The Concept of Trust

Get Complete Project Material File(s) Now! »

Confirmatory factor analysis

A confirmatory factor analysis (CFA) is a specific application of SEM that is linked to the measurement model (Brown, 2006, p. 1; Hoyle, 2000, p. 466). EFA is more inductive and concerns the discovery of new links or relationships, while CFA is deductive and driven by hypothesis to investigate known theories and even which items measure which latent factor (Brown, 2006, p. 1; Hoyle, 2000, p. 456). A confirmatory factor analysis is defined by the Dictionary of Statistics & Methodology as a “[f]actor analysis conducted to test hypotheses (or confirm theories) about the factors one expects to find. It is a type of or element of structural equation modeling” (Vogt, 2005, p. 57). In the case of SEM, the researcher uses both a measurement model and a structural model. The CFA represents this measurement model, while the “structural model concerns the directional relations between constructs” (Hoyle, 2000, pp. 465-466). CFA is also used to assess the quality of a measurement model in the case of a replication (Mayer & Gavin, 2005, p. 879).

How many factors?

When considering the number of factors to select, it is always difficult to decide what is just enough. Too few or too many both have associated problems. Hayton et al. (2004) summarise the problems associated with specifying too few factors as resulting in low factor loadings and in misinterpretation. A factor may be combined with another or even ignored, which leads to measured variables that actually load on factors not included in the model, falsely loading on the factors that are included, and distorted loadings for measured variables that do load on included factors. Hayton et al. (2004, p. 192) Although less critical, specifying too many factors can lead to difficult or incorrect interpretations by being distracted to focus on the wrong factor or a factor that is nearly impossible to replicate (Hayton et al., 2004).

Handling non-normal data in SEM

From the above analysis it is evident that the data is not normally distributed, and in a later section the effect of this on the final model will be investigated by using a bootstrapping simulation inside the SEM. A further method of determining the effect of non-normality is to make use of the Bollen-Stine p-value (instead of the maximum likelihood p-value) that is used to correct for non-normality when assessing overall model fit (http://ssc.utexas.edu/software-faqs/amos). This is however not an important aspect in any SEM if sufficient goodness of fit is achieved, as non-normality of the data in essence increases the probability of rejecting “models that may not be false”, in other words a type 1 error (http://ssc.utexas.edu/software-faqs/general#nonnormdatainsem).

Mahalanobis D2

As was shown in Section 5.1.4, the item responses are skew by nature (despite a relative normal kurtosis). To investigate if this would have an effect on model fit the 100 cases that had the most extreme scores, in other words, the scores farthest from the centroid as measured by the Mahalanobis distance (SPSS 22, 2013) were removed. The Mahalanobis distance is the distance of a case from the centroid where the centroid is the point defined by the means of all the variables taken as a whole. The Mahalanobis distance demonstrates how far an individual case is from the centroid of all the cases for the predictor variables. (Burdenski, 2000, p. 19) A Mahalanobis D2 can be used in this case as it indicates where in the multidimensional space an observation lies in relation to the sample mean of the relevant multiple variables (the centroid) in a model (Hair et al., 2010, p. 66; Kline, 2011b, p. 54). If the resulting value is relatively high, then Burdenski (2000, p. 19) contends that the related observation is an outlier.

Exploratory Factor Analysis: Replication of Martins (2000)

In order to establish whether the combined dataset is comparable to the previous subsets that were used in the development of the Martins (2000) model, the next sections will investigate the factor structure of this expanded combined dataset of 12 393 cases. In the first instance, a principal component analysis (PCA) will be run to replicate the previous work of Martins and colleagues so as to determine if the factor loadings remained constant. Although this step would not be recommended in a pure confirmatory approach, the fact that Martins et al. (1997) and Martins (2000) both made use of item parcelling makes it necessary. The aim is to confirm the content of the item parcels and the consequent calculation of the values of these parcelled observed variables that will form part of the structural equation model. These purely empirical results will then be used as a basis for exploratory common factor analysis such as principal axis factoring (PAF). Furthermore, in preparation of the structural equation modelling, a factor analysis based on maximum likelihood extraction will be conducted.

Table of Contents :

• Chapter 1: Scientific orientation to the research
  • 1.1 Importance of trust research
    • 1.1.1 Global state of trust
    • 1.1.2 Outcomes of employee trust
  • 1.2 Relevant base models of trust
  • 1.3 The paradigm perspective
  • 1.4 Definition of trust
  • 1.5 Measurement of trust
    • 1.5.1 Purpose
    • 1.5.2 Dimensions of the Trust Relationship Audit
    • 1.5.3 Personality aspects
    • 1.5.4 Managerial practices
    • 1.5.5 Trust relationship
    • 1.5.6 Additional dimensions
    • 1.5.7 Psychometric properties
    • 1.5.8 Data collection
    • 1.5.9 Data processing
  • 1.6 Structure of the research: Conceptual framework
    • 1.6.1 Research problem
    • 1.6.2 Aims and objective of the study
    • 1.6.3 Conceptual framework
      • 1.6.3.1 Integrative findings
      • 1.6.3.2 Conclusion
      • 1.6.3.3 Limitations of the research
      • 1.6.3.4 Recommendations
• 1.7 Chapter layout
  • 1.8 Research method
    • 1.8.1 Phase 1: Literature review
    • 1.8.2 Phase 2: Empirical study – research procedure
      • 1.8.2.1 Research participants
      • 1.8.2.2 Measuring instrument
      • 1.8.2.3 Research variables
• Chapter 2: The Concept of Trust
  • 2.1 The state of trust
    • 2.1.1 The need for a trans-disciplinary approach and multiple units of analysis
    • 2.1.2 Increased importance of trust in the academic literature
  • 2.2 The importance of trust on a macro-economic level
    • 2.2.1 Social capital
      • 2.2.1.1 Reputation
      • 2.2.1.2 Family ties and social capital
    • 2.2.2 Trust, the Internet and the virtual world of work
    • 2.2.3 Occupational level and trust
    • 2.2.4 Operational and production implications
  • 2.3 Importance of trust to the organisation
    • 2.3.1 Trust and control
    • 2.3.2 Implications for HR Practices
    • 2.3.3 Learning and innovation
    • 2.3.4 Global competiveness
    • 2.3.5 Organisational support and commitment
    • 2.3.6 Survival in complex organisational settings
  • 2.4 Defining trust
  • 2.5 Models of trust relevant to the current research
    • 2.5.1 The Mayer et al. (1995) model
      • 2.5.1.1 Trustworthiness
      • 2.5.1.2 Trust and risk taking
    • 2.5.2 Critique of the Mayer et al. (1995) model
    • 2.5.3 Evaluation of the Mayer et al. (1995) model
    • 2.5.4 Support for the Mayer et al. (1995) model
    • 2.5.5 Reynolds’ model (1997)
  • 2.6 Types of trust
    • 2.6.1 Deterrence- or calculus-based trust
    • 2.6.2 Knowledge-based trust
    • 2.6.3 Identification-based trust
    • 2.6.4 Hierarchy of trust
• Chapter 3: Trust in practice
  • 3.1 Repairing trust relationships
    • 3.1.1 Maintaining and enhancing trust
    • 3.1.2 Organisation level trust repair
    • 3.1.3 Enhancing trust in organisations – the role of presumptive trust
    • 3.1.4 Enhance or increase trust levels by means of information sharing
  • 3.2 Trust in leadership
    • 3.2.1 Trust in leadership/ supervisor or trust in the organisation?
    • 3.2.2 The spiral process of trust in the leader
  • 3.3 Co-worker trust – using different referents as foci
    • 3.3.1 Trust and teams
    • 3.3.2 Interpersonal trust in the organisational context – a social network
  • 3.4 The Martins (2000) model for managing trust
    • 3.4.1 Personality as antecedent of trust
      • 3.4.1.1 Extraversion
      • 3.4.1.2 Agreeableness
      • 3.4.1.3 Conscientiousness
      • 3.4.1.4 Emotional stability
      • 3.4.1.5 Resourcefulness (openness to experience/intellect)
    • 3.4.2 Linking personality and trust on a conceptual level
    • 3.4.3 Managerial practices as antecedents of trust
    • 3.4.4 The trust relationship
    • 3.4.5 Scale reliability and validity, including organisational trust and change dimensions
    • 3.4.6 Personality, knowledge sharing and trust
    • 3.4.7 Managerial practices, knowledge sharing and trust
    • 3.4.8 Information sharing
    • 3.4.8.1 Information sharing or knowledge sharing
    • 3.4.8.2 Knowledge sharing and trust in organisations
  • 3.5 South African trust-related research
  • 3.6 Summary
• Chapter 4: Empirical Research Design and Methodology
  • 4.1 Statistical modus operandi
  • 4.1.1 Sample size
    • 4.1.2 Descriptive statistics
    • 4.1.3 Likert scales or Likert items
    • 4.1.4 Cronbach’s alpha
    • 4.1.5 Effect size
    • 4.1.6 Negatively worded items
    • 4.1.7 Web-survey implications
    • 4.1.8 Measurement errors
    • 4.1.9 Control variables
  • 4.2 Survey methodology
  • 4.3 Structural Equation Modelling
• Chapter 5: Results
• Chapter 6: Findings, Conclusions, Limitations and Recommendations

GET THE COMPLETE PROJECT
AN EXPLORATORY STUDY ON ORGANISATIONAL TRUST RELATIONSHIPS