Ho, Research and Statistical Support Services
What's new in the upcoming SAS 8.2?
In October 2000 I wrote about the new
developments of SAS 8 in RSS
Matters. Now, a half-year later, I'm writing about it
again. SAS 8e Release 2 will be out shortly, loaded with
more procedures and features than anyone could imagine.
Buckle up, this is not a bug-fixing upgrade only. Some of
the new features will test the limit of your imagination.
I know it does mine. For instance, sending http://www.unt.edu/benchmarks/archives/2000/october00/rss.htm
via a SAS program. That's right, you will be able to send
E-mail via SMTP (Simple Mail Transfer Protocol) within a
SAS program by selecting E-mail addresses from a large
address database (so be prudent in hitting that F3 key
next time). How about supporting Chinese or Japanese SAS
codes and data? Or converting your output directly to
your Web server, or in postscript, RTF and PDF files? It
does that, and more!
Here are some new features that caught my
Namely, Sun's Solaris 7, IBM's AIX
4.3 and HP-UX 11.0. This enhancement enables
researchers to take advantage of the 64-bit features
(such as processing files larger than 2 gb) both at
our UNIX research machine sol and the new software.
While this has yet to be tested, we are keeping our
fingers crossed on how much we can benefit from this
technology. That said, we can say we have the
capability to do it, can't we?
The long-awaited Linux version of SAS
finally comes into production. We are still in the
process of acquiring this new platform for SAS. But
before long, researchers on campus will enjoy using a
server combining the open-source operating system and
SAS. A raincheck is in order!
The Version 8 Output Delivery System
adds more versatility to SAS output. Release 8.2
includes output destinations such as RTF (enhanced),
PDF (now production) and XML (enhanced experimental).
In addition, new ActiveX/Java-enabled interactive
graphs enhance data visualization when output is
ported to the Web. Graph'N Go is an example that
exports charts loaded with Active X features (read SAS
Corner in February 2001)
Last but not least, something all
researchers are most looking forward to: what new
stats can SAS 8.2 do. Well, some of the experimental
procedures in 8.1 are fully implemented or in stable
mode. Among them:
GAM is not new, both to
statisticians and SAS developers. Hastie
and Tibshirani proposed in 1990 the modeling
technique that estimates an additive
approximation to the multivariate regression
function, to a broader range of distributional
families on top of the conventional linear
function. GAM models allow the mean of the
dependent variable to depend on an additive
predictor through a nonlinear link function such
as Logit or Loess. Besides providing the user
with the flexibility of nonparametric regression,
generalized additive models also have the
advantage of easy interpretability, which comes
from modeling the regression surface as a sum of
smooth terms. GAM procedures are available in
S-Plus and other packages as early as in the
mid-1990s. SAS 8.1 has this procedure, PROC GAM,
in experimental mode. The new release will have
it in production.
Another widely studied topic,
Missing Values analysis poses as a significant
addition to tools for researchers using
SAS. The easiest and most common
method of dealing with missing values is to
ignore or delete them. While convenient and
simple to implement, researchers will lose
information about if systematic missing value
patterns exist. Above all, precious information
from those cases which happen to have one or a
few missing values will also be
To deal with this problem,
statisticians have proposed many ways to
"patch the holes in the data." One of
them is imputation, or plugging in values to the
missing value cases. Mean or median can do
the job but usually not desirable because of the
lack of underlying rationale that mean or any
moment is good enough. Rubin suggested in
1976 the multiple imputation method by replacing
each missing value with a set of plausible values
that represent the "uncertainty" of the
real value of interest. Models with imputed
values are then estimated and combined to draw
inference. SAS 8.1 is equipped with two new
procedures, PROC MI and PROC MIANALYZE, to
perform this task. In the new release, these two
procedures will be in a second experimental
When can we get it?
SAS claims the software will be available
in mid-March via Web request. The conservative RSS
estimate, however, is that it will not be in our hands
until late-April. Faculty members should keep checking
with us if they are eager to test the new software. As
for students, I will certainly make student versions
available as soon as possible.
Bye for now . . .
By the way, RSS wishes you a nice and
safe spring break and happy computing.
SAS Institute. 2001. What's New in SAS
Software for Release 8.2 (http://www.sas.com/products/sassystem/release82/index.html)
Hastie, T.J. and Tibshirani, R.J. (1990), Generalized
Additive Models, New York: Chapman and Hall.
Rubin, D.B. (1976), "Inference and Missing
Data," Biometrika, 63, 581 -592.
Rubin, D.B. (1987), Multiple Imputation for
Nonresponse in Surveys, New York: John Wiley &