04 Structual Equation Modeling with Latent Variables 2006
Structural Equation Modeling
(SEM) With Latent Variables
James G. Anderson, Ph.D.
Purdue University
Steps In
Structural Equation Modeling
1.
2.
3.
4.
5.
6.
Data preparation
Model specification
Identification
Estimation
Testing fit
Respecification
Data Preparation
• Estimation of missing data
• Creation of scales and indices
• Descriptive statistics to include
– Examination for outliers
– skewness and kurtosis
• Transformation of variables
Measurement Model (1)
•
•
Specifying the relationship between the latent
variables and the observed variables
Answers these questions:
1) To what extent are the observed variables actually
measuring the hypothesized latent variables?
2) Which observed variable is the best measure of a
particular latent variable?
3) To what extent are the observed variables actually
measuring something other than the hypothesized
latent variable?
Measurement Model (2)
• The relationships between the observed variables and
the latent variables are described by factor loadings
• Factor loadings provide information about the extent
to which a given observed variable is able to measure
the latent variable. They serve as validity coefficients.
• Measurement error is defined as that portion of an
observed variable that is measuring something other
than what the latent variable is hypothesized to
measure. It serves as a measure of reliability.
Measurement Model (3)
• Measurement error could be the result of:
– An unobserved variable that is measuring some
other latent variable
– Unreliability
– A secondorder factor
A Latent Variable with a Single Indicator
SENTENCE (LV)
1
Sentence Test
Score (OV)
1
Measurement
Error
Setting the Error Variance
• Error variance can be set to 0 if you have a
single indicator of the latent variable and no
information about its reliability
• Error Variance = (1Reliability) Variance of
the Observed Score if you know the
reliability of the indicator
Creating a Latent Variable from
Multiple Indicators
• Exploratory factor analysis can be used with
multiple indicators of a construct to
determine the number of factors and which
indicators are associated with each factor.
• Confirmatory factor analysis can then be
used to test the fit of the measurement
model.
Latent Vartiable with Multiple Indicators
1
VISPERC
CUBES
Spatial
LOZENGES
1
Verbal
1
1
WORDMEAN
e2
1
e3
1
PARAGRAPH
SENTENCE
e1
1
1
e4
e5
e6
Example of a Complete
Structural Equation Model
• We can specify a model to further discuss how to
diagram a model, specify the equations related to
the model and discuss the “effects” apparent in the
model. The example we use is a model of
educational achievement and aspirations.
• Figure 3 shows there are four latent variables
(depicted by ellipses), three exogenous variables
(knowledge, Value and Satisfaction) and one
endogenous (performance).
Variables
•
•
Performance – Planning,
Organization, controlling,
coordinating and directing a
Example 5: SEM with Latent Variables
farm cooperative
Knowledge – Knowledge of e3 1knowledge knowledge
economic phases of
e4
2knowledge
management directed
toward profitmaking
e5
1value
value
e9
performance
ValueTendency to
2value
e6
rationally evaluate means to
2performance 1performance
an economic end
e7 1satisfaction
e1
e2
Satisfaction Gratification
satisfaction
2satisfaction
e8
from performing the
managerial role
1
1
1
1
•
1
1
1
1
1
•
1
1
1
1
Structural Model (1)
• The researcher specifies the structural model to
allow for certain relationships among the latent
variables depicted by lines or arrows
• In the path diagram, we specified that
Performance is related to Knowledge, Value and
Satisfaction in a specific way. Thus, one result
from the structural model is an indication of the
extent to which these a priori hypothesized
relationships are supported by our sample data.
Structural Model (2)
• The structural equation addresses the
following questions:
– Is Performance related to the three predictor
variables?
– Exactly how strong is the influence of each
variable on Performance?
– How well does the model fit the data?
Example of a Complete
Structural Equation Model (2)
• Each of the four latent variables is assessed
by two indicator variables. The indicator
variables are depicted in rectangles.
Example 5: SEM with Latent Variables
e3
1
1knowledge
knowledge
1
e4
e5
e6
1
1
1
2knowledge
1value
1
value
performance
1
e9
1
2value
2performance 1performance
1
e7 1satisfaction
1
1
e8 2satisfaction
satisfaction
1
1
e2
e1
Example 5: SEM with Latent Variables
Measurement Moiel
e3
1
1knowledge
knowledge
1
e4
e5
e6
1
1
1
2knowledge
1value
1
value
performance
1
e9
1
2value
2performance 1performance
1
e7 1satisfaction
1
1
e8 2satisfaction
satisfaction
1
1
e2
e1
Example 5 SEM with Latent Variables
Structural Model
knowledge
value
satisfaction
performance
1
e9
Model Building
• If the original model does not provide an
acceptable fit to the data, alternative models
can be tested.
• The standardized residuals and modification
indices can be used to determine how to
modify the model to achieve a better fit to
the data.
Covariance
• SEM involves the decomposition of
covariances
• There are different types of
covariance matrices:
1)
2)
3)
4)
Among the observed variables
Among the latent exogenous variables.
Among the equation prediction errors
Among the measurement errors
Covariance (2)
•
Types of covariance
1) Among the observed variables
2) Among the latent exogenous variables
IQ
ACH
HOME
Covariance (3)
3) Among the equation prediction errors
Religion
E1
Legal
V1
E3
E2
F1
Profess
Experience
V2
E4
Error
F2
Error
Total, Direct and Indirect Effects
• There is a direct effect between two latent variables
when a single directed line or arrow connects them
• There is an indirect effect between two variables
when the second latent variable is connected to the
first latent variable through one or more other
latent variables
• The total effect between two latent variables is the
sum of any direct effect and all indirect effects that
connect them.
(SEM) With Latent Variables
James G. Anderson, Ph.D.
Purdue University
Steps In
Structural Equation Modeling
1.
2.
3.
4.
5.
6.
Data preparation
Model specification
Identification
Estimation
Testing fit
Respecification
Data Preparation
• Estimation of missing data
• Creation of scales and indices
• Descriptive statistics to include
– Examination for outliers
– skewness and kurtosis
• Transformation of variables
Measurement Model (1)
•
•
Specifying the relationship between the latent
variables and the observed variables
Answers these questions:
1) To what extent are the observed variables actually
measuring the hypothesized latent variables?
2) Which observed variable is the best measure of a
particular latent variable?
3) To what extent are the observed variables actually
measuring something other than the hypothesized
latent variable?
Measurement Model (2)
• The relationships between the observed variables and
the latent variables are described by factor loadings
• Factor loadings provide information about the extent
to which a given observed variable is able to measure
the latent variable. They serve as validity coefficients.
• Measurement error is defined as that portion of an
observed variable that is measuring something other
than what the latent variable is hypothesized to
measure. It serves as a measure of reliability.
Measurement Model (3)
• Measurement error could be the result of:
– An unobserved variable that is measuring some
other latent variable
– Unreliability
– A secondorder factor
A Latent Variable with a Single Indicator
SENTENCE (LV)
1
Sentence Test
Score (OV)
1
Measurement
Error
Setting the Error Variance
• Error variance can be set to 0 if you have a
single indicator of the latent variable and no
information about its reliability
• Error Variance = (1Reliability) Variance of
the Observed Score if you know the
reliability of the indicator
Creating a Latent Variable from
Multiple Indicators
• Exploratory factor analysis can be used with
multiple indicators of a construct to
determine the number of factors and which
indicators are associated with each factor.
• Confirmatory factor analysis can then be
used to test the fit of the measurement
model.
Latent Vartiable with Multiple Indicators
1
VISPERC
CUBES
Spatial
LOZENGES
1
Verbal
1
1
WORDMEAN
e2
1
e3
1
PARAGRAPH
SENTENCE
e1
1
1
e4
e5
e6
Example of a Complete
Structural Equation Model
• We can specify a model to further discuss how to
diagram a model, specify the equations related to
the model and discuss the “effects” apparent in the
model. The example we use is a model of
educational achievement and aspirations.
• Figure 3 shows there are four latent variables
(depicted by ellipses), three exogenous variables
(knowledge, Value and Satisfaction) and one
endogenous (performance).
Variables
•
•
Performance – Planning,
Organization, controlling,
coordinating and directing a
Example 5: SEM with Latent Variables
farm cooperative
Knowledge – Knowledge of e3 1knowledge knowledge
economic phases of
e4
2knowledge
management directed
toward profitmaking
e5
1value
value
e9
performance
ValueTendency to
2value
e6
rationally evaluate means to
2performance 1performance
an economic end
e7 1satisfaction
e1
e2
Satisfaction Gratification
satisfaction
2satisfaction
e8
from performing the
managerial role
1
1
1
1
•
1
1
1
1
1
•
1
1
1
1
Structural Model (1)
• The researcher specifies the structural model to
allow for certain relationships among the latent
variables depicted by lines or arrows
• In the path diagram, we specified that
Performance is related to Knowledge, Value and
Satisfaction in a specific way. Thus, one result
from the structural model is an indication of the
extent to which these a priori hypothesized
relationships are supported by our sample data.
Structural Model (2)
• The structural equation addresses the
following questions:
– Is Performance related to the three predictor
variables?
– Exactly how strong is the influence of each
variable on Performance?
– How well does the model fit the data?
Example of a Complete
Structural Equation Model (2)
• Each of the four latent variables is assessed
by two indicator variables. The indicator
variables are depicted in rectangles.
Example 5: SEM with Latent Variables
e3
1
1knowledge
knowledge
1
e4
e5
e6
1
1
1
2knowledge
1value
1
value
performance
1
e9
1
2value
2performance 1performance
1
e7 1satisfaction
1
1
e8 2satisfaction
satisfaction
1
1
e2
e1
Example 5: SEM with Latent Variables
Measurement Moiel
e3
1
1knowledge
knowledge
1
e4
e5
e6
1
1
1
2knowledge
1value
1
value
performance
1
e9
1
2value
2performance 1performance
1
e7 1satisfaction
1
1
e8 2satisfaction
satisfaction
1
1
e2
e1
Example 5 SEM with Latent Variables
Structural Model
knowledge
value
satisfaction
performance
1
e9
Model Building
• If the original model does not provide an
acceptable fit to the data, alternative models
can be tested.
• The standardized residuals and modification
indices can be used to determine how to
modify the model to achieve a better fit to
the data.
Covariance
• SEM involves the decomposition of
covariances
• There are different types of
covariance matrices:
1)
2)
3)
4)
Among the observed variables
Among the latent exogenous variables.
Among the equation prediction errors
Among the measurement errors
Covariance (2)
•
Types of covariance
1) Among the observed variables
2) Among the latent exogenous variables
IQ
ACH
HOME
Covariance (3)
3) Among the equation prediction errors
Religion
E1
Legal
V1
E3
E2
F1
Profess
Experience
V2
E4
Error
F2
Error
Total, Direct and Indirect Effects
• There is a direct effect between two latent variables
when a single directed line or arrow connects them
• There is an indirect effect between two variables
when the second latent variable is connected to the
first latent variable through one or more other
latent variables
• The total effect between two latent variables is the
sum of any direct effect and all indirect effects that
connect them.