Application of Partial Least Squares SEM in PACS research Utrecht, October 13, 2010 A conceptual introduction Rogier van de Wetering
Application of Partial Least Squares SEM in PACS research
Utrecht, October 13, 2010
A conceptual introduction
Rogier van de Wetering
Overview of presentation
1
Goal: Providing a non-technical overview of PLS SEM• Application oriented
• No prior knowledge of SEM required
Agenda:
Part 1:• Short summary SEM
• What is Partial Least Squares?
Part 2:
• Introduction PACS research
• Application of PLS to PACS research
• DEMO PLS using SMART PLS
Short recap:
Basically a Structural Equation Model involves three primary components
3
Manifest variables, i.e.,
observable measures Latent constructs
Path relationship(s)
MV1
MV3
MV2
MV1
SEM family can be divided into Covariance based SEM and Partial Least
Squares
SEM allows stating theory more exactly, testing theory more precisely, and yielding a more thorough
modeling/understanding of empirical data about complex phenomena and relationships
Related concepts:
• Path analysis
• Causal Models
• Confirmatory factor analysis
• Covariance structural analysis and
• Latent variable SEM
4
Exploratory Factor
Analysis
Lawley (1940)
Factor Analysis
(PCA) through
iterated OLS
Wold (1966)
PLS-SEM through
iterated OLS
Wold (1978)
Path Analysis –
without latent
constructs - Wright
(1921)
Confirmatory Factor
Analysis
Jöreskog (1969)
LISREL-SEM
Jöreskog (1969)Model confirmation
Prediction oriented
Partial least squares is a powerful and effective means to test multivariate
structural models with latent constructs
Herman Wold („66) developed the PLS
LISREL “hard modeling” vs PLS “soft modeling”
Appealing alternative for researchers:
1. Small samples,
2. Theory in the early stages of development,
3. Data that violate traditional statistical assumptions,
4. modern easy-to-use PLS software with graphical user-interfaces.
PLS is generally recommended for predictive research models (it is a regression method that predicts
Y from X)
PLS application has encountered increased popularity across many scientific disciplines (e.g., MIS,
marketing, chemistry and chemometrics)
PLS represents a powerful and effective means to test multivariate structural models
5
When to use Partial Least Squares SEM?
6
small data
sample
non-normal
data
distribution
Formative /
reflective
measures
All ratio
measures
Emphasis on
theory
development
Many manifest
variables
Usage of Partial Least Squares (PLS) SEM
Sample characteristics
© 2010 Deloitte Touche Tohmatsu
Part 2: PACS research
MRQ: “How can hospital enterprises systematically mature their Picture Archiving and
Communication System?”
The core of stage IV is to investigate the relationship between PACS
alignment and PACS performance
8
• Workflow
assessment
‒ Holistic perspective
‒ PACS impact on
workflow
• PACS maturity
‒ Meta-analysis
‒ Maturity model
• Development of strategic
situational paths
‒ Extending Maturity model
‒ Introducing strategic
alignment
• Explain PACS
performance
‒ Development
of higher order
model
‒ Application of
SEM
Stage I
Stage II
Stage III
Stage IV
Using SEM higher order constructs can be created
• SEM can be applied to test hierarchical constructs, or multidimensional constructs
• Higher order latent constructs can be created by specifying a latent variable that represents all the
manifest variables of the underlying lower-order latent variable
• In essence, two models of higher-order constructs can be distinguished:
‒ Reflective construct model
‒ Composite model
9
S&P
M&C
O&P
IT
P&C
PACS
alignment
OP3
OP5
OP4
MV7
MV8
δ7
δ8
MV9
MV10
δ9
δ10
MV11
MV12
δ11
δ12
ζ
ζ
ζ
λ7
λ8
λ9
λ10
λ11
λ12
γ4
γ5
γ6
1st
order 2nd
order 3rd
order
“Alignment of PACS (represented by
the multi factorial nature of five
organizational domains and their
related maturity levels) has a positive
relationship on PACS performance
(represented by the multi factorial
nature in terms of hospital efficiency
and clinical effectiveness and their
related items.”
Hypothesis
This modelling of PACS alignment is statistically appropriately captured by a
pattern of co-variation
Our main is to develop an
integrative model to assess
maturity and organizational
alignment of PACS, and their
impact on PACS performance
Alignment coincides
covariation (Venkatraman,
„89) as an operationalized
statistical scheme within SEM
Covariation defined as the
pattern of internal consistency
among underlying constructs
10
Integrative PACS performance model
Alignment Performanceβ = 0.6, t = 4.01; p < 0.0001
R2= .37
Further reading on PLS SEM
Chin, W. W. (1998). Issues and Opinion on Structural Equation Modeling. MIS Quarterly, 22(1), vii-xvi.
Marcoulides, G. A., & Saunders, C. (2006). PLS: A silver bullet? MIS Quarterly 30(2), iv-viii.
Marcoulides, G. A., Chin, W. W., & Saunders, C. (2009). A critical look at partial least squares modeling.
MIS Quartery, 33(1), 171-175.
12