Scalable Learning Analytics and Interoperability – an assessment of potential...
Open Universiteit Pls Prestation November 1st
1. PLS Tutorial
prepared for the PhD School
November 1st, 2012
Dr. Paul Ghijsen
Associate Professor Supply Chain Management
Department Marketing and Supply Chain Management
2. Preamble (1/2)
• Tool/ technique. Not aim (for me)
• Easy to use
• Often applicable – real life / complex situation
Suggestion is limited to the presentation of PLS-PM
results for a manuscript
PLS Tutorial Paul Ghijsen November 1st 2012
3. Preamble (2/2)
• Add to existing literature …
• Specific problem statement
• Limited number of research questions
• Force yourself to answer in one sentence : what is the
topic of your research and what should the answer
look like (in the end)…
• Make a graph of the relationships -> conceptual model
• Use literature to ‘compile’ questionnaire
PLS Tutorial Paul Ghijsen November 1st 2012
4. Overview tutorial
• Introduction SEM
• Focus on Partial Least Squares
– Especially SmartPLS, how to …
• Specify the model
• Derive results and
• Present results
• What you should think about
• Where to get it
PLS Tutorial Paul Ghijsen November 1st 2012
7. Introduction to SEM
• Factor-based covariance fitting approach for latent structural modeling:
LISREL, AMOS.
• Estimation methods:
– Maximum Likelihood and
– Generalized Least Squares / Generalized Method of Moments
• SEM with causal diagrams involve three primary components:
– Indicators (manifest variables or observed measures/variables)
– latent variable (construct, concept, factor)
– path relationships ( correlational, one-way paths, or two way paths).
– Take a look at David A. Kenny’s overview
PLS Tutorial Paul Ghijsen November 1st 2012
8. Reflective or formative indicators?
• Latent or Emergent Constructs: Covary or not
• Strong theory or not ?
PLS Tutorial Paul Ghijsen November 1st 2012
9. Introduction to SEM
• All the measured variance is useful variance to be explained.
A general model which encompasses, among other techniques,
canonical correlation, redundancy analysis, multiple regression,
multivariate analysis of variance, and principal components
• Sample size can be smaller: strong rule of thumb: equal to the
larger of the following: (1) ten times the scale with the largest
number of formative (i.e., causal) indicators, or (2) ten times the
largest number of structural paths directed at a particular
construct in the structural model.
PLS Tutorial Paul Ghijsen November 1st 2012
10. Focus on Partial Least Squares
• Powerful method of analysis because of the
minimal demands on measurement scales,
sample size, and residual distributions.
• PLS is considered better suited for explaining
complex relationships (Editorial LRP 2012)
PLS Tutorial Paul Ghijsen November 1st 2012
11. Steps to publish your results
• Download your data
• Data preparation or ‘cleaning’ your data
• Compute data descriptions using e.g. SPSS (table 1)
– Focus on Mean, STD, Skewness and Kurtosis
• Report on reliability test (Cronbach’s alpha) (table 2)
• Draw your inner model and your outer model
• Analyse the results from the PLS algorithm, Bootstrapping, FIMIX
• Use info on factor loadings and t-values to complete table 1 (outer model)
• Check convergent and discriminant validity
• Present table 4 on the results of the inner (or structural) model
PLS Tutorial Paul Ghijsen November 1st 2012
12. Steps to prepare your data
Step 1
Download the dataset in Spss format
Step 2
Replace values 8888888 & 9999999 with sysmis ('.')
Step 3
Recode control questions
Step 4
Replace the sysmis with series means
Step 5
Compute Cronbach’s alpha for the (existing) dimensions ( >.70)
Step 6
Delete (old) variables and / or respondents (with extreme care)
Step 7
Save file as 'tab-delimted“ or .csv
PLS Tutorial Paul Ghijsen November 1st 2012
13. Compute descriptives
• Focus on
– Mean
– Standard deviation
– Skewness
– Kurtosis
• Possibly correlations
• E.g. use SPSS
PLS Tutorial Paul Ghijsen November 1st 2012
14. Compute Reliability results
• Compute Cronbach’s alpha
– >= 0.6 for new variables
– >= 0.7 for existing variables
PLS Tutorial Paul Ghijsen November 1st 2012
16. Conceptual model
Influence of customer orientation on salesperson performance
PLS Tutorial Paul Ghijsen November 1st 2012
17. Conceptual model
Influence of customer education, expertise and experience on customer loyalty
PLS Tutorial Paul Ghijsen November 1st 2012
18. Convergent and discriminant validity
• Convergent:
• AVE >= 0.5
• Discriminant:
• Square root of AVE >= correlations (row and column)
PLS Tutorial Paul Ghijsen November 1st 2012
19. Goodness of Fit of the models
• FIMIX procedure in SmartPLS:
– Akaike Information Criterion (Analogous to Chi-square)
– Baysian IC (id.)
– Corrected AIC (id.)
– ENtropy, close to 0.5
PLS Tutorial Paul Ghijsen November 1st 2012
20. The SmartPLS tutorial videos
• http://youtu.be/-jegL4d-MXk Drawing a basic model
• http://www.youtube.com/watch?v=vNI7Cc48nRg Estimating the basic PLS path model
• http://www.youtube.com/watch?v=DX_OLuuP7lo Bootstrapping, p-values
• http://www.youtube.com/watch?v=juGqAMXKkZQ Epilogue
• http://www.youtube.com/watch?v=kalTHr8EKC0 How to handle missing data
• http://www.youtube.com/watch?v=acAl78zCqPU Basic concepts of PLS path modeling
• http://www.youtube.com/watch?v=RAL9X17r7CE PLS Path Modeling or Co-variance SEM
• WARPPLS:
• http://www.youtube.com/watch?v=yUojJaV3jlA
PLS Tutorial Paul Ghijsen November 1st 2012
21. What you should think about
• PLS is not a silver bullet:
– SEM co-covariance techniques also matter.
– Inadmissable solutions
– Factor indeterminancy
• Look at distribution of your data: ‘feel your data’.
• Aim for adequate data collection (adequate sample size)
PLS Tutorial Paul Ghijsen November 1st 2012
22. Where to get it?
• SmartPLS via www.smartpls.de
• PLSgraph via Wynne Chin (wchin@uh.edu ; http://disc-
nt.cba.uh.edu/chin/indx.html )
• WarpPLS via www.nedcock.com
• R via http://georgia-r-school.org/
PLS Tutorial Paul Ghijsen November 1st 2012