UnifyPow
A SAS Module for Sample-Size Analysis
Designed and Written by Ralph O'Brien
Department of Quantitative Health Sciences
Cleveland Clinic Foundation

20 September 2003

UnifyPow version 2002.08.17a

You can download a new version of UnifyPow: UnifyPow version 2002.08.17a (UnifyPow020817a.sas). This plain text file is distributed so that it runs by %INCLUDE-ing it as I have been demonstrating in the workshops.

Here also are the notes (PDF) that I covered in my workshop at the Joint Statistical Meetings in New York City in August 2002. UnifyPow Notes version 2002.08.11 (UnifyPowNotes020811.pdf). This is designed to be printed two-sided (duplex).

GraphPow

Chris Skibinski has developed GraphPow, which (as you might guess) graphs UnifyPow results using SAS/GRAPH. Click here to visit her website for the code and more information.

New Tools from SAS Institute

SAS Institute developer Dr. John Castelloe and his colleagues are designing and building products in this area that should eventually make UnifyPow obsolete. I am consulting actively and formally on all this, and I hope to now turn my attention to writing about sample-size analysis (with John) instead of programming and supporting UnifyPow.

The first products are:

* PROC POWER handles many of the basics.

* PROC GLMPOWER will handle linear models using syntax and modeling structure congruent with PROC GLM.

* Java-based interface that works from your web browser.

Experimental versions of these have been released in version 9.0, which has gone to selected sites. Version 9.1 will be the first full release of SAS 9.x, and it will contain production versions of these tools, which will have even greater functionality.

The current documentation is available in PDF from SAS.

Note: The rest of this site needs to be updated, but there is never enough time for this anymore. Sorry.

-Ralph O'Brien

UnifyPow is a freeware SAS module/macro that performs power analysis and other matters related to statistical planning for many types of common research designs and data analytic methods. It is described in O'Brien (1998), and is being demonstrated through various workshops and convention presentations. Don't misjudge UnifyPow because it is free: Its ease-of-use, capabilities, flexibility, a numerical accuracy are considerable. See commentary by Nicholas Petreley.

It is important that you notify me if you have concerns about anything. See quote by Fred Brooks.

Please register now!
Before downloading the files, it is important that you add your name and email address to the UnifyPow mailing list. Doing so will allow me to keep you informed about future releases of the freeware and its documentation. Registering is optional, of course, and I will not release the list or use it for any other purpose. Please click here to register.

Files for downloading

The following files may be downloaded by clicking on the filenames. *.pdf files need to be viewed and printed with Acrobat Reader, a breakthrough free tool that is commonly used on Windows 3.1/95/NT, Macintosh, and many varieties of UNIX. I have found that downloading Acrobat Reader 4.xx and installing it were easy tasks. Note: versions of Acrobat Reader earlier than 3.xx have failed to read my files.

• Tour of UnifyPow: most recent update of O'Brien (1998), the proceedings paper for the invited presentation at SUGI 23, March 1998, Nashville, TN. This describes the current and planned capabilities of UnifyPow and demonstrates its use. As UnifyPow evolves, this paper is updated and made available from this site.

• ReadMe file on UnifyPow: tells how to get started. This does not summarize UnifyPow's functionality and syntax. A plain text file.

• UnifyPow.sas source code (module version): current release of UnifyPow all in one %include module. This plain text file is distributed so that it runs as I have been demonstrating in the workshops. Technically, it is not a true macro in this form, but there are instructions in the opening lines to easily convert it to one. (I find no compelling reason to do this.)

• Input (.sas) file with numerous examples (module version): all examples used in the current one-day workshop and other talks and presentations. A plain text file.

• Output file with numerous examples: output (listing) produced by above input file. A plain text file.

• Workshop notes: notes (slides) from the current one-day workshop. This is not as good as a real manual, but hundreds of people have learned UnifyPow (relatively painlessly) by studying this. An Acrobat (.pdf) file.

• Cases: a few case studies (more fully described examples) distributed in the current workshop. An Acrobat (.pdf) file.

• Final draft version of O'Brien and Muller (1993) chapter: final manuscript that led to O'Brien RG, Muller KE (1993), "Unified power analysis for t-tests through multivariate hypotheses," Chapter 8 (pp 297-344) in "Applied Analysis of Variance in Behavioral Science," edited by Lynne K. Edwards and published by Marcel Dekker. I've updated the appendix to make it current with respect to freeware. An Acrobat (.pdf) file.

• Notes from my JSM-98 invited talk on generalized linear models. This covers power analyses for problems that fall within the generalized linear model, thus covering maximum likelihood logistic regression, Poisson regression, etc. The strategy is a direct extension of the method put forth in my 1986 SUGI paper on log-linear models (summarized, with my permission, by Alan Agresti in his terrific book, Categorical Data Analysis). I invited G. Shieh to join me in this 1998 effort and, to my dismay, he unilaterally published the essence of our work in a nice paper in Biometrics (December 2000). An Adobe Acrobat (.pdf) file.

• Manuscript by O'Brien & Shieh on a "Pragmatic, Unifying Algorithm (That) Gives Power Probabilities for Common F Tests of the Multivariate General Linear Hypothesis." I have stopped feeling guilty about not publishing this, because the few people who work in this area all seem to know about it and accept its content. An Adobe Acrobat (.pdf) file.

Downloading files using your Internet browser
Use either Netscape Navigator or Microsoft Internet Explorer. Here is what I have experienced myself; yours may be a bit different.

Plain text files:

1. Clicking on a plain text file (e.g., ReadMe.power) puts the contents directly onto the screen.

2. Doing a "Save As..." loads the file onto the local computer's hard disk. Specify either "text" or "source." If you are running SAS on a PC/Windows machine, you may want to store UnifyPow.sas in your SAS directory.

3. These plain text files can been read successfully with most editors and word processors in UNIX, Mac OS, and PC/Windows platforms. (Note: For the Mac, I encourage you to use the freeware BBEdit Lite. One of the usual Mac applications for this, SimpleText, will not read these large files.)

Acrobat files:

1. The Acrobat (*.pdf) files (e.g., Workshop.pdf) do not come up automagically in the browser's window unless the local computer is set up to link Acrobat Reader/Exchange with your browser. If not, the browser either asks whether/where you want to save the file directly onto your hard disk or it just saves it to someplace "obvious."

2. View and print with Acrobat Reader/Exchange.

3. If the image does automagically come up in the browser' window (because it is linked to Acrobat Reader/Exchange), you probably will still need to "Save As ..." the *.pdf file to your local hard disk. *.pdf files are binary; when saving them, specifying either "text" or "source" worked for me.

Thank you for your interest in my work.

Ralph G. O'Brien, PhD
robrien@bio.ri.ccf.org

Reference
O'Brien RG (1998), A Tour of UnifyPow: a SAS Module/Macro for Sample-Size Analysis," Proceedings of the 23rd Annual SAS Users Group International Conference, Cary NC: SAS Institute Inc., 1346-1355. You may download final update of this paper (Acrobat file).