This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
In previous discussion section, literatures concerning in-group identification, trust, job satisfaction, commitment has been introduced. In the following, the research model will be developed and then the key constructs within the model will be analyzed after which the research hypotheses will be deduced.
Before going on to discuss about the hypotheses and methodology adopted, the following is going to give a brief introduction about the meaning of the variables that refers to in this research.
Length of Employment within the organization: The duration that the professionals have been worked in the organization
Propensity to Trust: The professionals' general willingness to trust others working in the same organziation
In-group Identification: The professionals' sense of connectedness with a specific collection of people working in the same organization
Job Satisfaction: The overall feeling that professionals have about their jobs
Commitment: The force that binds the professionals to achieve the target for their working organization
From the previous research model, it is noted that three identified micro factors will act as variables or basic components in the hypothetical models. The three identified micro factors will be used as latent variables or factors in hypothesis 1 to 3. Their corresponding personal constructs will be used as observable variables or items to measure them. Because there are 3 micro factors influencing job satisfaction, the first hypothesis is to test whether there is a significant relationship between the 3 micro factors and job satisfaction. Furthermore, from the empirical findings from literatures, it is noted that the 3 micro factors may also affect professionals' commitment to their working organization. Therefore, the second hypothesis is to test whether there is a significant relationship between the 3 micro factors and professionals' commitment to their working organization. Moreover, it is found that job satisfaction may also be a mediating variable between the 3 micro factors and professional's commitment to their working organization. So, the research model will be investigated through the following hypotheses:
- Hypothesis 1: The 3 micro factors affect professional's job satisfaction in an organization.
- Hypothesis 1.1: Professional's length of employment within an organization affects his job satisfaction in an organization
- Hypothesis 1.2: Professional's propensity to trust other organizational members affects his job satisfaction in an organization
- Hypothesis 1.3: Professional's in-group identification affects his job satisfaction in an organization
- Hypothesis 2: The 3 micro factors affect professional's commitment to their working organization.
- Hypothesis 2.1: Professional's length of employment within an organization affects his commitment to the organization
- Hypothesis 2.2: Professional's propensity to trust other organizational members affects his commitment to the organization
- Hypothesis 2.3: Professional's in-group identification affects his commitment to the organization
- Hypothesis 3: Job satisfaction is a mediating variable between the 3 micro factors and commitment to an organization.
- Hypothesis 3.1: Job satisfaction is a mediating variable between professional's length of employment within an organization and his commitment to the organization
- Hypothesis 3.2: Job satisfaction is a mediating variable between professional's propensity to trust other organizational members and his commitment to the organization
- Hypothesis 3.3: Job satisfaction is a mediating variable between professional's in-group identification and his commitment to the organization
Although some researchers argue that both the qualitative and quantitative approaches may be indiscernible irrespective of whatever research is carried out, such as "no particular method has a clear superiority over the others" (McCall and Bobko, 1990), many others (e.g. Bryman, 1999) have insisted that the research issues should determine the selection of the research approach. The suitability of a specific approach is a matter of discretion relating to the objective, setting and a variety of other questions (Hart, 1994) or even the availability of resources (Fellows and Liu, 2002).
Warwick and Lininger (1975) propose that there are three conditions which make quantitative approach an appropriate and useful means of gathering information:
- Whether the goals of the research call for quantitative data;
- When the information sought is reasonably specific and familiar to the respondent; and
- When the researcher himself has considerable prior knowledge of particular problems and the range of responses likely to emerge.
Bryman (1999) states that when a research concerns a large scale issue e.g. to test the validity of a prior theory, "to explore or test certain theoretical relationships" and when the research focuses on establishing cause-and-effect relationships, the quantitative approach will be the appropriate one.
In this research, the aim is to understand the nature of commitment amongst construction professionals to their working organization. In order to achieve this aim, research model is built up based upon the existing issues identified from psychology and management. Through building the research model, this research will attempt to construe and interpret the behavioural phenomena and the causalities among the variables e.g. in-group identification, propensity to trust, length of employment within the organization, job satisfaction and commitment etc. amongst construction professionals to their working organization.
After model development, hypotheses are derived from the model so as to get the research model tested empirically. The hypotheses and sub-hypotheses all concern the cause-effect relationships between construction professionals and their commitment to their working organization, together the mediating influence of job satisfaction on these relationships. In addition, the latter part of this research is to seek empirical support for the validity of the cause-effect relationships within the model and hypotheses in one specific setting - Hong Kong construction industry. What the researcher purses is to obtain empirically quantitative evidence to test the hypotheses among construction professionals. Therefore, the highlights stressing the "hypothesis-testing" (Drenth et al., 1998) within a relatively broad research population justify the adoption of the quantitative approach.
Design of the research
To test the hypothesis deduced from the previous page, the methodology adopted is to investigate how micro factors will affect commitment amongst construction professionals to their working organization and the influence of job satisfaction as a mediator on this issue. This research consists of two stages, of which Stage 1 is the pilot study and Stage 2 is the main survey. The measurement instruments for the main variables will be discussed later.
Stage 1 is a pilot study, through which all the questions towards the main variables are going to be asked to a small sample number of participants. The aim of this stage is: (1) to ensure the questions are well-defined and understood by the participants; (2) to ensure the instruments developed by the western psychologists can be applied into the Hong Kong construction context (since the questions are mostly asked in general, it may not be well-suited to the construction industry context) and (3) to test the reliability of the instruments within a small scale.
Following this, the main questionnaire survey will be carried out in Stage 2. The aim of this stage is mainly to test the hypotheses and sub-hypotheses developed. Certain analysis techniques will be carried out to test the hypotheses based on the data obtained from the survey participants. In the following, these techniques will be briefly described.
Method of Analysis
As mentioned previously, the quantitative approach is adopted due to the characteristics of the research objectives which are to test the cause-effect relationships between micro factors and commitment amongst construction professionals to their working organization.
From various literatures (e.g. Hair et al., 1998), it is noted that a myriad of sophisticated techniques have been developed to facilitate quantitative research in social science, such as Factor Analysis, Conjoint Analysis, Analytical Hierarchy Process (AHP), Multiple Regression Analysis, Structural Equation Modeling (SEM) etc. Different analytical techniques are developed by mathematicians for different social science research settings. The simple and multiple regression techniques have been used extensively in social science as data collection tools for the purpose of prediction and variance explanation.
Netemeyer, Bearden and Sharma (2003) highlighted the importance of internal consistency as well as the effects of inter-item correlation and wording redundancy on reliability when carrying out a research. Therefore, certain analysis will be carried out to check on these issues before adopting the analytical technique to test the cause-effect relationships.
Factor analysis is based on the assumption that some underlying factors, which are fewer in number than the number of variables, are responsible for the covariation among the observed variables (Kim and Muller, 1978). It is concerned with exploring the pattern of relationships among a number of variables (Hair et al. 1998). In regression analysis, the analysis of variance (ANOVA) is based on the analysis of variance in dataset. There are two different kinds of factor analysis, namely exploratory and confirmatory. The exploratory factor analysis attempts to reduce a set of variables into fewer underlying factors. The confirmatory factor analysis hypothesizes that there are fewer underlying factors for a set of variables and then seeks to determine whether the hypothesis does hold. At this point, since it is going to test whether or not the existing data are consistent with a constrained priori structure that meets the conditions of model identification and to ensure the validity of the measurement instrument, therefore, only confirmatory factor analysis (CFA) will be adopted.
CFA refers to a number of multivariate statistical techniques used to analyze the interrelationships among a large number of variables. Its purpose is to identify the underlying dimensions or factors that accounts for the relationships that are observed among the variables. The result of the analysis shows a set of interrelated variable relationships. According to Long (1983), standard CFA models have the characteristics listed below:
- each indicator is a continuous variable represented as having two causes - a single underlying factor that the indicator is supposed to measure all other unique source of causation that are represented by the error term
- the measurement errors are independent of each other and of the factors
- all associations between the factors are analyzed
The purpose of carrying out CFA are mainly to remove collinearity, to reduce the number of original variables and to identify dimension among the set of observations on the basis of a set of variable measures (Reisinger, 2003). In the confirmatory factor model, researcher can specify (1) which pairs of common factors are correlated; (2) which observed variables are affected by which common factors; (3) which observed variables are affected by a unique factor and (4) which pairs of unique factors are correlated (Long, 1983). Therefore, CFA is particularly useful in the validation of scales for the measurement of specific constructs (e.g. Steenkamp, 1991). Validity is the extent to which the factors "accurately" measure what they are supposed to measure (Hair et al. 1998). CFA is also going to be adopted in this study in order to confirm the validity of the existing scale of the questionnaire survey.
Beyond examination of the loading for each factor, a common measure used in assessing the measurement model is about the reliability of each construct. Reliability is a measure of internal consistency of the construct factors, depicting the degree to which they 'indicate' the latent construct (Hair et al. 1998). It is the ratio of the variance of the true scores to the total scores (Netemeyer, Bearden and Sharma, 2003). A common used threshold value for acceptable reliability is 0.70, which often termed as Cronbach's alpha. According to Streiner (2003), several key points about Cronbach's alpha should be noted in doing the reliability analysis:
- Alpha is not a fixed property of a scale. It relies upon as much on the sample being tested as on the test.
- Alpha measures only the internal consistency of the scale. A high value of alpha is a prerequisite for internal consistency but does not guarantee it.
- The bigger alpha may not mean better reliability. Higher value may only reflect unnecessary duplications across items and redundancy.
In addition, Cortina (1992) also mentioned that the length of a scale influences the value of alpha dramatically. So, when the scale test is conducted, the number of items included should be carefully considered. In this research, the Cronbach's alpha will be employed to assess the reliability of the scale test.
Structural Equation Modeling (SEM)
Structural equation modeling is a multivariate technique that combines (confirmatory) factor analysis modeling from psychromatic theory and structural equations modeling associates with econometics (Hair et al. 1998). During past decades, Structural Equation Modeling (SEM) has been developed as a powerful multivariate data analysis technique in social science, especially in the fields of sociology, psychology and education (Mueller, 1996). It is a logical coupling of regression and factor analysis approaches (Maruyama, 1998). It attempts to deal with the impossibility to measure the abstract concept / constructs in empirical research. SEM provides a statistical bridge between theoretical and empirical aspects in the study of human being behaviours and the measurement of the characteristics of man (Hair et al. 1998). In addition, SEM is also seen as one process from theoretical relationships to the mathematical / statistical relationships among the variables. It enables the researchers to overcome the drawbacks in other analysis (such as the measurement and specification errors) and quantify the effects among latent constructs. There are several reasons to use SEM in social science research (Reisinger, 2003 and Ding, 2007):
- Measurements are sometimes imperfect in social science: researchers need to use scales to measure some abstract constructs in order to minimize error
- Abstract or latent constructs are more interesting than observed variables;
- Sometimes relationships between latent constructs need to be estimated;
- SEM is a general framework for many other statistical techniques, e.g. ANOVA, MANCOVA, regression, path analysis and factor analysis
- Unreliability of indicators can be taken into consideration
SEM hypothesizes a set of relationships among variables, both observed and latent (Hair et al., 1998). In this research, the length of employment within an organization is a latent variable and its relationships with job satisfaction and commitment need to be investigated. In addition, it can also estimate multiple and interrelated relationships among dependence variables (job satisfaction and commitment). Since the relationship between job satisfaction and commitment needs to be investigated as to show the mediating effect of job satisfaction. Hence, SEM is an ideal choice for model analysis.
In the following, the seven stages in the process of structural equation modeling will be briefly discussed.
Development of a theoretical model
Structural equation modeling is based on casual relationships, in which the change in one variable is assumed to result in a change in another variable (Heise, 1975). The first part of Stage 1 will focus on the development of a theoretical model with the linkages (defined relationships) between the latent constructs and their measurable variables. This part represents the development of a structural model. Taking the relationship between job satisfaction and commitment as an example, the structural model here refers to "it is hypothesized that there is a strong relationship (correlation) between job satisfaction and commitment".
The second part of Stage 1 involves the operationalization of the latent constructs via the measured variables and describing the way in which they are represented by empirical indicators (Hair, et al., 1998). This refers to the development of a measurement model. Hypotheses and sub-hypotheses developed are referred here to the measurement model.
Constructing a Path Diagram of casual relationships
Stage 2 involves the construction of a path diagram. In the path diagram, all relationships between constructs and their indicators are graphically presented with arrows. This form a visual presentation of the hypotheses developed. According to Hair et al. (1998), a path diagram is more than a visual portrayal of the relationships because it allows the researcher to present not only the predictive relationships among constructs (i.e. the dependent-independent variable relationships), but also associative relationships (correlations) among constructs. Two assumptions underlie path diagrams (Hair et al., 1998). First, all casual relationships are indicated. This also justifies why a casual relationship does not exist between two constructs as it is to justify the existence of another relationships. The second assumption relates to the nature of casual relationships that are assumed to be linear. The estimates of path coefficients and variances are obtained from the path analysis.
Path analysis model describes "the hypothetical casual effect, and assigns the explained and unexplained variances" (Hair et al., 1998). The overall goal of the path analysis is to estimate casual versus noncasual aspects of observed correlations. The originally postulated model can be expressed simultaneously multivariate linear equations via technique of path analysis. Path coefficients are interpreted as regression coefficients in multiple regression, which means that they control for correlations among multiple presumed causes (Reisinger, 2003). In this study, in order to test the hypotheses, path analysis will be adopted as one of the main technique. From this analysis, it is proposed that the casual correlation between job satisfaction and micro factors as well as commitment correlation with job satisfaction and micro factors.
Converting the Path Diagram into a Set of Structural and Measurement Models
This stage involves the formal mathematical specification of the model by describing the nature and number of parameters to be estimated (which variable measure which constructs), translating the path diagram into a series of linear equations that link constructs, defining the measurable model specifying which variable measure which constructs and indicating hypothesized correlations among constructs or variables.
Translating a path diagram into a series of structural equations is a straightforward procedure. First, each endogenous construct is the dependent variable in a separate equation. Then the predictor variables are all constructs at the ends, or "tails" of the straight arrows leading into the endogenous variable. That is, for example,
Y1 = b1X1 + b2X2 + E1 (bi is the structural coefficient, Ei is the prediction error)
Regarding the issues about defining the measurable model to specify which variable measure which constructs, it relates to the issues regarding the number of indication per constructs and the process of specifying the reliability of the constructs (Hair et al., 1998). The measurement that requires here refers to the confirmatory factor analysis and reliability analysis that specified previously.
In addition to the structural and measurement models, the research must be specified any correlations between the exogenous constructs or between the endogenous constructs. In this research, each exogenous constructs (micro variables) are designed to be correlated with the endogenous constructs.
Choosing the Input Matrix Type and Estimating the Proposed Model
In this stage, the researcher will address the actual process of estimating the specified model, including the issues of inputting the data in the appropriate form and selecting the estimation procedure.
With reference to the inputting data issues, SEM differs from other multivariate techniques that it uses only the variance-covariance or correlation matrix as the input data (Hair et al. 1998). It is noted that the correlation matrix is more widely used because it allows for direct comparisons of the coefficients within a model. This matrix will also be used as the inputting data in this research because this research is to understand the pattern of relationships between the constructs and variables, but not to explain the total variance of the constructs or test a theory. From a practical perspective, correlations are also more easily interpreted and diagnosis of the results is more direct.
In addition, an assessment of the sample size also needs to be done at this stage because it will affect the estimation and interpretation of the result. Although there is no correct rule for estimating sample size for SEM, recommendations are for a size ranging between 100 and 200 (e.g. Hair et al., 1998). According to Hair et al. (1998), the sample size has to be large enough when compared with the number of estimated parameter, but with an absolute minimum number of 50 respondents.
When selecting the estimated procedure, it is decided that the multivariate normality assumption test needs to be carried out before the decision is made. If this assumption is violated, either a transformation of the original dataset or a selection of appropriate estimation will be required. In the former case, once the original dataset is transformed so that the multivariate normality assumption is satisfied, the final result will be difficult to interpret due to the change of the original scale. For the latter, some estimation methods can be used as the fitting functions such as the unweighted least squares (ULS), generalized least squares (GLS) and maximum likelihood (ML). In this research, it is suggested that the maximum likelihood (ML) will be selected as the estimation function because it is robust to non-normal data (Wang, Fan and Wilson, 1996) and can provide valid results with small sample size.
Maximum Likelihood (ML)
According to Winer et al. (1991), the term maximum likelihood describes the statistical principle that underlies the derivation of parameter estimates: the estimates are the ones that maximize the likelihood (the continuous generalization) that the data are drawn from this population. That is, maximum likelihood estimators are those that maximize the likelihood of a sample that is actually observed (Hair et al. 1998). It is a normal theory method because ML estimation assumes that the population distribution for the endogenous variables is multivariate normal.
The statistical assumptions of ML estimation include independence of the observations, multivariate normality of the endogenous variables, independence of the exogenous variables and disturbances and correct specification of the model (Hair et al. 1998). In standard ML estimation, standard errors are calculated only for the unstandardized solution. This means that the results of the statistical tests (i.e. ratios of parameter estimates over their standard errors) are available only for the unstandardized solution. Because other statistical methods usually require additional assumptions about the population distribution of either observed endogenous variables (i.e. multivariate normal) or the disturbances (i.e. normal), therefore ML estimation will be selected as the estimation model.
Assessing the Identification of the Structural Model
This stage addresses the issue of model identification, that is, the extent to which the information provided by the data is sufficient to enable parameter estimation (Hair et al. 1998). If a model is not identified, then it is not possible to determine the model parameters. For the purpose of identification, the researcher has to concern with the size of the correlation matrix relative to the number of estimated coefficients. The difference between the number of correlations and the actual number of coefficients in the proposed model is termed "degree of freedom". A necessary condition for identification is that the number of independent parameters is less than or equal to the number of elements of the sample matrix of correlation among the observed variables.
Evaluating Goodness-of-Fit Criteria
The first step in evaluating the results is an initiation inspection for "offending estimates", which are coefficients that exceed acceptable limits. According to Hair et al (1998), the common examples are:
- Negative error variances or nonsignificant error variances for any construct
- Standardized coefficients exceeding or very close to 1.0
- Very large standard errors associated with any estimated coefficients
Once the model is established as providing acceptable estimates, the goodness-of-fit will then be assessed for the measurement and structural models separately. This assessment is to ensure that the model can provide an adequate representation of the entire set of casual relationships. From the statistical literature, it is found that there are three types of model fit measurement: absolute fit measures, incremental fit measures and parsimonious fit measures. Under each type of measure, there are several kinds of measures. In this research, the following few measures will be used to assess the fitness of the model as they are considered as important index of fit in the literatures that should be reported.
Absolute Fit Measures
The absolute fit measures provide information on the extent to which the model as a whole provides an acceptable fit to the data (Hair et al. 1998). Among the absolute fit measures used to evaluate the SEM, the Likelihood ratio of Chi-square Statistic, Root Mean Square Error of Approximation (RMSEA) and Expected Cross Validation Index (EVCI) is decided to adopt for this study.
Incremental Fit Measures
While all the absolute measures might fall within acceptable levels, the incremental fit index is needed to ensure acceptability of the model from other perspectives (Hair et al. 1998). The incremental fit measures assess the incremental fit of the model compared to a null model (i.e. the most simple model that can be theoretically justified). In this research, it is decided that the Normal Fit Index (NFI) and Comparative Fit Index (CFI) will be adopted for measurement.
Interpreting and Modifying the Model
The final stage involves interpreting the results in both empirical and practical terms, as well as examining the results for any potential model modifications (Hair et al. 1998). At this stage, the interpretation is the basis for support of the theoretical model and results from previous analysis are used to develop a better fitting model. This model respecification procedure is necessary to ensure that the new estimated model fits the data better. Together with this, any specification error can also be identified with goodness-of-fit maximized.
Sampling involves selecting units of analysis (e.g. people, groups etc.) in a manner that maximizes the researcher's ability to answer research questions that are set forth in a study (Tashakkori and Teddie, 2003). In this research, unit of analysis is on individual basis. In addition, the targeted population, sampling frame and sample size have to be sort out before the questionnaire survey starts.
In this research, the targeted research population is the professionals working in the Hong Kong construction industry. A "professional" is considered as "someone who can act independently while bringing a body of special knowledge to bear in a work situation" (Shapero, 1985). It is recognized that professionals are highly qualified and are engaged primarily in work of intellectual nature (Alvesson, 1995) and that professionals have a specific area of specialization (Maister, 1993).Therefore, the targeted population includes architects, engineers and surveyors working in Hong Kong construction industry. The total number of targeted population will be complied from various current professional membership directories such as Hong Kong Institute of Surveyors, Hong Kong Institute of Architects and Hong Kong Institute of Engineers.
The sampling frame is decided to be the surveyor consultant companies in Hong Kong. From the HKIS website, it is found that there are around 150 surveyor consultant companies in Hong Kong. This decision is based on two considerations. Firstly, collecting data from all consultant companies in Hong Kong construction industry is not possible as time and resources are limited. Secondly, the programme offered by the Department of Real Estate and Construction is associated with the surveying profession.
Based on the sampling frame, the sample population will be the surveyors working in the consultant companies in Hong Kong. From the HKIS website, it is found that there are around 4787 corporate members (as at 23 July 2009) and 78% worked in the private and public companies. This implies that there are around 3500 corporate members worked in public and private companies. However, there is no information about how many percentages of them are working in the surveyor consultant companies. Therefore, the consultant companies set in the sampling frame will be used for obtaining the sampling unit for data collection. With reference to Kline (2005), the structural equation modeling analysis method usually requires a sample size of around 200. In addition, recognizing from previous PhD Thesis and research studies (e.g. Phua, 2002), it is found that the average response rate for mailed questionnaires surveys in Hong Kong is usually between 10-13%. The total number of questionnaire sent out is decided to be more than 2000. Therefore, the average number of surveyors for each consultant companies would need to be around 15. It is noted that some small-sized consultant companies may not have enough number of surveyors, therefore the questionnaire sent for different consultant company need to be calculated and adjusted. (Details will be discussed later)
- Alvesson, M. (1995), Management of Knowledge-Intensive Companies, Berlin: de Gruyter
- Bryman, A. and Burgess, R. (1999), Qualitative Research, London: Sage Publication
- Cortina, J.M. (1993), "What is coefficient alpha? An examination of theory and applications", Journal of Applied Psychology, vol. 78, pp. 98-104
- Ding, Z.K. (2007), "Interpersonal Trust and Willingness to Share Knowledge among Architects: A Two-Stage Triangulation Approach", Department of Real Estate and Construction, The University of Hong Kong
- Drenth, .J.D., Thierry, H. and Charles, J.D.W. (1998), Handbook of work and organizational psychology, East Sussex: Psychology Press Ltd.
- Fellows, R. and Liu, A.M.M. (2002), Research methods for construction (2nd Edition), Oxford: Blackwell Science
- Hair, J.F. Jr., Anderson, R.E. and Tatham, R.L. (1998), Multivariate data analysis, Upper Saddle River: Prentice Hall
- Heise, D.R. (1975), Casual Analysis, New York: Wiley
- Kim, J. and Mueller, C.W. (1978), Introduction to factor analysis: what it is and how to do it, Newbury Park: Sage Publications
- Kline, R.B. (2005), Principles and Practices of Structural Equation Modeling (2nd Edition), New York: Guilford Press
- Long, J.S. (1983), Confirmatory Factor Analysis, Beverly Hills: Sage Publications
- Maister, D.H. (1993), Management the Professional Service Firm, New York: Simon and Schuster
- Maruyama, G.M. (1998), Basics of Structural Equation Modeling, Sage: Thousand Oaks
- McCall, M.W. Jr. and Bobko, P. (1990), "Research methods in the service of discovery", in Dunnette, M.D. and Hough, L.M. (eds.), Handbook of industrial and organizational psychology, Palo Alto, CA: Consulting Psychology Press
- Mueller, R.O. (1996), Basic principles of structural Equational modeling: An introduction to LISREL and EQS, New York: Springer
- Netemeyer, R.G., Bearden, W.O. and Sharma, S. (2003), Scaling Procedures: issues and applications, Thousand Oaks: Sage Publications
- Phua, F.T.T. (2002), Toward a critical assessment of social identity: the nature of organizational identification and its implications for inter-organizational cooperation in the context of the Hong Kong construction industry, Unpublished PhD Thesis, Department of Real Estate and Construction, The University of Hong Kong
- Reisinger, Y. and Turner, L. (2003), Cross-cultural behaviour in tourism: concepts and analysis, Oxford: Butterworth
- Shapero, A. (1985), Managing Professional People: Understanding Creative Performance, New York: Free Press
- Steenkamp, J.E.M. and van Triip, H.C.M. (1991), "The Use of LISREL in Validating Marketing Constructs", International Journal of Research in Marketing, Vol. 8, No. 4, pp. 283-299
- treiner, D.L. (2003), "Starting at the beginning: an introduction to the coefficient alpha and internal consistency", Journal of Personality Assessment, Vol. 80, pp. 99-103
- Tashakkori, A. and Teddlie, C. (eds.) (2003), Handbook of mixed methods in social and behavioural research, Thousand Oaks: Sage Publications
- Wang, L.L., Fan, X. and Wilson, V.L. (1996), "Effects of Nonnormal Data on Parameter Estimates for a Model with Latent and Manifest Variables: An Empirical Study", Structural Equation Modeling, Vol. 3, No. 3, pp. 228-247
- Warwick, D.P. and Liniger, C.A. (1975), The sample survey: theory and practice, New York: McGraw-Hill