WHAT IS SEM?
Mplus, EQS, Amos, Calis, Mx, SEPATH,
Tetrad, R, stata
SEM is not one statistical ‘technique’
Integrates a number of different
What is SEM?
multivariate techniques into one model
SEM can be thought of as path analysis for
fitting framework
latent variables.
Useful for Research Questions that….
What are Latent Variables?
Involve complex, multi-faceted constructs
Most social scientific concepts are not
that are measured with error, SEM can
directly observable, e.g. intelligence, social
make corrections in the errors of
capital
measurement
Hypothetical or ‘latent’ constructs
That specify ‘system’ of relationships rather
We can measure latent variables using
than a dependent variable and a set of
observable indicators
predictors
When we ask people “how happy they are
Focus on indirect (mediated) as well as
with their lives”, there will be variability.
direct effects of variables on other variables
True score and measurement error
SEM is also known as
X=t+e
Covariance Structure Analysis
Analysis of Moment Structures
Analysis of Linear Structural Relationships
(LISREL)
Causal Modeling X being measured
T being true score; true value on construct
Software for SEM (their true level of happiness
Best known is Lisrel, developed by Joreskog E being error; it can be a systematic error (mean
and Sorbom of the individual errors doesn’t equal error)
and random error (where the mean of the
errors will be equal to zero)
The problem with one indicator, the equation is Endogenous Variable
unidentified, we can’t solve the equation - Caused by variables in the system
uniquely. We need to have multiple Exogenous Variable
indicators. - Caused by variables outside the system
Benefits of Latent Variables INTERPRETING RESULTS
Most social concepts are complex and multi-
faceted
Using single measures will not adequately
cover the full conceptual map
Removes/reduces random error in
measured construct
Random errors with dependent variables ->
estimates unbiased but less precise
Random errors with independent variables Normed Chi-square = X2 * df < 2 or 3
-> attenuates regression interval toward Normed Fit Index (NFI) exceeds .90
zero Goodness of Fit (GFI) Index exceeds .90
Comparative Fit Index (CFI) exceeds .93
Remember: SEM can be thought of as path
analysis for latent variables.
Advantages of SEM over Conventional Analysis
So, what is PATH ANALYSIS? 1. Greater flexibility regarding
The diagrammatic representation of a assumptions; one can still interpret
theoretical model using standardized results even in the face of
notation multicollinearity
Presents regression equation between 2. Allows use of confirmatory factor
measured or observable variables analysis to reduce measurement error
Focused on direct, indirect, and total effects by testing multiple superior model
visualization through its graphical
Path Diagram Notaton modeling interface.
3. Tests models overall rather than
coefficients individually
4. Has the ability to tests models with
multiple dependent variables to include
mediating variables and to model error
terms for all indicator variables.
5. Considers potential errors of
measurement in all variables and when
a hypothesized structural model shows
model fit indices that are less than
satisfactory, it allows specifications
searches to find better fitting models to
the sample variance-covariance matrix.
Developing a Structural Equation Model
1. Model Specification – This is the
process of formally stating a model by
determining which parameters are to be
fixed or free.
2. Model Identification- This is the idea of
having at least one unique solution for
each parameter estimate in the model
from the observed data.
3. Model Estimation- This is the process in
which start values of the free
parameters are chosen in order to
generate an estimated population
covariance matrix
4. Testing Model Fit – This is the process
of evaluating a structural equation
model with goodness-of-fit indices.
