Structural Equation Modeling

May 2012 Dr. Lin 1

Structural Equation Modeling Dr. Binshan Lin

BellSouth Professor [email protected]

May 2012Kasetsart University PhD Workshop, Thailand

Instructor Profile Dr. Binshan Lin is the BellSouth Corporation Professor at Louisiana State

University in Shreveport (LSUS). He received his Ph.D. from the Louisiana State University in 1988. He is an nine-time recipient of the Outstanding Faculty Award at LSUS. Professor Lin receives the Computer Educator of the Year by the International Association for Computer Information Systems (IACIS) in 2005, Ben Bauman Award for Excellence in IACIS 2003, Distinguished Service Award at the Southwest Decision Sciences Institute (SWDSI) in 2007, Outstanding Educator Award at the SWDSI in 2004, and Emerald Literati Club Awards for Excellence in 2003.

Dr. Lin has published over 260 articles in refereed journals, and currently serves as Editor-in-Chief of Industrial Management & Data Systems.

Professor Lin serves as President of SWDSI (2004-2005), Program Chair of IACIS Pacific 2005 Conference, Program Chair of Management International Conference (MIC) 2006, General Chair of MIC Conference (2007 and 2008). In addition, Dr. Lin serves as Program Chair of Technology Innovation and Industrial Management (TIIM) International Conference 2009, Conference Director of TIIM Conference (2010-present), and Conference Director of MakeLearn International Conference (2012-present). Dr. Lin also serves as a vice president (2007-2009; 2010-2012) of Decision Sciences Institute (DSI).

2May 2012 Dr. Lin

May 2012 Dr. Lin 3

Dr. Sewall Wright1889-1988

1st paper in 1920

May 2012 Dr. Lin 4

The Wright Idea

Y1 = α1 + β1X + ε1i

Y2 = α2 + β2X + β3Y1 + ε2i

X Y1 ε1iY2

ε2i

May 2012 Dr. Lin 5

Structural equation modeling (SEM) is a statistical technique for testing and estimating causal relations using a combination of statistical data and qualitative causal assumptions. 

Definition

May 2012 Dr. Lin 6Starfield and Bleloch (1991)

Understanding of Processes

univariate descriptive statistics

exploration, methodology and

theory development

realistic predictive

models

abstract models

multivariate descriptive statistics

more detailed theoretical

models

univariate data

modeling

multivariate data

modeling

Data

SEM

May 2012 Dr. Lin 7

J.C. Westland, ECRA, 2010.

May 2012 Dr. Lin 8

Do the conventional methods meet your needs?

All your greatscientific ideas! ANOVA result you

hope to get!

May 2012 Dr. Lin 9

May 2012 Dr. Lin 10

Testing the Purchase Funnel

Awareness Consideration Purchase

Media

May 2012 Dr. Lin 11

There is no consensus on a single definition for TQM.

We see TQM as a business-level strategy or management process.

Its components of process and content are necessary but not sufficient conditions for success.

TQM

May 2012 Dr. Lin 12

TQM is defined as a holistic management philosophy that strives to satisfy customer needs and expectations through continuous improvement efforts in every function and process within an organization

TQM

May 2012 Dr. Lin 13

Role Conflict

May 2012 Dr. Lin 14

Occurs when different expectations impinge concurrently, resulting in “dissonance” for the individual who aims to perform the incompatible roles (Lynch, 2007)

Higher Quantity vs. Higher Quality As a mediator variable in a causal model

of employee behaviour

Role Conflict

May 2012 Dr. Lin 15

Cause vs. Effect

Effect of a Cause (Description)◦What follows a cause?

Cause of an Effect (Explanation)◦Why did the effect happen?

Do bacteria “cause” disease?◦Actually toxins cause disease◦Actually certain chemical reactions are cause

Holland, P. W. (1988). “Causal inference, path analysis, and recursive structural equations models” Sociological Methodology, 18, 449-484.

Multiple Regression Causal ModelingX1

X2

X3

X4

X5

Y

How well do predictors predict in Y? What are independent effects when effects of other variables are controlled?

X1

X3 X4

X2 X5

Y

How well do predictorsrelate with regard to ultimateprediction of Y?

May 2012 Dr. Lin 17

Latent variables (as opposed to observable variables), are variables that are not directly observed but are rather inferred from other variables that are observed (directly measured). 

Latent Variables

May 2012 Dr. Lin 18

J.C. Westland, ECRA, 2010.

May 2012 Dr. Lin 19

J.C. Westland, ECRA, 2010.

May 2012 Dr. Lin 20

Conceptualizing Latent Variables Latent variables:

representation of the variance shared among the variables

TotalVariance

CommonVariance

UniqueVariance

SpecificVariance

RandomError

May 2012 Dr. Lin 21

A mediation model is one that seeks to identify the mechanism that underlies the relationship between an IV and a DV via the inclusion of a 3rd explanatory variable, known as a mediator variable.

Mediator Variable

May 2012 Dr. Lin 22

The perception that an individual lacks information required to perform a job or task, leading one to feel deserted (Onyemah, 2008)

Job description Operating manual IS managers dealing with unclear and varying

expectations from end users Positive relationship between role conflict and

role ambiguity experienced by employees

Role Ambiguity

May 2012 Dr. Lin 23

Six dimensions of TQM practices are assessed using an adapted version of scales developed by Prajogo et al. (2007), Prajogo and Sohal (2006), Samson and Terziovski (1999), Sohail and Teo (2003) and Zhang et al. (2000).

42 items are grouped into six segments to measure the different dimensions of TQM practices: leadership, strategic planning, customer focus, human resource focus, process management and information analysis.

The response format is a 5-point Likert type scale ranging from “strongly disagree” to “strongly agree”.

TQM Measurement

May 2012 Dr. Lin 24

Role conflict and role ambiguity are measured using scales developed by Rizzo et al. (1970).

The scales developed have been extensively validated and have established records for its psychometric properties.

A 5-point Likert type scale is utilized ranging from “strongly disagree” to “strongly agree”.

Role Conflict & Role Ambiguity

May 2012 Dr. Lin 25

Step #1: Determine the individual constructs Theory identifies the items to be used as

measurement variables Theoretical constructs should be operationalized from

scales of prior research or through new scales

Six Steps of SEM Process

May 2012 Dr. Lin 26

Step #2: Develop & specify the measurement model A path diagram should be drawn Representation of the entire set of relationships that

constitutes a SEM Step #3: Designing a Study to Produce Empirical

Results Step #4: Assessing the measurement model validity Step #5: Specify structural model Step #6: Assess structural model validity

Six Steps of SEM Process

May 2012 Dr. Lin 27

An assessment of the degree of consistency between multiple measurements of the same variable

Concerned with whether alternative measurements at different times would reveal similar information

Internal consistency reliability: Cronbach’s alpha coefficient α > 0.5 or 0.6

Reliability

May 2012 Dr. Lin 28

The extent to which measure(s) correctly represent the constructs of study

Concerned with how well the construct is defined by the measure(s)

Validity

May 2012 Dr. Lin 29

Leadership

Strategic Planning

Role Conflict

Information Analysis

Process Management

Human Resource

Focus

Customer Focus

Role Ambiguity

TQ

M Practices

H6c

H6b

H6a

H2a

H2c H3a

H3b

H3c

H4a

H4b

H4c

H5a

H5c

H2b

H5b

H1

H7a

H7b

H7c

May 2012 Dr. Lin 30

The unit of analysis for this research is individual - the full-time salaried employees of ISO 9001:2000 certified organizations in Malaysia.

ISO 9000 standard is a base for organizations to apply and certify a management system in relation to quality management.

ISO 9000 certification is granted to the firms after they demonstrate that they have mapped operating processes associated with the quality of their products, and that they have complied with these repeatable, standardized and documented processes.

In 2011 the questionnaires were distributed to 100 ISO certified firms listed in the Federation of Malaysian Manufacturers (FMM) Directory.

Samples & Procedures

May 2012 Dr. Lin 31

98 organizations (35 manufacturing firms + 63 service firms).

A total of 650 questionnaires are distributed and 453 are completed and returned.

31 questionnaires have to be excluded as outliers. The outliers are detected using the graphical method, that is, residuals scatter plot (±3 std dev).

422 returns are used for analysis, with net response rate of 65%.

Sampling

May 2012 Dr. Lin 32

Lower Bound of Sample Size

Large Sample Size

SEM researchers suggest a sample size of at least ten times the number of parameters we will be estimating.

May 2012 Dr. Lin 33

Profile Percentage (%) Profile Percentage (%)

Age   Length of Service  

< 21 years old 0.71% More than 6 months but less than 1 year

20.62% 21-25 years old 23.46% 1–2 years 24.41% 26-30 years old 35.55% 3-5 years 20.61% 31-35 years old 16.11% 6-10 years 15.40% 36-40 years old 11.14% 11-20 years 14.69%41 or above 13.03% Above 20 years 4.27%Qualifications   Type of Work  

No college degree 10.19% Administration 37.44%

Diploma 15.40% Production 20.62%

Bachelor degree/ Professional qualification

59.01%

Computer and IT

26.54%

Master degree 13.74% Sales and marketing 15.40%

PhD degree 1.66%    

Profiles of the Survey Respondents

May 2012 Dr. Lin 34

Measurement Model involves the development of measurement models using confirmatory factor analysis (CFA) to achieve the best fitting group of items to represent each measurement scale.

The 2nd model (Structural Model 1) examines the relationships between TQM practices and role conflict.

The 3rd model (Structural Model 2) examines the relationship between TQM practices and role ambiguity.

The 4th model (Structural Model 3) examines the relations among TQM practices, role conflict and role ambiguity as well as the mediating effect of role conflict between TQM practices and role ambiguity simultaneously.

Four Models

May 2012 Dr. Lin 35

  LD SP CF HR PM IA RC RALD 0.864              SP 0.729**

(0.069)0.861            

CF 0.502**(0.037)

0.711**(0.061)

0.839          

HR 0.646**(0.081)

0.699**(0.078)

0.594**(0.064)

0.894        

PM 0.640**(0.056)

0.735**(0.060)

0.651**(0.054)

0.754**(0.095)

0.852      

IA 0.588**(0.051)

0.699**(0.059)

0.649**(0.058)

0.671**(0.082)

0.734**(0.069)

0.875    

RC -0.293**(0.005)

-0.322**(0.005)

-0.294**(0.005)

-0.263**(0.005)

-0.361**(0.007)

-0.373**(0.008)

0.668  

RA -0.377**(0.009)

-0.442**(0.010)

-0.343**(0.007)

-0.366**(0.010)

-0.456**(0.011)

-0.428**(0.010)

0.591**(0.008)

0.761

Correlations and Composite Reliabilities for All Variables

* p < 0.05; ** p < 0.01; *** p < 0.001; LD=Leadership; SP=Strategic planning; CF=Customer focus; HR=Human resource focus; PM=Process management; IA=Information analysis; RC=Role conflict; RA=Role ambiguity.

May 2012 Dr. Lin 36

  Model Fit Indices

χ² / df GFI AGFI RMSEA NFI CFI TLI

≤ 3 a ≥ 0.80 b ≥ 0.80 b ≤ 0.05 c ≥ 0.80 b ≥ 0.90 d ≥ 0.90 e

Measurement Model 1.655 0.882 0.861 0.039 0.887 0.952 0.946

Structural Model 1 1.578 0.874 0.854 0.037 0.870 0.948 0.942

Structural Model 2 1.598 0.866 0.845 0.038 0.862 0.943 0.937

Structural Model 3 1.538 0.858 0.838 0.036 0.847 0.940 0.934

Model Fit Indices for the Measurement & Structural Models►(Chau & Hu, 2011)►Goodness-of-Fit Indices (Forza & Filippini, 1998)►Adjusted Goodness-of-Fit Indices (Forza & Filippini, 1998) ►Root Mean Square Error Approximation (Browne & Cudeck, 1993)►Normal Fit Index (Forza & Filippini, 1998)►Comparative Fit Index (Hair et al, 2010)►Tucker-Lewis Index: (Vanderberg & Scarpello, 1994)  

May 2012 Dr. Lin 37

Path Coefficients for Structural Model 3

* p < 0.05; ** p < 0.01; *** p < 0.001; LD=Leadership; SP=Strategic planning; CF=Customer focus; HR=Human resource focus; PM=Process management; IA=Information analysis; RC=Role conflict; RA=Role ambiguity.

Hypotheses Causal Path

Path Coefficients Critical Ratios p-value

H1 RC RA 0.752 6.070 0.000***H2a LD RC -0.140 -1.270 0.204H2b LD RA 0.102 1.072 0.284H3a SP RC 0.154 0.685 0.493H3b SP RA -0.351 -1.784 0.074†H4a CF RC -0.022 -0.175 0.861H4b CF RA 0.220 1.974 0.048*H5a HR RC 0.242 2.949 0.003**H5b HR RA 0.045 0.647 0.518H6a PM RC -0.356 -2.572 0.010*H6b PM RA -0.166 -1.410 0.159H7a IA RC -0.282 -3.049 0.002**H7b IA RA 0.028 0.362 0.717

May 2012 Dr. Lin 38

The hypotheses H1, H3b, H4b, H5a, H6a and H7a are empirically supported.

However, the findings do not support hypotheses H2a, H2b, H3a, H4a, H5b, H6b and H7b because the respective path coefficients are not significant in the predicted directions.

Hypotheses Testing

May 2012 Dr. Lin 39

Tests of Mediating Effects of Role Conflict on the TQM - Role Ambiguity Relation

* p < 0.05; ** p < 0.01; *** p < 0.001; Mediator = Role conflict; DV=Role ambiguity; IV=Independent variables

Constructs

(Hypotheses)

Baron & Kenny Test

Coefficients of

Structured Model 1 (IV Mediator)

Coefficients of

Structured Model #2 (IV

DV)

Coefficients of Structured Model #3 (IV DV, mediator controlled)

Leadership (H2c) -1.230 -0.134 -0.004 0.102Strategic Planning (H3c) 0.671 0.151 -0.233 -0.351Customer Focus (H4c) -0.170 -0.018 0.198 0.220*Human Resource Focus (H5c)

2.625** 0.232** 0.216** 0.045

Process Management (H6c)

-2.347** -0.352* -0.416*** -0.166

Information Analysis (H7c)

-2.683** -0.276** -0.169* 0.028

Role Conflict (H1) - - - 0.752***

May 2012 Dr. Lin 40

The Baron and Kenny (1986) statistic is used to test for the significance of the mediating effect.

Three regression equations are used to test for the mediation model and the following three conditions must hold to establish the mediation.

First, the independent variables must be shown to be significantly related to the mediator in structural model 1.

Second, the independent variables must be shown to be significantly related to the dependent variable in structural model 2.

Third, the mediator must affect the dependent variable in structural model 3.

Barton & Kenny Test

May 2012 Dr. Lin 41

The mediator (role conflict) is significantly related to the dependent variable (role ambiguity) in Structural Model 3, while human resource focus (β = 0.045, p > 0.05), process management (β = -0.166, p > 0.05), and information analysis (β = 0.028, p > 0.05) are found to have no significant relationship with role ambiguity.

Role conflict is found to be a full mediator between the following: human resource focus and role ambiguity; process management and role ambiguity; information analysis and role ambiguity.

Thus, H5c, H6c and H7c are statistically supported.

Result

May 2012 Dr. Lin 42

The negative relationships between two TQM practices (i.e., process management and information analysis) and role conflict provide incentives for industrial practitioners.

In order to reduce the levels of role conflict among employees, the organizational administrators and managers are incentivised to develop appropriate implementation procedures to enhance the process management as well as to improve efficient use of information analysis.

Implication #1

May 2012 Dr. Lin 43

The industrial practitioners must be attentive to the pressures of customer focus which increase employees’ role ambiguity.

Using behaviour-based evaluation gives employees more control over their evaluations, thereby reducing employees’ role ambiguity.

Implication #2

May 2012 Dr. Lin 44

The organizational administrators and managers must be aware that the presence of role conflict inevitably leads to higher levels of role ambiguity.

On the other hand, role conflict appears to be a full mediator influencing several TQM practice.

One effective way to alleviate role ambiguity is to eliminate, if not reduce, the conflicting roles and expectations communicated to an individual.

Implication #3