See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267815037

Partial least square in a nutshell | Saiyidi MAT RONI 2 0 1 4

Chapter · November 2014

DOI: 10.13140/RG.2.1.4125.4245

CITATIONS READS

0 6,259

1 author:

Saiyidi Mat Roni

Edith Cowan University
46 PUBLICATIONS   150 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Security and trust equilibrium for online transactions View project

Research project: Goods & services tax noncompliance model View project

All content following this page was uploaded by Saiyidi Mat Roni on 06 November 2014.

The user has requested enhancement of the downloaded file.

2 0 1 4
Saiyidi MAT RONI

Saiyid
Digitally signed by
Saiyidi MAT RONI
DN: cn=Saiyidi
MAT RONI, o=Edith

i MAT Cowan University,

ou,
email=m.matroni@

RONI ecu.edu.au, c=AU

Date: 2014.11.06
14:52:15 +08'00'

1 Partial least square in a nutshell | Saiyidi MAT RONI

1 November 2014
 1 Partial least square in a nutshell

Partial least square (PLS) a robust structural equation modelling (SEM) approach. It is
sometimes referred as component-based SEM or simply PLS-SEM. PLS-SEM is a causal
modelling statistical approach with an aim to maximise explained variance of dependent
latent variables (Chin, 1998b; Hair, Ringle, & Sarstedt, 2011).

Unlike covariance-based SEM, e.g. provided by AMOS and LISREL, PLS-SEM does
not put emphasis on normality of data distribution (Ringle, Sarstedt, & Straub, 2012), which
allows researchers to use their raw data in the analysis without having to transform their data
to make it at least, approximate normally distributed, prior to running SEM. Although some
studies suggest choosing PLS-SEM over covariance-based SEM on a basis of non-normal
data distribution is a weak argument, the fact that violation of normality assumption can
produce unintended biases in the final statistical result (or no solution at all), it is prudent to
opt for PLS-SEM to alleviate two serious issues with covariance-based SEM: improper
solutions where solution is beyond what is gauged by parameters, and factor indeterminacy
(Fornell & Bookstein, 1982).

For more in-depth review of PLS-SEM features Hair et al. (2011), Chin (1998b), Wold
(1985) and Hair Jr, Sarstedt, Hopkins, and Kuppelwieser (2014) are good references to begin
with.

2 Partial least square in a nutshell | Saiyidi MAT RONI

1 November 2014
 2 Approach to structural equation modelling (SEM)

2.1 Preliminary data analysis

As in most cases, before an actual analysis can be performed, it is advisable to run

preliminary data analysis (PDA). PDA is recommended to ensure the dataset is ‘cleaned’ and
‘cleansed’ of substantial noise that attract biases in the final results of the actual analysis. The
following diagram illustrates a common approach to PDA.

Screen data Missing value analysis

•Delete monotones •Choose MI or EM

AVE = average variance Check outliers Check normality

extracted
CMB = common method
•Trim or Winsor •transform if necessary
bias
EM = Expected
maximisation
MI = Multiple
imputations
ML = Maximum
likelihood
NRB = Non-response
bias
PAF = Principal axis Run factor analysis Reliability & validity
factoring
PCA = Principal •Extraction: PCA, PAF or ML •Cronbach's alpha
component analysis •Rotation: Orthogonal or • AVE
oblique

Address biases
•CMB -> Harman's single
factor score
•NRB -> split half, then test
means difference

3 Approach to structural equation modelling (SEM) | Saiyidi MAT RONI

1 November 2014
2.2 Measurement model

Once PDA is done, the next step to make before the interpretation of the output of a
structural model, is to look at the measurement model. Assessments of measurement model
than later interpret the result of the structural model are two-stage SEM approach suggested
by Anderson and Gerbing (1988) and Hair et al. (2011). Measurement model stage involves
an assessment of model parameters which included reliability test using Cronbach’s alpha
and composite reliability (Kock, 2013; Urbach & Ahlemann, 2010), validity analyses through
average variance extracted (AVE) and standardised factor loadings and cross-loadings
(Fornell & Larcker, 1981; Hair, Black, Babin, & Anderson, 2010; Kline, 2010; Schumacker
& Lomax, 2012), and also an assessment to determine the nature of construct being a
formative or reflective (Chin, 1998a). The quality of the measurement model is further
assessed on lateral and vertical collinearity using variance inflation factor (VIF) (Kock &
Lynn, 2012). The criteria for the assessment is summarised in Table 1 below.

Table 1: Measurement model assessment criteria.

Assessment Criterion Note Reference

Item reliability Individual item standardised Min. of .50 Hair et al. (2010)
loading on parent factor.

Convergent Individual item standardised Min. of .50 Hair et al. (2010)

validity loading on parent factor, and

Loadings with sig. p-value p < .05 Gefen and Straub

(2005)

Composite reliability > .70 Fornell and Larcker

(1981)
Nunnally and
Bernstein (1994)
Hair et al. (2010)

Average variance extracted > .50 Hair et al. (2010)

(AVE) Urbach and
Ahlemann (2010)

Discriminant Square-root of AVE More than the Hair et al. (2010)

validity correlations of
the latent
variables.

4 Approach to structural equation modelling (SEM) | Saiyidi MAT RONI

1 November 2014
Reliability Cronbach’s alpha > .70 Nunnally and
Bernstein (1994)
Urbach and
Ahlemann (2010)
Hair et al. (2010)

Variance inflation factor (VIF) < 10 Hair et al. (2010)

<5 Kock and Lynn

(2012)

Nature of Formative / reflective: Chin (1998a)

construct Coltman, Devinney,
Midgley, and Veniak
(2008)

5 Approach to structural equation modelling (SEM) | Saiyidi MAT RONI

1 November 2014
2.3 Structural model

After measurement model is found to be satisfactory, structural model parameter

estimates then can be used for analyses and interpretation. The structural model is evaluated
through coefficient of determination, R2(Chin, 1998a, 1998b), predictive relevance, Q2
(Geisser, 1975; Stone, 1974), effect size, f2 (Cohen, 2013), and path coefficients (see Hair et
al., 2011; Mohamadali, 2012). These criteria are summarised in Table 2 below.

Table 2: Structural model criteria.

Criterion Note Reference

2
Coefficient of determination, R .67 substantial Chin (1998a)
.33 average
.19 weak
Predictive relevance, Q2 >0 Geisser (1975)
Stone-Geisser Stone (1974)
test

Effect size, f 2 .02 small Cohen (2013)

.15 medium
.35 large

Path coefficient Magnitude Hair et al. (2010)

Sign
p-value

6 Approach to structural equation modelling (SEM) | Saiyidi MAT RONI

1 November 2014
 3 WarpPLS

3.1 Start WarpPLS

Double-click on the icon > Proceed to use software

7 WarpPLS | Saiyidi MAT RONI

1 November 2014
3.2 Transfer data set from Excel to WarpPLS readable format

Data file: PLS.FullData.xlsx

If your dataset is in SPSS format (.sav):
File > Save As > Type in file name > Save as type: > select Excel 2007 through 2010 (*.xlsx)
> Save.

Open PLS.FullData.xlsx > Your data should look like this:

8 WarpPLS | Saiyidi MAT RONI

1 November 2014
On WarpPLS:
Proceed to Step 1 > Create project file > File name > Type BasicModel > Save.

Proceed to Step 2 > Read from file > Files of type: > (*.xlsx) > Select PLS.FullData > Open.

9 WarpPLS | Saiyidi MAT RONI

1 November 2014
Your data preview should look like this.

Next > Finish > Check your data set > OK > Yes.

Proceed to Step 3 > Pre-process data > OK.

10 WarpPLS | Saiyidi MAT RONI

1 November 2014
Check your data > Yes.

On main window: Project > Save Project.

11 WarpPLS | Saiyidi MAT RONI

1 November 2014
 4 Basic model

Data file: BasicModel.prj

3 Predictors (independent variables) and 1 criterion (dependent variable).

4.1 Create latent variables

Proceed to Step 4 > Define SEM Model > Latent variable options > Create latent variable >
Click anywhere on the canvas (white area).

12 Basic model | Saiyidi MAT RONI

1 November 2014
Latent variable name: > type EASE > Add indicators: Click PEOU1 > Add.
Repeat the Add indicators steps for PEOU2, …, PEOU6 > Save > Save latent variable
settings > OK.

13 Basic model | Saiyidi MAT RONI

1 November 2014
4.2 Create direct links among latent variables
Proceed to Step 4 > Define SEM Model > Direct link options > Create direct link.
Click EASE > Click INTENT > WarpPLS creates the line in the model.

Repeat the process for USEFUL-INTENT and ATTITUDE-INTENT.

Model options > Save model and close > OK.

Proceed to Step 5 > Perform SEM analysis > your result should look like this.

14 Basic model | Saiyidi MAT RONI

1 November 2014
4.3 Measurement model

On main window > View/save analysis results.

Examine the value of each criterion and compare them against the thresholds listed in
the table below.

Assessment Criterion Note Steps

Item reliability Individual item standardised Min. of .50 View > View indicator
loading on parent factor. loadings and cross-
loadings > View
combined loadings
and cross-loadings.

Convergent Individual item standardised Min. of .50 View > View indicator
validity loading on parent factor. loadings and cross-
loadings > View
combined loadings
and cross-loadings

Loadings with sig. p-value < .05

Composite reliability > .70 View > View latent

variable coefficients.

Average variance extracted > .50 View > View latent

(AVE) variable coefficients.

Discriminant Square-root of AVE More than the View > View

validity correlations of correlations among
the latent latent variables and
variables. errors > View
correlations among
latent variables with
sq. rts. of AVEs.

Reliability Cronbach’s alpha > .70 View > View latent

variable coefficients.

Variance inflation factor < .50 View > View latent

(VIF) variable coefficients.

15 Basic model | Saiyidi MAT RONI

1 November 2014
Nature of Formative / reflective: Look for View > View causality
construct Simpson’s assessment
paradox coefficients > View
indication path-correlation sign

4.4 Structural model

On main window > View/save analysis results.

Examine the value of each criterion and compare them against the thresholds listed in
the table below.

Criterion Note Steps

Coefficient of determination, R2 .67 substantial View > View latent
.33 average variable coefficients.
.19 weak
Predictive relevance, Q2 >0 View > View latent
Stone-Geisser test variable coefficients.

Effect size, f 2 .02 small View > View standard

.15 medium errors and effect sizes for
.35 large path coefficients.

Path coefficient Magnitude From the main structural

Sign model, or
p-value
View > View path
coefficients and P values.

16 Basic model | Saiyidi MAT RONI

1 November 2014
 5 Model with moderating variable

Data file: ModeratingVariable.prj

Moderating
variables

17 Model with moderating variable | Saiyidi MAT RONI

1 November 2014
 6 Model with mediator variable

Data file: MediatingVariable.prj

Mediating
variable

18 Model with mediator variable | Saiyidi MAT RONI

1 November 2014
 7 Model with second-order factor

Data file: SecondOrderFactor.prj

Moderating
. variable.

Second-order factor comprises of identify

and identity latent variables.

Proceed to Step 5 > Perform SEM analysis > Close window.

On the main window > Modify > Add one or more latent variable…

19 Model with second-order factor | Saiyidi MAT RONI

1 November 2014
Latent variable to be added: > select internal > Add > OK.

> select identify > Add > OK > Close.

Proceed to Step 4 > Yes > Define SEM model.

Latent variable name: > Type SOCIAL > Add indicators: > choose lv_internal > Add >
choose lv_identify > Add.

Save > Save latent variable settings.

20 Model with second-order factor | Saiyidi MAT RONI

1 November 2014
 8 Model with control variable

Data file: ControlVariable.prj

Moderating
variable
Control
variable

21 Model with control variable | Saiyidi MAT RONI

1 November 2014
 9 References

Anderson, J. C., & Gerbing, D. W. (1988). Structural equation modeling in practice: A

review and recommended two-step approach. Psychological Bulletin, 103(3), 411-
423. doi: 10.1037/0033-2909.103.3.411
Chin, W. W. (1998a). Issues and opinion on structural equation modeling. MIS Quarterly,
22(1), VII-XVI.
Chin, W. W. (1998b). The partial least squares approach to structural equation modelling. In
G. A. Marcoulides (Ed.), Modern methods for business research (pp. 295-336).
Mahwah, New Jersey: Lawrence Erlbaum Associates.
Cohen, J. (2013). Statistical Power Analysis for the Behavioral Sciences (2 ed.). Hoboken:
Taylor and Francis.
Coltman, T., Devinney, T. M., Midgley, D. F., & Veniak, S. (2008). Formative versus
reflective measurement models: Two applications of formative measurement. Journal
of Business Research, 61(12), 1250-1262.
Fornell, C., & Bookstein, F. (1982). Two structural equation models: LISREL and PLS
applied to consumer exit-voice theory. Journal of Marketing Research, 19, 440-452.
Fornell, C., & Larcker, D. F. (1981). Evaluating Structural Equation Models with
Unobservable Variables and Measurement Error. JMR, Journal of Marketing
Research, 18(1), 39.
Gefen, D., & Straub, D. (2005). A practical guide to factorial validity using PLS-graph:
Tutorial and annotated example. Communications of the Association for Information
Systems, 16, 91-109.
Geisser, S. (1975). The Predictive Sample Reuse Method with Applications. Journal of the
American Statistical Association, 70(350), 320-328. doi:
10.1080/01621459.1975.10479865
Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2010). Multivariate data analysis
(7 ed.). Upper Saddle River, NJ, USA: Prentice-Hall, Inc.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2011). PLS-SEM: Indeed a silver bullet. Journal of
Marketing Theory and Practice, 19(2), 139-151. doi: 10.273/MTP1069-6679190202
Hair Jr, J. F., Sarstedt, M., Hopkins, L., & Kuppelwieser, V. G. (2014). Partial least squares
structural equation modeling (PLS-SEM). European Business Review, 26(2), 106-121.
doi: doi:10.1108/EBR-10-2013-0128

22 References | Saiyidi MAT RONI

1 November 2014
 Kline, R. B. (2010). Principles and Practice of Structural Equation Modeling, Third Edition
(3 ed.). New York: Guilford Publications.
Kock, N. (2013). WarpPLS 4.0 User Manual. Loredo, Texas: ScriptWarp Systems.
Kock, N., & Lynn, G. S. (2012). Lateral Collinearity and Misleading Results in Variance-
Based SEM: An Illustration and Recommendations. Journal of the Association for
Information Systems, 13(7), 546-580.
Mohamadali, N. A. K. (2012). Exploring new factors and the question of ‘which’ in user
acceptance studies of healthcare software. (Doctor of Philosophy), University of
Nottingham, Nottingham. Retrieved from
Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory. New York: McGraw Hill.
Ringle, C. M., Sarstedt, M., & Straub, D. W. (2012). Editor's comments: a critical look at the
use of PLS-SEM in MIS quarterly. MIS Quarterly, 36(1), iii-xiv.
Schumacker, R. E., & Lomax, R. G. (2012). A Beginner's Guide to Structural Equation
Modeling : Third Edition (3 ed.). Hoboken: Taylor and Francis.
Stone, M. (1974). Cross-Validatory Choice and Assessment of Statistical Predictions. Journal
of the Royal Statistical Society. Series B (Methodological), 36(2), 111-147. doi:
10.2307/2984809
Urbach, N., & Ahlemann, F. (2010). Structural Equation Modeling in Information Systems
Research Using Partial Least Squares. Journal of Information Technology Theory and
Application, 11(2), 5-40.
Wold, H. (1985). Partial least square. In S. Kotz & N. L. Johnson (Eds.), Encyclopedia of
statistical sciences (Vol. 6, pp. 581-591). New York: Wiley.

23 References | Saiyidi MAT RONI

1 November 2014

View publication stats