Multidimensionality and HigherOrder Factor Models, Part 2 - PowerPoint PPT Presentation

Loading...

PPT – Multidimensionality and HigherOrder Factor Models, Part 2 PowerPoint presentation | free to download - id: 1ff118-ZDc1Z



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Multidimensionality and HigherOrder Factor Models, Part 2

Description:

Nicole Ponder. MKT 8543. Quantitative Marketing Seminar. April 19, 2005 ... More distinctions between formative and reflective...Figure 1, page 201 ... – PowerPoint PPT presentation

Number of Views:66
Avg rating:3.0/5.0
Slides: 21
Provided by: nicoleh4
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Multidimensionality and HigherOrder Factor Models, Part 2


1
Multidimensionality and Higher-Order Factor
Models, Part 2
MKT 8543 Quantitative Marketing Seminar
  • April 19, 2005

Mississippi State University
Nicole Ponder
2
Measurement Scales versus Indices
  • Measurement scale consists of effect
    indicators whose values are caused by an
    underlying construct (Bollen 1989)
  • The reflective model (Bollen and Lennox 1991)

var(x1) ?2 var(?1) var(?1) The only thing
that the 4 indicators should have in common is
the latent construct!
?1
x1
x2
x3
x4
?1
?2
?3
?4
3
Measurement Scales versus Indices
  • Index consists of indicators that, taken
    together, cause the underlying construct
    (Diamantopoulos and Winklhofer 2001)
  • The formative model (Bollen and Lennox 1991)

?1 ?1 x1 ?2 x2 ?3 x3 ?4 x4
?1
Often, researchers mistakenly use this model to
run with SEMand problems occur!
x1
x2
x3
x4
?1
?2
?3
?4
4
Jarvis, Mackenzie, and Podsakoff (2004)
  • Question of interest How do you choose between
    using reflective indicators and using formative
    indicators in your own research?
  • Simple way by using the definitional essay!
    Can you craft an item that captures the entirety
    of the overall definition? Or do your items
    capture the different components/dimensions of
    the overall concept?
  • Other guidelines, p. 203
  • More distinctions between formative and
    reflectiveFigure 1, page 201
  • Many constructs in marketing have been treated as
    reflective when they really should be treated as
    formative (p. 208-209)
  • Many seminal articles! This could change
    structural relationships and conclusions!

5
Jarvis, Mackenzie, and Podsakoff (2004)
  • Now, what happens when we move from first-order
    constructs to higher-order constructs?
  • If the distinction between formative and
    reflective indicators is not clear, it only gets
    worse when moving to multidimensional constructs!
  • Figure 2, page 205Four different ways of looking
    at second-order factor structures

6
Bagozzi and Edwards (1998)
  • In any empirical study, it is essential to be
    specific as to the depth and dimensionality of
    constructs and their measures if meaningful
    results are to be obtained
  • Interested in studying the different levels of
    abstraction of a multidimensional construct
  • Used SEM to test four different levels of
    aggregation of the Work Aspect Preference Scale
    (WAPS)
  • Concluded that work values can only be studied at
    its lowest levels of abstraction i.e., overall
    work values cannot be properly analyzed

7
Levels of abstraction in the Work Aspect
Preference Scale (WAPS)
Work Values
Non-work Orientation
Human/ personal concern
Freedom
Money
Life style
Detachment
Creativity
Independence
Self-develop
Co-workers
Security
8
BEs Aggregation Method
  • Total disaggregation model tests the
    subcomponents of work values at their lowest
    level of abstraction
  • n437
  • the ?x parameter estimates are all statistically
    significant, the fit indices show strong model
    fit, and the modification indices for ?x and ??
    are all nonsignificant
  • Thus, at this lowest level of abstraction, each
    indicator can be seen as truly reflective of the
    construct it is supposed to represent.

Det
Mon
Lif
Cre
Ind
Se Dev
Cow
Sec
1
2
3
4
5
6
7
8
a
b
c
d
e
f
g
h
i
j
k
l
m
n
o
p
q
r
s
t
u
v
w
x
9
BEs Aggregation Method
  • Partial disaggregation model
  • Use of parcels reduces number of parameters to
    be estimated
  • Overall fit does improve over the total
    disaggregation model

Det
Mon
Lif
Cre
Ind
Se Dev
Cow
Sec
12
34
56
78
ab
cd
ef
gh
qr
st
ij
kl
mn
op
uv
wx
10
Quick notes on the use of parcelsfrom Little et
al. (2002)
  • Pros
  • Individual items are statistically less reliable
    than aggregate scores
  • Overall levels of specific and random error are
    reduced
  • Overall fit statistics provide evidence of a
    better fit of the model to the data
  • Parsimony!
  • Cons
  • What happens if the construct is
    multidimensional? And esp. if the dimensions are
    not related?
  • Parcels can mask true problems that exist with
    the measurement model
  • Researchers could play with parcels to get the
    best model fit

11
BEs Aggregation Method
  • Partial aggregation model

Human/ personal concern
Nonwork orientation
Freedom
det
sde
mon
lif
cow
sec
cre
ind
?2 (17, n437) 150.52, p 0.00 CFI .75 BE
concluded that one must reject the partial
aggregation model based on the goodness of fit
indices.
12
Our Proposed Aggregation Method
  • Creation of reflective combinations to use in the
    partial aggregation model

Det
Mon
Lif
Cre
Ind
Se Dev
Cow
Sec
1
2
3
4
5
6
7
8
a
b
c
d
e
f
g
h
i
j
k
l
m
n
o
p
q
r
s
t
u
v
w
x
Reflective measures may be created for nonwork
orientation, freedom, and human/personal concern
as followsNWO1 1 5 aNWO2 2 6
bNWO3 3 7 cNWO4 4 8 d
FR1 e i FR2 f j FR3 g k FR4 h l
HPC1 m q u HPC2 n r v HPC3 o s
w HPC4 p t x
13
Our Method
  • Full variance-covariance matrix provided by BE
  • Used a FORTRAN-based multivariate normal data
    generator to generate 1000 individual
    observations that will reproduce the given matrix
  • Able to replicate results that BE got for their
    models
  • Used SEM to re-analyze partial and total
    aggregation models using our method

14
Re-analyzed Partial Aggregation Model
Freedom
Non-work Orientation
Human/personal Concern
NWO1
NWO2
NWO3
NWO4
FR1
FR2
FR3
FR4
HPC1
HPC2
HPC3
HPC4
?2 (51, n1000) 320.08, p 0.00 CFI 0.96
GFI 0.95 AGFI 0.93 Here, SMCs are high MIs
are low, no cross-loadings.
15
BEs Aggregation Method
  • Total aggregation model

Work Values
Non-work Orientation
Human/personal Concern
Freedom
?2 (2, n437) 66.78, p 0.00 CFI .62 The
three factor loadings were constrained to be
equal. BE concluded that the fit of the model
was poor therefore, one can only study work
values at its facet levels.
16
Re-analyzed Total Aggregation Model
  • Can use reflective combinations in the total
    aggregation model as well

Freedom
Non-work Orientation
Human/personal Concern
NWO1
NWO2
NWO3
NWO4
FR1
FR2
FR3
FR4
HPC1
HPC2
HPC3
HPC4
Reflective measures may be created for work
values as follows WV1 NWO1 FR1 HPC1WV2
NWO2 FR2 HPC2WV3 NWO3 FR3 HPC3WV4
NWO4 FR4 HPC4
17
Re-analyzed Total Aggregation Model
Work Values
WV1
WV2
WV3
WV4
?2 (5, n1000) 8.60, p 0.13 CFI .99 GFI
.99 AGFI .99 The three factor loadings were
constrained to be equal. If measured properly,
work values can be studied at a global level.
18
Comments on Our Aggregation Results
  • Much better fit, now measures are properly
    reflective
  • In order for reflective combinations to be proper
    indicators, it is mandatory that the total
    disaggregation model displays properties of
    excellent model fit
  • parameter estimates for ?x must be large and
    statistically significant (0.70 or higher if phi
    is standardized)
  • SMCs for each indicator must be large (well above
    0.50, preferably 0.70 or higher)
  • modification indices for ?x must be statistically
    non-significant (values lt 3.84)
  • modification indices for ?? must be statistically
    non-significant (values lt 3.84)

19
Comments/Guidelines for Aggregation
  • Need clean results at the total disaggregation
    level
  • Take time to develop proper conceptual
    definitions of constructs
  • Pay attention to the assumption of reflective
    measures
  • How do you know which indicators to combine?
  • To create reflective indicators of NWO, the
    combination of 1, 5, and a is arbitrary
  • Just ensure each dimension is represented!

NWO1 1 5 aNWO2 2 6 bNWO3 3 7
cNWO4 4 8 d
Money
Detach
Lifestyle
1
2
3
4
5
6
7
8
a
b
c
d
20
Conclusions
  • Better alternative than HOF models
  • Forces the researcher to place importance on the
    development of conceptual definitions, and to
    get it right at the total disaggregation level!
  • Reflective combinations approach may be applied
    to other multidimensional constructs
  • Trust reliability, integrity, and confidence
  • Communication informing, answering, listening
  • Service quality tangibles, reliability,
    responsiveness, assurance,empathy
  • Market orientation customer orientation,
    competitor orientation, interfunctional
    coordination
About PowerShow.com