Brief Overview of LISREL—Page 1
Brief Overview of LISREL & Related Programs & Techniques
S
TRUCTURAL AND MEASUREMENT MODELS
:
LISREL. We have focused on structural models.
Such models assume that all variables are measured without error. Of course, this assumption is
often not reasonable. As we saw earlier in the course,
Random measurement error in the dependent variable does not bias regression coefficients.
However, it does result in larger standard errors.
Random measurement error in the independent variables results in biased estimates. In the
case of a bivariate regression, estimates will be biased toward zero. With more IVs, the bias
can be upwards or downwards.
Systematic error, of course, can produce either an upward or downward bias.
Factor analysis is one way of dealing with measurement error. With factor analysis, a large
number of items are reduced to a smaller number of factors, or ―latent variables‖. For example, 7
personality measures might be reduced into a single ―locus of control‖ scale. This scale would be
more reliable than any of the individual measures that constructed it.
Factor analysis can be either
exploratory — the computer determines what the underlying factors are
confirmatory — the researcher specifies what factor structure she thinks underlies the
measures, and then tests whether the data are consistent with her hypotheses.
Programs such as LISREL make it possible to combine structural equation modeling and
confirmatory factor analysis. (I understand programs like AMOS and MPlus and the gllamm
addon routine to Stata can do these sorts of things too but I have never used them. These
programs may be easier to use and/or cheaper than LISREL, so you may want to check them out
if you want to do heavyduty work in this area.) Some traits of LISREL:
There is both a measurement model and a structural model.
o
The measurement model indicates how observed indicators are linked to underlying
latent variables. (e.g. X1 and X2 may be indicators of Locus of control; X3 and X4 may
be indicators of Socioeconomic status).
o
The structural model indicates how the latent variables are linked to each other.
o
By controlling for measurement error, a correctly specified LISREL model makes it
possible to obtain unbiased estimates of structural coefficients. (Of course, getting the
model correctly specified is the trick.)
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentBrief Overview of LISREL—Page 2
LISREL can handle a wide array of problems and models. These include
o
Models with measurement error
o
Nonrecursive models
o
Manovatype problems
o
Multiple group comparisons (e.g. you can have separate models for blacks & whites)
o
Tests of constraints (e.g. two or more coefficients equal each other, a subset of
coefficients equals zero, parameters are equal across populations)
o
Confirmatory factor analysis models
o
Ordinal regression
o
Hierarchical Linear Models
I’ll give just a few examples, not all of which I will talk about in class. A free trial edition and a
This is the end of the preview.
Sign up
to
access the rest of the document.
 Spring '11
 RichardWilliams
 Regression Analysis, Variance, LISREL

Click to edit the document details