Documente Academic
Documente Profesional
Documente Cultură
Krissy Zeiser
Pennsylvania State University
klz124@pop.psu.edu
12-1pm
11/13/2008
Setting up the Dataset
• Missing values should be recoded in another program (preferably with SPSS since
AMOS uses SPSS datasets)
• Dummy variables and recoding of ordinal/continuous variables should be done
before working in AMOS
• Along with SPSS datasets (.sav), AMOS also reads .dbf, .xls, .txt, and .csv files
Missing Data
• AMOS prefers complete datasets, though it can perform FIML
• It is common to replace missing values with the variable’s mean or mode
• May perform listwise deletion first if you know which variables you will be using
Benefits of AMOS
• You can specify when variables or error terms should be correlated
• You can set error terms equal to each other
• You can assign values to correlations, factor loadings, etc.
• Easy to conceptualize models and provide illustrations to you structural equation
model
• Compares different models that use different constraints (nested models)
• Allows you to perform the same model on different subgroups of the sample
• Can perform latent growth curve modeling
• Can use visual basic code to run AMOS
Drawbacks to AMOS
• Must have a continuous (or ordinal) dependent variable
• Normally doesn’t use syntax-though the model is saved if you need to replicate
the results
• Amos does not allow you to weight your data to correct for sample design and
attrition
Today’s Example
• Today, we will be using variables from the National Education Longitudinal
Study of 1988 (NELS88)
• I will be demonstrating SEM using FIML, so missing values have already been
recoded into “.”
For more information, the .pdf file of the Amos manual can be found at:
http://www.psy.umassd.edu/Psy506/help_documents/AMOS_SEM%20manual.pdf
Analyzing in AMOS
Simple and Multiple Linear Regression
Example 1: two observed independent variables (math and reading test scores)
predicting an observed dependent variable (educational expectations)
• You can use the “copy machine” to copy, drag, and paste your boxes
rather than create new ones
• Right-click the error term and click “Object properties” to name it so that AMOS
knows how to identify the error term in the output. When you’re done typing,
exit the window by clicking the red “x” in the upper right-hand corner.
o When you open a window to change preferences or label variables, Amos
will remember your changes when you close the window. There is not an
option to “save changes.”
• After drawing you structural equation model, you must select which
dataset you will use
• Next, you must open the dataset and drag the variables into the
appropriate boxes
e1
1
reading_score
educ_ex
math_score
• If you want to see standardized regression coefficients, under the “output” tab of
analysis properties, select the box next to “standardized estimates.” When you
close the menu, what you have selected will be remembered
• You’re finally ready to run your model
• The output will show you the estimates as well as the model fit statistics.
This output will be saved as a separate file in the same location as your “input”
file .
• You can copy your model to a word document by clicking the button with
the clipboard and pasting it right into your document
Multivariate Linear Regression
Example 2: four observed independent variables (racial categories for Asian, Hispanic
black, and Indian) predicting two observed dependent variables (educational
expectations and reading test scores)
• To make the covariances look “neater,” you can use the “moving truck”
to move the arrows or the variables themselves
0,
e1
1
asian
educ_ex
hispanic
reading_score
1
0,
black
e2
indian
Linear Regression with Latent Variables
Example 3: three observed independent variables (family structure variables for two
parents (non-biological, single parent, and “other”) predicting an (unobserved) latent
dependent variable (emotional support) composed of five observed variables
• Use the red X to delete variables you no longer need-just click on the icon
and then click on the variable(s) you wish to delete
• To make a latent construct click on the latent construct icon. Then, click
on your latent variable oval once for every observed variable that comprises the
latent construct.
o Notice that error terms are automatically attached to these boxes and that
one of the factor loadings is already constrained to equal 1
o You have to create an error term for the latent oval as well
o Right-click the oval and select “object properties” to name your latent
construct. Don’t forget to name all of the error terms, too.
• AMOS only conducts confirmatory factor analysis and provides factor loadings.
To perform exploratory factor analysis (to determine whether these variables
should indeed be combined to make a scale or latent construct), you should use
another package such as SPSS.
0, 0, 0, 0, 0,
e5 e4 e3 e2 e1
1 1 1 1 1
1
0
twopar
Emotional Support
single
1
0,
e6
other
Running Regression with Multiple subgroups
Example 3: three observed independent variables (family structure variables for two
parents (non-biological, single parent, and “other”) predicting an (unobserved) latent
dependent variable (emotional support) composed of five observed variables, models run
separately by gender
• You may also filter the dataset so that you are only looking at men or women,
respondents of different races, etc.
• Double-click “Group number1.” You may now rename the first group (i.e. males)
and click the “new” button to create and name a new group (i.e. females)
• Where you select the dataset, you now must select a dataset (it may be
the same one) for both groups
• For each of your groups, click on the button “grouping variable,” select which
variable identifies your groups (i.e. female), and click “ok.”
• Click on the group you choose to identify, and then click the “group value” button
and select which value identifies your group (i.e. 0=male).
• Do this for each of your groups, and then click “ok” to leave the “data files”
window
• A box on the left-hand side of your output will allow you to select each group.
While the model fit applies to all of the groups together, the estimates will change
by group.