[Social-sciences-list] SAS

[Arielle Silverman]

Hi Robert and all,
Since my preferred stats program at this time is still SAS, I want to
offer some basic pointers about using SAS in case you might consider
using it in the future. I would be happy to correspond off-list with
anyone who is looking for SAS programs to run particular analyses.
I used SPSS when I was an undergrad student five years ago and found
it semi-accessible, but I switched to SAS when I started grad school
because the introductory stats sequence for grad students here at CU
requires everyone to complete problem sets using SAS. My university
has purchased a license for me and renewed it each year so I can run
SAS on my own laptop rather than trying to put a screen reader onto a
networked computer. I'm glad to hear that SAS works with NVDA as well
as JAWS.
It is possible to enter data directly into SAS, but I prefer to do all
my data cleaning in Excel and then use the SAS file import wizard to
bring the data in for analysis. The import wizard can be accessed
through the file menu. After locating the Excel file, it is necessary
to give the file a temporary name (I usually just call mine "data")
and then you will be ready to type or paste in your commands.
Below are a couple examples of basic commands I use routinely so you
can get a sense for how the programming works. As you can see, many
commands are quite simple and intuitive.

To get descriptive stats for a list of variables:
proc means;
var = [variables];
run;
To get a correlation matrix for a list of variables:
proc corr; var = [variables];

To get an alpha reliability coefficient for a series of scale items:
proc corr alpha; var = [variables];

To run a multiple regression analysis:
proc reg;
model [outcome variable] = [predictor variables];
run;
(note this runs a simultaneous regression where the effect of each
predictor is tested controlling for other predictors in the model).

To run a two-way ANOVA:
proc glm class = [factor names];
model [dependent variable] = [factor1 factor2 factor1*factor2 covariates];
lsmeans [factor1*factor2];
run;

These are just a few basic examples, and within each one it is
possible to request additional statistics. For example, in proc reg,
you can request partial correlations for each predictor by writing the
code as follows:
proc reg;
model [outcome variable] = [predictors] / pcorr2;
run;

Note that since you have to type all your variable names into the
code, you will want to be sure your variables have names that are
short and easy to remember.
As I said I would be happy to send additional code off-list if you
want to learn how to run a particular analysis that I have not listed
here.
Most recently, I have learned how to analyze structural equation
models using the tcalis procedure in SAS. I am the only one in my SEM
class who is using SAS, which was a little tricky, but I was able to
teach myself by following the online SAS support guidance at
support.sas.com

My personal bias is to use a program that is most accessible and least
frustrating, even if others in the class aren't using it. I think it
is better to have the ability to run analyses reliably on your own and
to consult online support if you get stuck, rather than using a
program that others are using but regularly dealing with accessibility
hurdles or unreliability.
Best,
Arielle