Scott maclean advanced quant - 2011

Event
sponsored
by
Affinnova

All
copyright
owned
by
The
Future
Place
and
the
presenters
of
the
material

For
more
informa=on
about
Affinnova
visit
h>p://www.
affinnova.com/

For
more
informa=on
about
NewMR
events
visit
newmr.org

Advanced
Quant
Techniques

July
14,
2011

Structural
Equa5on
Modelling

Sco9
MacLean,
Nulink
Analy5cs

Scott MacLean, Nulink Analytics, Australia
NewMR Advanced Quant Techniques, July 14, 2011
Scott MacLean
Nulink Analytics

Structural Equation Modelling
•  On the basis of things we can measure, we
attempt to make predictions of things we cannot
measure

Start with some constructs/
factors
HAPPINESS PROSPERITY
WELLNESS

Add some links
WELLNESS

Add some indicator variables
WELLNESS
Sometimes I laugh out loud
for the joy of living
1
Stress in not a problem
for me
I take pleasure
in the small things of life
I have more than enough
money to live on
1
I don't really need to
buy anything extra
I don't really worry
about anything
1
My health has never
been an issue

Factor A
Factor B
Factor C
Factor D
Factor G
Customer
Satisfaction
Factor H
Factor F
Factor E
.93
.50
.40
.97
.97
.95 .52
.35
.02
.52
.75
Structural model example
.07

Q1a
Q1b
.93
.93
.02
Q13_1
Q13_2
Q13_3
Q13_4
.82
.93
.93
.89
.75
Q33a
Q33b
Q40
Q29
Q29a
.87
.83
.44
.49
.86
.35
.95
.52
.52
.07
Q1
Q2
Q3
Q4
Q5
.86
.78
.82
.57
Q6
.84
.70
.93
Q22_1
Q22_3
.84
.82
.50
.40
Q25_2
Q25_5
Q25_7
Q25_7
.78
.78
.79
.67
.97
Measurement model
Q18_5
Q18_6
Q18_7
.84
.68
.79
.97
Overall
Satisfaction
Factor G
Factor H
Factor A
Factor B
Factor E
Factor C
Factor D
Factor F

Goodness of Fit
Goodness of fit indices abound in covariance-based SEM.
No one fit index is best, and controversy over both which to
use, and what values indicate lack of fit, is extensive.
Four of the most common are:
–  CFI or Comparative Fit Index
–  GFI or Goodness of Fit Index
–  RMSEA or Root Mean Square Error of Approximation
–  cmin/df or minimum chi square value divided by the degrees of
freedom
These are discussed in the accompanying Glossary of
Terms.
With PLS-based SEM, goodness of fit is not such an issue

Reflective versus Formative Indicators
•  A construct should be modelled as having formative
indicators if all of a number of conditions hold, eg.:
–  The indicators are viewed as defining characteristics of the construct
–  Changes in the indicators are expected to cause changes in the
construct
–  Changes in the construct are not expected to cause changes in the
indicators
–  The indicators do not necessarily share a common theme
–  Eliminating an indicator may alter the conceptual domain of the
construct
–  A change in the value of one of the indicators is not necessarily
expected to be associated with a change in all of the other indicators

Why is this important* ?
“... strong evidence that measurement
model misspecification of even one
formatively measured construct within a
typical structural equation model can
have very serious consequences for
the theoretical conclusions drawn from
that model ”
*Jarvis, C. B., Mackenzie, Scott B. and Podsakoff, Philip M. (2003) A critical review of
construct indicators and measurement model misspecification in marketing and
consumer research Journal of Consumer Research, Vol 30 (September) 199-218.

Not only that ...
“ The entire model could appear to
adequately fit the data, even though the
structural parameter estimates within that
model exhibit very substantial biases that
would result in erroneous inferences ...”

... but also ...
“ … paths emanating from a construct
with a misspecified measurement model
are likely to be substantially inflated ...
… paths leading into a construct with a
misspecified measurement model are
likely to be deflated ”

Is it a big issue ?
•  Four journals were searched* for the 24-year
period from 1977 through 2000
•  Aim – to identify all empirical applications of
latent variable SEM or confirmatory factor
analysis
–  Journal of Consumer Research
–  Journal of Marketing
–  Journal of Marketing Research
–  Marketing Science
*Jarvis, C. B., Mackenzie, Scott B. and Podsakoff, Philip M. (2003) A critical review of
construct indicators and measurement model misspecification in marketing and
consumer research Journal of Consumer Research, Vol 30 (September) 199-218.

What did they find?
•  178 articles containing 1,192 constructs
modelled as latent factors with multiple
indicators
Should have
been modelled
as reflective
Should have
been modelled
as formative
TOTAL
Was modelled
as reflective 810 336 1,146
Was modelled
as formative 17 29 46
TOTAL 827 365 1,192

And it means ...
Over a 24 year period in academic, refereed
papers appearing in four of the world’s
leading market research / marketing science
journals, nearly 30% of modelled
constructs were wrong

Formative example
•  Customer complaints, where the indicators
might be:
–  Frequency of complaining to store manager
–  Incidence of telling friends and relatives about a bad service
experience
–  Likelihood of reporting the supplier to a consumer complaints
agency
–  Likelihood of pursuing legal action against the supplier
–  etc.
•  A high score on one item would certainly
influence the level of the latent construct, but
would not necessarily have an effect on the
other items

Analysis: formative vs reflective
•  It can be difficult to analyse models that
contain formative factors, unless the PLS-
based approach to SEM is utilised
•  The covariance-based can be used, but
only under very specific circumstances*
* See http://nulinkanalytics.com.au/D2-S2-TM1-03%20Scott%20MacLean.pdf
for further details

Software suggestions
•  AMOS
– Available from
http://www-01.ibm.com/software/analytics/
spss/products/statistics/amos/
•  SmartPLS
– Beta release available from www.smartpls.de

Further reading suggestions
•  http://nulinkanalytics.com.au/D2-S2-TM1-03%20Scott
%20MacLean.pdf
•  http://www.smallwaters.com/whitepapers/marketing/
•  “Structural Equation Modelling in Marketing – Parts I, II
and III” in
Dictionary of Quantitative Management Research; Sage
Publications Ltd; ISBN: 9781412935296

Training suggestions
•  A 3-day Course on PLS Path Modelling and XLSTAT-
PLSPM software: Basic Concepts and Foundations,
Advances and Applications by W.W. Chin, V. Esposito
Vinzi, M. Tenenhaus 26-27-28 September 2011, Paris,
France http://www.xlstat.com/en/training/ &
http://www.xlstat.com/file/PLSPMCourse.pdf
•  Online R and PLS courses for Aug 2011 through June
2012, https://www.regonline.com/2-for-1

Thank you
Scott MacLean
Nulink Analytics

Q & A
Sue
York

The
Future
Place

Sco>
MacLean

Nulink
Analy=cs

Scott MacLean
Sco>
MacLean

Director,
Nulink
Analy=cs

T:
+61
419
504
588

E:
sco>@nulinkanaly=cs.com

W:
www.nulinkanaly=cs.com

Glossary of terms*
Indicators are observed variables, sometimes called manifest variables or
reference variables, such as items in a survey instrument.
Latent variables are the unobserved variables or constructs or factors which
are measured by their respective indicators.
The measurement model is that part (possibly all) of a SEM model which deals
with the latent variables and their indicators.
Exogenous variables are independents with no prior causal variable (though
they may be correlated with other exogenous variables, depicted by a double-
headed arrow -- note two latent variables can be connected by a double-headed
arrow (correlation) or a single-headed arrow (causation) but not both.
Endogenous variables are mediating variables (variables which are both effects
of other exogenous or mediating variables, and are causes of other mediating
and dependent variables), and pure dependent variables.
* Source: Garson, G. David (2007). "Structural Equation Modeling", in Statnotes: Topics in Multivariate Analysis,
retrieved 21 August 2007 from http://www2.chass.ncsu.edu/garson/pa765/sructur.htm

Glossary of terms*
Path coefficients are the effect sizes calculated by the model estimation
program. Often these values are displayed above their respective arrows on the
arrow diagram specifying a model. In AMOS, these are labeled "regression
weights," which is what they are, except that in the structural equation there will
be no intercept term (unlike conventional regression, where this is normally
specifically provided for).
Standardised (path) coefficients measure the extent to which a change of one
standard deviation in the independent or predictor variable is reflected by a
change in the dependent variable, also measured in units of its standard
deviation.
Residual terms essentially measure the difference between the predicted and
estimated actual values for the latent variables
Error terms essentially measure the difference between the observed and
predicted values for the indicator variables

Glossary of terms* (cont.)
Goodness of fit indices abound in Structural Equation Modelling. No one fit
index is best, and controversy over both which to use, and what values indicate
lack of fit, is extensive.
Four of the most common are:
–  CFI or Comparative Fit Index
–  GFI or Goodness of Fit Index
–  RMSEA or Root Mean Square Error of Approximation
–  cmin/df or minimum chi square value divided by the degrees of freedom
These are briefly discussed on the following pages.
Degrees of freedom (df) is a measure of the number of independent pieces of
information on which the precision of a parameter estimate is based. The
degrees of freedom for an estimate equals the number of observations (values)
minus the number of additional parameters estimated for that calculation. As we
have to estimate more parameters, the degrees of freedom available decreases.

CFI compares the existing model fit with a null model which assumes the latent
variables in the model are uncorrelated (the "independence model").
That is, it compares the covariance matrix predicted by the model to the
observed covariance matrix, and compares the null model (covariance matrix of
0's) with the observed covariance matrix, to gauge the percent of lack of fit
which is accounted for by going from the null model to the researcher's SEM
model.
CFI is among the measures least affected by sample size.
CFI varies from 0 to 1 (if outside this range it is reset to 0 or 1). CFI close to 1
indicates a very good fit.
By convention, CFI should be equal to or greater than (about) .90 to accept the
model, indicating that 90% of the covariation in the data can be reproduced by
the given model.

GFI varies from 0 to 1 but theoretically can yield meaningless negative values.
A large sample size pushes GFI up. Though analogies are made to R-square,
GFI cannot be interpreted as percent of error explained by the model. Rather it
is the percent of observed covariances explained by the covariances implied by
the model.
That is, R-square in multiple regression deals with error variance whereas GFI
deals with error in reproducing the variance-covariance matrix.
As GFI often runs high compared to other fit measures, some suggest using .95
as the cutoff.
By convention, GFI should by equal to or greater than (about) .90 to accept the
model.

RMSEA - by convention, there is good model fit if RMSEA less than or equal to .
05. There is adequate fit if RMSEA is less than or equal to (about) .08.
It is one of the fit indexes less affected by sample size, though for smallest
sample sizes it overestimates goodness of fit.
It may be said that RMSEA corrects for model complexity, as shown by the fact
that df is in its denominator. However, degrees of freedom is an imperfect
measure of model complexity. Since RMSEA computes average lack of fit per
degree of freedom, one could have near-zero lack of fit in both a complex and in
a simple model and RMSEA would compute to be near zero in both, yet most
methodologists would judge the simpler model to be better (in other words,
there is no compelling reason to accept a more complex explanation when a
simpler explanation achieves the same outcome).

cmin - the model chi-square, also called discrepancy or the discrepancy
function, is the most common fit test, printed by all computer programs. The chi-
square value should not be significant if there is a good model fit, while a
significant chi-square indicates lack of satisfactory model fit, ie. the given
model's covariance structure is significantly different from the observed
covariance matrix.
There are three ways, listed below, in which the chi-square test may be
misleading. Because of these reasons, many researchers who use SEM believe
that with a reasonable sample size (eg. n > 200) and good approximate fit as
indicated by other fit tests (eg. CFI, RMSEA), the significance of the chi-square
test may be discounted and that a significant chi-square is not a reason by itself
to modify the model.
–  The more complex the model, the more likely a good fit.
–  The larger the sample size, the more likely the rejection of the model
–  The chi-square fit index is also very sensitive to violations of the assumption of
multivariate normality.

cmin/df – is the chi-square fit index divided by degrees of freedom, in an
attempt to make it less dependent on sample size.
Some authors state that relative chi-square should be in the 2:1 or 3:1 range for
an acceptable model.
Still others would argue that 3 or less is acceptable, and some allow values as
high as 5 to consider a model adequate fit, while others insist relative chi-square
be 2 or less.
Because of the inadequacies of cmin as a measure of goodness of fit (see
previous page) we would mostly rely on the other fit measures for assessing the
appropriateness (or, more accurately, lack of non-appropriateness) of the model.

Collinearity – Collinearity (or multicollinearity – the two terms are
interchangeable) occurs because two (or more) variables are related – they
essentially measure the same thing.
Why is multicollinearity a problem? If the goal is simply to predict Y from a set
of X variables, then multicollinearity is not a problem. The predictions will still be
accurate, and the overall r-square quantifies how well the model predicts the Y
values.
However, if the goal is to understand how the various X variables impact Y, then
multicollinearity is a big problem. One problem is that the regression
coefficients can be indicated as statistically insignificant, even though the
variables concerned are important. The second problem is that the confidence
intervals on the regression coefficients will be very wide. The confidence
intervals may even include zero, which means we can’t even be confident
whether an increase in the X value is associated with an increase, or a
decrease, in Y. Because the confidence intervals are so wide, excluding a
subject (or adding a new one) can change the coefficients dramatically – and
may even change their signs.
* Source: http://www.graphpad.com/articles/Multicollinearity.htm

(Exploratory) Factor Analysis – Factor Analysis is used to uncover the latent
structure (dimensions) of a set of variables. It reduces attribute space from a
larger number of variables to a smaller number of factors and as such is a "non-
dependent" procedure (that is, it does not assume a dependent variable is
specified).
Factor Analysis can be used for any of a number of purposes, eg:
–  To reduce a large number of variables to a smaller number of factors for
modeling purposes, where the large number of variables precludes modeling
all the measures individually
–  To select a subset of variables from a larger set, based on which original
variables have the highest correlations with the principal component factors
–  To create a set of factors to be treated as uncorrelated variables as one
approach to handling multicollinearity in such procedures as multiple
regression
* Source: Garson, G. David (2007). “Factor Analysis", in Statnotes: Topics in Multivariate Analysis,
retrieved 18 September 2007 from http://www2.chass.ncsu.edu/garson/pa765/factor.htm

Scott maclean advanced quant - 2011

More Related Content

Viewers also liked (10)

Similar to Scott maclean advanced quant - 2011 (20)

More from Ray Poynter (20)

Scott maclean advanced quant - 2011