDIX B of this study. 6.6 ANALYSIS OF DATA 6.6.1 Descriptive analysis of data Within each of the areas in Section B of the questionnaire, described in section 6.4.2, students had to respond to items by indicating to which extent they agreed or disagreed with the items. Each item was judged using a 4-point Likert scale consisting of: 1 = Strongly agree 2 = Agree 3 = Disagree 4 = Strongly disagree The analysis pertaining to each area in Section B of the questionnaire will therefore always commence with a descriptive analysis of the data. This will include providing basic information about the data describing the: number of valid and missing responses; range of responses (i.e. range, minimum and maximum); central tendency of responses (e.g. mean values); standard deviation and variance of responses; skewness of responses with the standard error of skewness; and kurtosis of responses with the standard error of kurtosis. 6.6.1.1 Standard deviation A standard deviation is a statistical measure which expresses the average deviation (or dispersion) about the mean (Wegner, 1993: 92). CHAPTER 6 232 Research methodology 6.6.1.2 Skewness Skewness provides information on the symmetry of the distribution (Pallant, 2005: 51). In a normal symmetrical distribution, one would find that the degree of skewness tends to be equal to zero where the mean score is equal to the median score. A positively skewed distribution indicates that scores are clustered to the left of the graph at low values and in general the mean score is greater than the median score. Negative skewness indicates a clustering of scores at the high-end (or right hand side) of the graph and in general the mean score is less than the median score (Pallant, 2005: 52). 6.6.1.3 Kurtosis Kurtosis provides information on the ‘peakedness’ of the distribution. Positive kurtosis indicates that the distribution is peaked (i.e. scores clustered around the centre) or ‘leptokurtic’ (i.e. has long, thin tails). Kurtosis values below 0 indicate a distribution that is relatively flat (i.e. too many cases in the extremes or ‘platykurtic’). In a normal symmetrical distribution, one would find that kurtosis tends to be equal to zero (Pallant, 2005: 51). Kurtosis may sometimes be the result of an underestimation of the variance but this risk is reduced with a sample of 200 or more cases (Pallant, 2005: 52). Even though the results of this analysis done on the data used in this study showed the presence of negative kurtosis, one can ignore it, however, as the sample was greater than 200 (n = 491). (Also see section 6.6.2.2.) As was pointed out in section 6.2.2, this was a descriptive study, which meant a large number of items had to be included. In order to give statistical meaning to the analysis, factor analysis was employed to determine the underlying dimensions of the various attitudes emerging in the data (Eiselen et al., 2005: 104). CHAPTER 6 233 Research methodology 6.6.2 Factor analysis 6.6.2.1 Introduction Exploratory factor analysis and confirmatory factor analysis are two statistical approaches used to examine the internal reliability of a measure. Both are used to investigate the theoretical constructs, or factors, that might be represented by a set of items. Other characteristics of the two methods are: either can assume the factors are uncorrelated (i.e. orthogonal); both are used to assess the quality of individual items; and both can be used for exploratory or confirmatory purposes (Pallant, 2005: 96). 6.6.2.2 Aim of factor analysis The most common form of factor analysis, namely exploratory factor analysis, is used to uncover the latent structure (or dimensions) of a set of variables or opinion-related questions (items) (Eiselen et al., 2005: 104-105) and allows one to condense a large number of variables or scale items into a smaller, more manageable number of dimensions or factors (Pallant, 2005: 96). In exploratory factor analysis the researcher’s à priori assumption is that any indicator may be associated with any factor. Using exploratory factor analysis (as was the case in this empirical study) reduces attribute space from a larger number of variables to a smaller number of factors and as such is a ‘non-dependent’ procedure, that is, it does not assume that a dependent variable is specified (Pallant, 2005: 96). The mathematical detail of factor analysis is beyond the scope of this study but there will be instances where some mathematical detail will be provided for the sake of completeness. CHAPTER 6 234 Research methodology 6.6.2.3 a) Practical considerations when using exploratory factor analysis Sample size For factor analysis to provide valid results, the number of respondents should be at least four times the number of variables (Eiselen et al., 2005: 105). Sometimes statisticians are of the opinion that the ratio of respondents to the number of items should be 10:1 (Pallant, 2005: 174). If this ‘ratio test’ is used in this analysis, factor analysis can be employed because this would imply a total of 250 respondents. The number of respondents in this analysis was 491. b) Factor extraction method Factor extraction involves determining the smallest number of factors that can be used to best represent the interrelations among the set of variables (Pallant, 2005: 174). If all variables in factor analysis are continuous and assumed to be normally distributed, the maximum likelihood factoring method may be employed (Eiselen et al., 2005: 105). As these two assumptions may not be valid for this analysis, the extraction method of Principal Axis Factoring will be used in the analysis. c) Method of rotation Rotation is merely used to assist with the interpretation or identification of variables making up or contributing most to a factor. Two types of rotation can be identified, namely orthogonal and oblique rotation (Eiselen et al., 2005: 105-106). For the purposes of this and further analyses in this study, the Varimax method of orthogonal rotation with Kaizer Normalisation (or the Eigenvalue rule) will be used, as orthogonal rotations yield factors uncorrelated with one another and authors also believe that oblique rotations provide unstable results (Pallant, 2005: 175; Eiselen et al., 2005: 106). Orthogonal rotation also results in solutions that are easier to interpret; they do however require an assumption that the underlying constructs are independent or not CHAPTER 6 235 Research methodology correlated (Pallant, 2005: 176). d) Reliability of factor analysis The reliability of the application of factor analysis will be determined by: the Kaiser-Meyer-Olkin Measure of Sampling Adequacy; Bartlett’s Test of Sphericity; and Initial Eigenvalues. (i) Kaiser-Meyer-Olkin Measure of Sampling Adequacy Although some statisticians feel that the Kaiser-Meyer-Olkin Measure of Sampling Adequacy (hereafter KMO) value should be greater than 0.5 or 0.6 (Eiselen et al., 2005: 107), for the purposes of this analysis the statistical impact was placed at 0.7. Any value greater than 0.7 will therefore mean that a factor analysis will be warranted as there is enough correlation between an item and the rest of the other items (Eiselen et al., 2005: 107/108) (ii) Bartlett’s Test of Sphericity Bartlett’s Test of Sphericity (hereafter Barlett’s Test) tests the hypothesis that the correlation matrix is an identity matrix and all items will therefore be uncorrelated (Eiselen et al., 2005: 107). Small p-values (i.e. significance or Sig. less than 0.05) indicate that a factor analysis may be useful. Two other diagnostic measures that are used to decide whether a factor analysis is warranted is the Anti-image Correlations and Initial Eigenvalues (Eiselen et al., 2005: 107-108). (iii) Initial Eigenvalues The Initial Eigenvalues (also known as characteristic roots) are used to explain the total variance. When one performs a factor analysis, the ideal situation is to use as few factors as possible (i.e. to reduce the data) but to still allow these factors to contain as much of the information provided as possible. CHAPTER 6 236 Research methodology The ‘ideal’ number of factors to be used can be determined using the Eigenvalue criterion. This criterion states that the number of factors to be used is equal to the number of factors with Eigenvalues greater than 1 (Eiselen et al., 2005: 108). 6.6.3 Item analysis Factor analysis does not indicate whether each individual item is highly correlated with the scale as a whole. To determine the extent to which the items in this study are collectively correlated with the scale an item analysis will be performed. In an item analysis one is interested in how well the responses of each item in a factor (scale of items) correspond to the other items and the scale as a whole (Eiselen et al., 2005: 112). 6.6.4 Factor scores Following the item analysis, factor scores will be calculated and interpreted. In order to calculate the factor scores, i.e. the average score for each respondent for each of the factors, the average (or mean) response for each respondent will be calculated (Eiselen et al., 2005: 116). The question now arises as to whether the means calculated (i.e. the factor scores) for each factor differs on a gender basis. This will be tested by developing various hypotheses and then exploring the differences between the gender groups. 6.6.5 Development of hypotheses In order to explore the differences between the gender groups, various hypotheses will be developed. The null-hypothesis will state that the means between the gender groups are equal e.g. H0: µM = µF whereas the alternative hypothesis will state that the means between the gender groups are not the same e.g. H1: µM ≠ µF. 6.6.6 6.6.6.1 Testing of hypotheses Introduction When one explores the differences between gender groups, for example, one is trying CHAPTER 6 237 Research methodology to see whether there is a statistically significant difference between male students and female students (Pallant, 2005: 96/97). 6.6.6.2 t-tests for independent samples An appropriate test that can be used is the t-test for independent samples between two groups based on gender for example (Collis & Hussey, 2003: 244; Eiselen et al., 2005: 76; Pallant, 2005: 196). Within the t-test for independent samples, any p > 0.05 will mean that the null hypothesis will not be rejected, as that there is no statistically significant difference between the gender groups. Other statistical alternatives do exist to measure differences in interval data items (as was used in the questionnaire). These include the Mann-Whitney U test and the Pearson Chi-square test (Cooper & Schindler, 2003: 596). 6.6.6.3 Mann-Whitney U test The Mann-Whitney U Test is used to test differences between two independent groups on a continuous measure (Pallant, 2005: 291) by comparing the means of two groups i.e. male and female students. 6.6.6.4 Pearson Chi-square test Another non-parametric test that may be used to explore the interrelationship amongst variables is the Pearson Chi-square ( χ 2 ) test for independence (Pallant, 2005: 104, 287). 6.7 SUMMARY This chapter provided a logical basis and framework which will be used in the empirical analysis to be completed in Chapter 7. ---o0o--- CHAPTER 6 238