Statistical Computing: Resources and Trends
Resources for UCLA Researchers
Trends in Consulting Questions
The nature of consulting questions have changed considerably over time.
- Many questions used to be on "data management" and "data transfer" type
- As resources have grown for solving these mundane questions, clients ask
higher and higher level questions.
- As packages make more statistical features available, clients seek help
and information to exploit these features.
- As it is easier and easier to move data among packages, clients more
willing to move among packages to obtain special features.
Trends in Data Analysis
The following are areas where we have seen increased interest over the last
year in our statistical consulting.
- Availability of techniques in statistical packages appears to drive user
- When you have a solution available, more eager to tackle the problems that
- Missing Data
- Moving from simple techniques (listwise deletion, mean substitution,
single regression imputation) to more complex techniques (e.g. multiple
- Software has become available to make multiple imputation possible.
- Multilevel Modeling
- Seen explosion of interest in this area (Afifi class had 60+ students).
- Contextual effects (e.g. students within schools, patients within
- Longitudinal Analysis, tolerates missing data, time spacing need not be
- Packages that solve this include SAS PROC MIXED, SPSS Mixed, HLM, MLwiN
- Starting to see focus on other kinds of outcome variables (binary,
latent variables), Stata gllamm, M-plus
- Survey Data Analysis
- Packages that solve this include Stata, SUDAAN, and WesVar
- While SUDAAN has been available for quite a while, Stata has brought
considerable ease of use and availability with a wide variety of analyses.
- Logistic Regression
- This has been an area of frequent confusion and misunderstanding.
Pedhazur found a majority of articles using logistic regression interpreted
the results incorrectly.
- With Long's Categorical Data Analysis book, an increased emphasis on
- With Long and Friese's
Regression Models for
Categorical and Limited Dependent Variables Using Stata, tools exist for
aiding the interpretation of results.
What's New in Stat Packages
- Stata 8
- Recent Developments in SAS 8.x
- IML Workshop
- Programs/Samples/RotatingPlot/ Plot3D
- Graphs via active agent
- Upcoming Developments in SAS 9
- Data Step will permit access to PERL
- MI Permits Class Statement
- MI has experimental options LOGISTIC and DISCRIM in the
MONOTONE statement impute missing categorical variables by
logistic and discriminant methods, respectively.
- MIANALYZE has test option for testing linear hypotheses about
- ROBUSTREG procedure added
- TPHREG is like PHREG and allows class variables
- Graphics integrated into procedures
- PROC GLM will produce cell mean plots
- PROG LOGISTIC will produce diagnostic plots
- PROC MIXED will produce residual graphs
- Additional Resources
The content of this web site should not be construed as an endorsement
of any particular web site, book, or software product by the
University of California.