Note: Except for starting time and latest ending
time (16:50) and a lunch break starting between 11:30 and 12:00, all times and
topics are VERY approximate. If
necessary, we may spend additional time on a topic, and alternatively if things
move more quickly than planned, we will move ahead.
You
can get a student version of LISREL from http://www.ssicentral.com
Abbreviated
course notes available at; Class1 Class2 output example
A
good overview article is available here
Monday, 17 August:
9:00 Introduction
to Course
Course Syllabus
9:30 Basics of covariance, correlation,
and matrix operations
11:30 Lunch
12:30 History
of SEM – how it developed
13:30 Path Analysis
Exercise
1 (see last page for exercises)
Reading GLM
& SEM
Tuesday, 18 August:
9:00 Discussion of Exercise #1
10:00 Models
imply covariances – direct, indirect & spurious
effects, path
Tracing
rules
12:00 Lunch
13:00 LISREL matrices and implied covariances
15:00
Comparing the observed
covariance matrix [S] with the
Model implied covariance
matrix [∑(θ)]
15:30 Exercise #2
Take a look at:
Excel example
Wednesday,
19 August:
9:00 Discussion of Exercise #2
9:30 Parameter
estimation and fit – How it works
11:30 Lunch
12:30 The
measurement model – Confirmatory Factor Analysis (CFA)
13:30 Evaluating
the measurement model
14:30 Combining
the measurement and structural models
Exercise
#3
Readings
for Thursday: Unidimensionality
(You
may find the following tough going, depending on your background, so just try
to pick up the main points. I will try
to summarize tomorrow, and then you might wish to give them another try). Meehl Latent Validity
Alpha
In
the (unlikely?) event that you want to get further into the topic of the Meehl article, look at this if you want: More
on Meehl
Thursday, 20 August:
9:00 A
philosophy for model evaluation
11:30 Lunch
strategies
Exercise #4
Readings: (Again, focus on the conclusions) Fit Indices1 Fit Indices2 Fit3 Reporting1 Reporting2
For
more on Causation, see Judea Pearls home page at: http://bayes.cs.ucla.edu/jp_home.html
Friday, 21 August:
9:00 Discussion of Exercise #4
10:00 Identification
and equivalent models
11:00 SEM and other multivariate methods
12:00 Lunch
13:00 Reporting SEM analyses
14:00 Multiple group models and measurement invariance – an overview
16:00 A
few of the things we haven’t covered – Second order factors, missing data
issues, mixture and latent class models, ordinal and categorical data analysis,
bootstrapping and Monte Carlo estimation, moderation (interaction) analysis,
latent growth modeling, exploratory SEM, etc.
Equivalent Models1 Equivalent Models 2
Here
are some additional readings you might find of interest:
What’s wrong with coefficient alpha?
Performance of estimators under
misspecification & non-normality.
Item Parceling1 Item
Parceling2
Exercise #1.
Using a data
base of your choice, find a substantive research paper in your research area
(or a related one) that both interests you and makes use of structural equation
methods. Make two copies of the paper, one for you and
one for me. The SEM model that is
considered need not be complex.
Does the SEM analysis
-makes use of latent variables?
-report fit statistics?
What software did the authors use?
In a sentence or two, write your
understanding of what the authors claim on the basis of the SEM analysis. Recognizing that you may not yet have the
technical knowledge to critique the statistical method, comment on your intution regarding the validity of the substantive claim
(i.e. would you like to believe or disbelieve the results?)
Exercise #2.
Run the following model, and show the direct
effects, the indirect effects, and total effects & where they come from.
Modified
Model for Performance and Satisfaction
References
Bagozzi, R.P. Perfomance
and satisfaction in an industrial sales force: An
examination of their antecedents and
simultaneity. Journal of Marketing, 1980,
44,
65-77
Joreskog,
K.G. and Sorbom, D. Recent
developments in structural equation
modeling. Journal of Marketing Research, 1982,
19, 404-416.
Da ni=8 no=122
La
(8A8)
performmjbsatis1jbsatis2
achmot1 achmot2t-s s-e1t-s s-e2verbintm
km
*
1.000
.418 1.000
.394 .627 1.000
.129 .202
.266 1.000
.189 .284
.208 .365 1.000
.544 .281
.324 .201 .161 1.000
.507 .225
.314 .172 .174
.546 1.000
-.357
-.156 -.038 -.199 -.277 -.294 -.174 1.000
sd
*
2.09
3.43 2.81 1.95 2.06 2.16 2.06 3.65
select
performm jbsatis1
achmot1 't-s s-e1' verbintm /
mo ny=2 nx=3 ne=2 nk=3 be=fu td=di,fi te=di,fi
ga=fu,fi lx=fu,fi ly=fu,fi
le
perform jobsatis
lk
achmot
't-s
s-e' 'verb int'
fr ga 2 1 ga 1 2 ga
1 3 be 2 1
va 1 ly 1 1 ly
2 2 lx 1 1 lx 2 2 lx 3 3;
pd
ou ef ss mi ad=off
Exercise
#3
Run
the following Model and bring the results to class. Create the km and sd files from example 2 (with appropriate paths, as
necessary). Change the input
specifications to input pattern matrices as opposed to fixing and freeing
individual paths.
Modified
Model for Performance and Satisfaction
References
Bagozzi, R.P. Perfomance
and satisfaction in an industrial sales force: An
examination of their antecedents and
simultaneity. Journal of Marketing, 1980,
44,
65-77
Joreskog,
K.G. and Sorbom, D. Recent
developments in structural equation
modeling. Journal of Marketing Research, 1982,
19, 404-416.
Da ni=8 no=122
La
(8A8)
performmjbsatis1jbsatis2
achmot1 achmot2t-s s-e1t-s s-e2verbintm
km file=EX56.DAT
sd file=EX56.DAT
mo ny=3 nx=5 ne=2 nk=3 be=fu
le
perform jobsatis
lk
achmot
't-s
s-e' 'verb int'
fr ly 3 2 lx 2 1 lx 4 2 be 2 1
fi te 1 td 5 ga 1 1 ga 2 2
ga 1 3
va 1 ly 1 1 ly
2 2 lx 1 1 lx 3 2 lx 5 3;va
1.998 td 5
pd
ou ef sc all ad=off
Exercise
4
Revisit
the article you selected in Exercise #1.
Now that you hopefully understand a little more, critically evaluate the
methodology of the article. What would you suggest to the authors? (Optional:
If your article gives the correlation or covariance matrix or a link to
the raw data, specify and run the model in the article and look for alternative
specifications).