USING LISREL FOR STRUCTURAL EQUATION MODELING 1

USING LISREL FOR STRUCTURAL EQUATION MODELING
1. Open data file in SPSS, and generate a correlation matrix:
2. Once you have the output, cut and paste it to the Notepad utility (this will remove
all formatting, etc. from the table):
Correlations
info
info
comp
arith
simil
vocab
digit
pictcomp
parang
block
object
coding
comp
arith
pictcomp
parang
block
object
coding
Pearson Correlation
1
.467
.494
simil
.513
vocab
.625
digit
.345
.230
.202
.229
.185
.007
Sig. (2-tailed)
.
.000
.000
.000
.000
.000
.002
.007
.002
.014
.926
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.467
1
.392
.510
.531
.236
.407
.187
.369
.322
.061
Sig. (2-tailed)
.000
.
.000
.000
.000
.002
.000
.013
.000
.000
.425
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.494
.392
1
.369
.387
.269
.155
.227
.272
.043
.090
Sig. (2-tailed)
.000
.000
.
.000
.000
.000
.040
.003
.000
.573
.235
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.513
.510
.369
1
.538
.260
.369
.298
.261
.269
-.041
Sig. (2-tailed)
.000
.000
.000
.
.000
.001
.000
.000
.000
.000
.593
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.625
.531
.387
.538
1
.294
.285
.132
.297
.185
.100
Sig. (2-tailed)
.000
.000
.000
.000
.
.000
.000
.081
.000
.014
.188
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.345
.236
.269
.260
.294
1
.075
.148
.073
.035
.173
Sig. (2-tailed)
.000
.002
.000
.001
.000
.
.322
.050
.339
.648
.022
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.230
.407
.155
.369
.285
.075
1
.249
.382
.363
-.072
Sig. (2-tailed)
.002
.000
.040
.000
.000
.322
.
.001
.000
.000
.345
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.202
.187
.227
.298
.132
.148
.249
1
.351
.253
.038
Sig. (2-tailed)
.007
.013
.003
.000
.081
.050
.001
.
.000
.001
.619
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.229
.369
.272
.261
.297
.073
.382
.351
1
.399
.107
Sig. (2-tailed)
.002
.000
.000
.000
.000
.339
.000
.000
.
.000
.159
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.185
.322
.043
.269
.185
.035
.363
.253
.399
1
.053
Sig. (2-tailed)
.014
.000
.573
.000
.014
.648
.000
.001
.000
.
.486
N
175
175
175
175
175
175
175
175
175
175
175
Pearson Correlation
.007
.061
.090
-.041
.100
.173
-.072
.038
.107
.053
1
Sig. (2-tailed)
.926
.425
.235
.593
.188
.022
.345
.619
.159
.486
.
N
175
175
175
175
175
175
175
175
175
175
175
3. Delete all of the extra information from the text file, including the variable names,
so that only the correlations are left:
1 .467
.467
.494
.513
.625
.345
.230
.202
.229
.185
.007
.494
1
.392
.510
.531
.236
.407
.187
.369
.322
.061
.513
.392
1
.369
.387
.269
.155
.227
.272
.043
.090
.625
.510
.369
1
.538
.260
.369
.298
.261
.269
-.041
.345
.531
.387
.538
1
.294
.285
.132
.297
.185
.100
.230
.236
.269
.260
.294
1
.075
.148
.073
.035
.173
.202
.407
.155
.369
.285
.075
1
.249
.382
.363
-.072
.229
.187
.227
.298
.132
.148
.249
1
.351
.253
.038
.185
.369
.272
.261
.297
.073
.382
.351
1
.399
.107
.007
.322
.043
.269
.185
.035
.363
.253
.399
1
.053
.061
.090
-.041
.100
.173
-.072
.038
.107
.053
1
4. Save as a text file (.txt)
5. Open the LISREL program from the Start Menu
6. Choose “File-New”, and then select “Path Diagram” (it’s hiding at the bottom of
the list – you will have to use the scroll bars to find it):
7. Save the new path diagram (give it a file name and location)
8. Go to the “Setup” menu, and start with “Title and Comments.” At each step, hit
“Next” to go on to the next step, until you have completed all four steps:
9. For “Group Labels,” you can just hit “next” if all your variables are I/R level.
10. For “Variables,” name the variables from your original SPSS dataset. They have
to be in the same order as they appear in the correlation matrix.
11. use the “Delete” key on the keyboard (not Ctrl-X), to delete the two default
variables that LISREL puts in your list, or else rename these variables to the
names of the variables in your data file.
12. create as many latent variables as you want to include in the model, and give them
whatever names you want.
13. link the variable names to the data file with the correlation matrix:
•
•
•
“statistics from” and “matrix to be analyzed” should both be set to
“correlations” from the drop-down menus
“file type” should be set to “external ASCII data”, and then you can locate the
text file with the correlation matrix using the “browse” button.
Specify your sample size under “number of observations.” This is a key step –
if you forget it, you will get a “model does not converge” error at the end of
the process.
14. Hit “OK” to continue, and you will return to the drawing window. Draw your
model by dragging the observed and latent variables onto the graph-paper screen.
LISREL will automatically add an error term for each observed variable (unlike
AMOS, where you have to create them manually).
15. To “clean up” the drawing, select objects using the “select” box tool, and then
right-click the graph paper screen near these objects (but not on them). Use the
“align” and “even space” tools to clean up the diagram.
16. To specify a covariance between two variables (e.g., between the two latent
variables), draw the curving arrow between the error terms on those variables.
17. Under the “Setup” menu, choose “build SIMPLIS syntax.”
18. Once the syntax is generated, click the “run LISREL” button to test the model.
19. The output will show you the path diagram and the chi-square goodness-of-fit
test, as well as the RMSEA test. RMSEA of .05 or less indicates a good fit, with
values between .1 and .05 being marginal. Ideally, the chi-square statistic’s pvalue should be greater than .05 (chi-square is used here as a “badness of fit”
statistic), but with large samples this is not always going to happen. RMSEA
adjusts for the complexity of the model and the size of the sample.
20. Additional measures of goodness-of-fit can be obtained by going to “Fit Indices”
under the “Output” menu. See notes on AMOS for definitions of the various fit
indices and how to interpret them.
There is also an “output” file created by running the syntax (shrink the path
diagram to see it), which shows you the same goodness-of-fit tests and suggests
additional paths that might improve the model.
21. If you are not satisfied with the level of fit obtained, improve the model by adding
or removing variables, changing specified relationships, etc.
22. The easiest way to get the model as an image for a Word document seems to be
this: turn off the gridlines (under the “View” menu), move the toolbars out of the
way, and do a print screen (Ctrl-Print Screen key). Then go to Word and paste the
image, and crop out anything but the picture of the path diagram. [There is an
“export to metafile” option under the “File” menu that should save the path
diagram in an image format without doing these manipulations, but I haven’t been
able to get it to work].