Upload
benedict-barker
View
212
Download
0
Embed Size (px)
Citation preview
A Review of ‘A Step-by-Step Approach to Using SAS for Factor Analysis and Structural Equation
Modeling’ (Larry Hatcher)
By Barry DeCicco, September 23, 2010
B. DeCicco, MSUG, 23 Sep 10 2
Do I recommend this book?
• Yes - if you are looking to start performing factor analyses and structural equations modeling (SEM) in SAS, and you do not have extensive experience in other statistics packages, especially if your math/stats background is weak.
• No - if you are familiar with factor analysis/SEM in other packages, then you can probably go to the SAS/STAT user’s guide, and learn the new syntax (see ‘other resources’ slide near the end).
B. DeCicco, MSUG, 23 Sep 10 3
What this book is NOT
• A matrix algebra explanation of factor analysis and structural equation modeling (SEM).
B. DeCicco, MSUG, 23 Sep 10 4
What this book IS
• A light regression explanation of factor analysis and SEM.
• A guide on how to perform that in SAS, what the goals are and how to interpret the results.
• (in the appendices) an introduction to how to do things in SAS.
• Approachable with a low level of statistical background.
B. DeCicco, MSUG, 23 Sep 10 5
Outline of this book• Principle component analysis (looking for combinations of
observed variables).• Factor analysis (looking for latent variables which explain
observed variables).• Scale reliability (a short chapter, need-to-know material only).• Path analysis – manifest (observed) variables only.• Measurement models/CFA (linking manifest with latent variables
to evaluate measurement quality). • Structural equation analysis – manifest variables measuring
latent variables which are linked to each other; this book also covers model-fitting.
• Appendices on how to use SAS, write programs, enter data, and handle output.
B. DeCicco, MSUG, 23 Sep 10 6
What needs supplementing
• Fit indices – they cover a variety of fit indices and helpfully list them by portion of the model being tested (pg 391), but this field is rapidly evolving.
• ODS – this is a pre-ODS book, and I found ODS extremely helpful in getting output into useable form (e.g., into Excel).
B. DeCicco, MSUG, 23 Sep 10 7
What I liked
• Extremely step-by-step.• Good explanation on how to estimate
reliabilities of latent variables. I was able to estimate reliabilities and ‘true’ correlations of latent factors, comparing them to a more naïve approach.
B. DeCicco, MSUG, 23 Sep 10 8
Comment on SAS Syntax
• With SEM, Proc Calis requires multiple long lists of linear equations.
• Using Excel formulae to write them out saves a lot of time:
B. DeCicco, MSUG, 23 Sep 10 9
Comment: SAS vs. AMOS• I worked on learning Proc CALIS (the classic SEM
procedure for SAS) with this book at the same time I was learning AMOS, which is the SPSS SEM package.
• AMOS has a graphical user interface.• This is great; one can draw models, and AMOS can
automatically fill in multiple items.• Sometimes AMOS’s ‘help’ isn’t so helpful.• I was surprised that it was frequently no harder to
do things in SAS than in a GUI-driven program.
B. DeCicco, MSUG, 23 Sep 10 10
The Sweet Side of AMOS
0, 1
Hospital Environment
0, 1
Nursing
Nurseslisten
0,
e1
Nrsng court,resp
0,
e2
1
1
0, 1
Doctor
Doctorsexplain
0,
e3
Doctorslisten
0,
e4
Doctors courtesy,respect
0,
e5
1
1
1
0, 1
OutcomesRate
0,
e6
Recommend
0,
e7
1
1
Nursesexplain
0,
e81
Noise
0,
e91
Roomclean
0,
e101
0, 1
Hospital Experience
Help toileting
0,
e111
Pain cntrl
0,
e121
Staff help w/pain
0,
e131
Explainnew med
0,
e141
Describe medside effect
0,
e151
Call button
0,
e161
B. DeCicco, MSUG, 23 Sep 10 11
The Sour Side of AMOS
0,
Admission
a1
0,
e11
1a2
0,
e21
0,
Discharge
d1
0,
e31
1d2
0,
e41
d3
0,
e51
0,
Visitors
v2
0,
e61
1
v3
0,
e71
0,
Meals
m2
0,
e81
1m3
0,
e91
m4
0,
e101
0,
Doctor
p1
0,
e11
11p2
0,
e121p3
0,
e131p4
0,
e141p5
0,
e151
0,
Tests_treatments
t1
0,
e16
1
1
t3
0,
e171
t5
0,
e181
t7
0,
e191
0,
Room
r1
0,
e201
1
r2
0,
e211
r3
0,
e221
r4
0,
e231
r5
0,
e241
0,
Inconvenience
i60,
e25
1
1i50,
e261i40,
e271i3 0,
e281i10,
e291
0,
Nursing
n1
0,
e30
1
1
n2
0,
e311
n3
0,
e321
n4
0,
e331
n5
0,
e341
n6
0,
e351
0, 1
Outcomes
o20,
e36
1
1o30,
e371o40,
e381
B. DeCicco, MSUG, 23 Sep 10 12
Other resources for this topic• ‘A Handbook of Statistical Analyses with SAS’ has a chapter on factor
analysis; nothing on SEM.• SAS Help documents ‘The FACTOR Procedure’, ‘Introduction to
Structural Equation Modeling with Latent Variables‘ and ‘The CALIS Procedure’ cover the syntax, and have very condensed introductions to the material.
• http://www.ats.ucla.edu/stat/sas/default.htm has a very large body of very well done work on how to use SAS, and how to run statistical analyses.
• Statnotes: Topics in Multivariate Analysis, by G. David Garson (http://faculty.chass.ncsu.edu/garson/PA765/statnote.htm) The best site on the internet for people who find themselves needing to do a lot of graduate-level statistics, without the graduate degree in statistics. It generally assumes the use of SPSS, but the concepts translate.
B. DeCicco, MSUG, 23 Sep 10 13
Where to get this presentation
• Go to the Ann Arbor ASA website: https://sites.google.com/site/annarborasa/
• Choose ‘Presentations and Class Materials’ on the left-hand menu: https://sites.google.com/site/annarborasa/A2ASA/presentations-and-class-materials
• Look for the folder ‘September 2010 MSUG Presentation’. This presentation and any associated documents will be there.