** ** ****** ** ** ******** ** ** ***** ** ** ** *
** ** ** *** ** ** ** ** ** * ** ** ** *
**** **** ** * ** ** ** ** ** **** **
** ** ** ** *** ** ** ** ** ** ** **
** ** ** ** ** ** ** ** ** * ** ** **
** ** ****** ** ** ** ***** ***** ** ** **
****** ****** ***** ** **** ******** ****** ******
** ** ** ** * ** ** * ** ** ** **
****** **** ** ** ** ** **** ******
** ** ** ** **** ** ** ** ** ** **
** ** ** ** ** ** * ** ** ** ** **
** ** ****** ***** ** **** ** ****** ** **
**** KY REGISTER **** KY REGISTER **** KY REGISTER **** KY REGISTER ***
MARCH 1989
TABLE OF CONTENTS
CCS Spring Seminar Series. . . . . . . . . . . . . . . . . . . . . . 44
UKCC Short Courses . . . . . . . . . . . . . . . . . . . . . . . . . 80
SUPER! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Keeping Your Userid Secure . . . . . . . . . . . . . . . . . . . . . 157
BMDP5V: Unbalanced Repeated Measures with Structured
Covariance Matrices . . . . . . . . . . . . . . . . . . . . 197
Suggestions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
Classifieds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
Service Directory . . . . . . . . . . . . . . . . . . . . . . . . . 549
*************************************************************************
CENTER FOR COMPUTATIONAL SCIENCES 1989 SPRING SEMINAR SERIES
All of the following seminars are free and open to the public.
Unless otherwise noted, all seminars will be held on Wednesdays, 4:00
p.m. in 137 Chemistry-Physics Building. Refreshments at 3:30 p.m.
March 1
Opportunities in Computational Chemistry & Materials Science
Thomas Weber, NSF and Bell Labs
March 8
The Use of Computers in Archaeological Research
Christopher Peebles, Indiana University
March 20 (Monday)
Supercomputers & Conformation of Biological Molecules
Harold Scheraga, Cornell University
March 29
Phase Transitions in Particle & Condensed Matter Physics
Gyan Bhanot, Florida State University
April 5
Imaging the Earth's Interior by Supercomputer
Larry Brown, Cornell University
April 12
Computational Fluid Dynamics
Philip M. Gresho, Lawrence Livermore Laboratory
April 19
Seismic Modeling of the Earth's Interior
Freeman Gilbert, University of California, San Diego
*************************************************************************
UKCC SHORT COURSES
The following short courses are free to all UK faculty, staff,
and students, but preregistration is required. If you register for a
course and then find that you will be unable to attend, please cancel
your registration by calling 257-UKCC. Failure to do so may
jeopardize your right to register for future UKCC short courses.
There are several ways you can register, depending on the class:
* You can register online. Enter PUBLIC, and then type
SHORTCOUrse, or enter VIEW UKCC SHORTCOURSE. Some
classes require online registration.
* You can register by calling 257-UKCC (this is Voice Mail
Exchange, your registration will be confirmed by phone
within five days).
Many of these classes don't require any knowledge or experience
with any computer system. If there are prerequisites for a particular
class, they'll be listed in the class description. If you have
questions about class content or bypassing prerequisites, call the
instructor for that class.
INTRODUCTION TO VM/CMS AND XEDIT
Monday, March 27 and Tuesday, March 28
6:00 to 8:00 p.m.
103 McVey Hall
A basic introduction to interactive use of the IBM mainframe systems,
this class presumes no previous knowledge of the IBM systems or any
other computer system. You'll learn how to access the computer, how
to create and manage files on your account, and how to use online
tools such as CALENDAR and VIEW. You'll also learn how to use the CMS
text editor, XEDIT, to create and modify individual files. This
course will be taught in two two-hour sessions. Both sessions will
provide hands-on practice of the commands that are covered. You will
be given a class computer account which will remain active for the
duration of the course. This course will be offered three times
during the Spring semester. Your instructor will be Pat Murphy
(257-2244).
ADVANCED GRAPHICS PROGRAMMING TOOLS
Monday, Wednesday, and Friday
March 20, 22, and 24
1:30 p.m. to 3:30 p.m.
326 McVey Hall (CCS)
This course will concentrate on more advanced features and
capabilities of DI3000 and other mainframe Precision Visuals software
modules. Topics will include the use of "metafiles" and the PVI
Metafile Translator, fundamental modeling and animation techniques,
retained segments and segment storage data structures, and contour and
3-D mesh surface mapping on the IBM systems. Newer additions to the
PVI software family will be introduced and demonstrated. Participants
in this workshop should be familiar with CMS, VS FORTRAN, and basic
DI3000 programming. Bob Williamson will be your instructor
(257-2227). Online registration is required for this course.
*************************************************************************
SUPER!
This year's annual meeting of SUPER! (Supercomputing by University
People for Education and Research) will be hosted by the University of
Kentucky, April 23 to 26, in Lexington. The theme of the meeting will
be "Supercomputing in the 90s."
University people actively engaged in IBM-based high-performance
computing, including researchers and teachers, computing center staff
and management, and user support staff are encouraged to become
involved in SUPER!
For more information, contact Sandy Leachman at 257-8737;
SANDY@UKCC.BITNET; or FAX 606/257-4000; or write
Center for Computational Sciences
325 McVey Hall
Lexington, KY 40506-0045
*************************************************************************
KEEPING YOUR CMS USERID SECURE
Your CMS userid and your files are safe from tampering as long as
your password is secure. Here are a few tips:
* For maximum security, turn off and then turn on public
terminals before you log on (but please leave public terminals
on when you leave). If you encounter a terminal with a
strange logo or that seems to be running some kind of a
program, turning it off and back on will force off whatever
might have been running.
* Give your password only to those who need it. It's best to
not give it out at all.
* Never write down your password or enter it on the same line
as your userid when logging on.
* Change your password frequently (use the PASSWORD command).
Avoid passwords that might easily be guessed by someone
trying to tamper with your userid (like your name or userid,
for example). Use the full eight characters for your password
and include some non-alphabetic characters (like digits or
punctuation).
* Never leave your userid logged on unattended in a public
area. When you logoff be sure that your userid is completely
logged off before you leave.
* If you ever encounter peculiar messages while logging on, if
your password isn't accepted, or if you can't log on because
your userid is logged on already, contact the UKCC immediately.
Although these suggestions are for CMS users, similar measures
can be used with most timesharing systems.
-- Dave Elbon
*************************************************************************
BMDP5V: UNBALANCED REPEATED MEASURES
WITH STRUCTURED COVARIANCE MATRICES
The latest release of the MVS/370 BMDP statistical subprogram
library includes BMDP5V, an interesting new program that will perform
analysis of variance of repeated measures or other correlated
responses. BMDP5V is useful in solving two frequent difficulties
that arise when analyzing such data: modeling the covariance
structure of the residual errors and dealing with missing
repeated measures.
Statistical programs such as SAS, SPSSx, and BMDP include
facilities for either the univariate approach assuming sphericity,
adjusted F-statistics, or the multivariate approach for unstructured
covariances. BMDP5V, on the other hand, enables you to easily analyze
data assuming virtually any identifiable covariance structure. It
also incorporates analysis adjusting for missing response measures.
Repeated Measures Analysis of Variance
The usual assumption in analysis of variance is that the random
error terms are uncorrelated across measures. When more than one
measurement is taken on a subject you have "repeated measures," and
this assumption is likely to be false.
The usual split-plot or mixed model analysis is still valid if
sphericity is assumed in the random errors of the repeated measures.
Sphericity holds when orthogonal contrasts of the repeated measures are
uncorrelated. In practice, assuming the errors are equally correlated
with equal variances (compound symmetry) is sufficient for sphericity
and is probably the most natural way for sphericity to occur. This
sphericity assumption is very important. If it doesn't hold, the
usual split plot tests will be anti-conservative, possibly very
anti-conservative. Examples exist showing how an F-statistic for a
null interaction that would be significant at a 0.05 level under
sphericity actually has a 0.6+ level! So whether or not you have any
effects, if sphericity does not hold and you assume it is true,
depending upon the covariance structure, you're likely to get
"significant" F-statistics.
By assuming sphericity, you are assuming a very specific
structure to the covariances of the repeated measures. If sphericity
is not reasonable, the usual univariate analysis is probably not
appropriate.
An alternative to the sphericity assumption is to assume that the
covariances of the repeated measures are unstructured. This suggests
a multivariate analysis. Suppose Y1, Y2, and Y3 are responses to a
repeated measure. The multivariate test that Y1-Y2 and Y2-Y3 are
simultaneously zero is actually testing the "main effect" of the
repeated measure. For such contrasts in the response, the usual
multivariate tests of "between subject" factors will correspond to
tests of interactions of the repeated measure and between subject
factors. However, these tests will often lose power to detect
differences relative to the univariate tests.
Another approach is to adjust the statistics in the univariate
analysis for lack of sphericity, usually using the Box correction
factor, epsilon, as an estimate of the amount of sphericity. If
sphericity does not hold, an approximate test can be generated by
comparing the usual F-ratio with the significance point derived from an
F-distribution where both the numerator and denominator degrees of
freedom have been multiplied by epsilon. This leads to the
Box-Greenhouse-Geisser-Imhof or the Huhyn-Feldt F corrections printed
by SAS, SPSSx, BMDP2V, and BMDP4V.
Both of these approaches assume either a rather restrictive
covariance structure, sphericity, or no structure at all. However,
there are times you may wish to assume other covariance structures
among the repeated measures. Econometricians have used AR(1)
autoregressive error structures, where, for example, the errors one
time unit apart have correlation p, while those two units apart have
correlation p**2, three apart p**3, etc. Thus errors closer in time will
be more highly correlated.
In other analyses you may wish to use more general autoregressive
covariance structures, a factor analytic structure, a random
coefficients model, or some general linear structure. Contrasts for
repeated measures assuming these structures will generally not be
spherical, and, presumably, multivariate modeling with an unstructured
covariance matrix will lose power. Again, BMDP5V allows virtually any
identifiable covariance structure.
The second major limitation to the usual repeated measures
analyses is dealing with missing repeated measures. First, the
multivariate approach requires complete data. Incomplete response
vectors must be deleted. To access the special features for repeated
measures in the standard packages, the data have to be entered in this
multivariate mode. Even if you want a univariate analysis, all
measures for a subject are deleted if any of the repeated measures are
missing, when using the specific repeated measures features.
You could reformulate the data and explicitly enter a subject
effect, but that has other limitations. For example, between subject
factors are usually completely parameterized; that is, one between
subjects parameter is introduced for each subject. Such designs often
exceed computing capacity when you add interaction error terms.
By comparison, BMDP5V does not require complete data on all
measures. It requires the assumption that the probability a response
is missing is independent of the "true" value of the response,
so-called "missing at random."
If sphericity holds, the between subject factors are fully
parameterized, and the data are missing at random, the usual tests for
repeated measures factors against within subject error are exact, but
the usual tests (using subject means) for between subject factors
aren't. Apparently, unless there are absolutely no repeated
measures/sub-plot effects, these tests are not even asymptotically
valid. BMDP5V provides asymptotically valid, maximum likelihood
estimates and associated tests for such missing data situations.
BMDP5V is not appropriate for all analyses. However, when you
have a repeated measures or a split-plot analysis with an unusual
covariance structure, or have missing data in your repeated measures
or sub-plots, you may wish to use this program for your analysis.
Some Specific BMDP5V Commands /DESIGN
DPName = varname
Names dependent variable for MODEL paragraph.
DPVar = var list
Specifies list of (repeated) variables for each subject.
GROUP = var list
Specifies between subject factors.
REPEATED = namelist
Names the repeated measures factors.
LEVEL = # list
Lists number of levels for each variable named in the REPEATED
statement.
CVNAME = covariate
Lists covariates associated with dependent variables.
name list. (time-varying covariates)
cov name 1 = var list 1.
cov name 2 = var list 2.
etc.
CONTRAST(var) = c list. Specifies partition of effects into
c1 = #1, #2, #3, ..., #n. contrasts.
c2 = #1, #2, #3, ..., #n.
etc.
/MODEL
dependent var = 'list of terms to include in model'.
INTERCEPT or NO INTERcept.
for example: Y='A + B + A.B'.
/STRUCTURE TYPE (one only)
CS Compound symmetry.
UNSTRUC Unstructured.
AR1 Autoregressive of order 1.
BANDED Toeplitz (general autoregressive).
FACTOR Factor analytic.
RANDOM Random effects.
LINEAR Linear in a set of specified basis matrices.
USER Other user specified structure.
(The last four structures require additional parameters to define
the covariance structure.)
/COMPUTE ALGorithm (one only)
NR Maximum Likelihood using Newton Raphson
FS Maximum Likelihood using Fisher Scoring
GEM Maximum Likelihood using Generalized EM
REML Restricted Maximum Likelihood using Generalized EM
REML Restricted Maximum Likelihood using Quasi-scoring
MAXIT=#. Specifies maximum number of iterations.
/PRINT
COVR. Prints estimated covariance matrix of regression
coefficients.
COVC. Prints estimated covariance matrix of covariance
parameters.
(Not available when ALG = REML or QSR)
RES. Prints response, residuals, etc.
An Example
Suppose you've measured a response, y, to one of three levels of
some treatment. You took measurements the same time each day for two
days, once in the morning and once in the evening. Overall, you have
four repeated measures. You decide it is appropriate to analyze this
as two repeated measures factors, specifically, day and ampm, for time
of day. Finally, you suppose an AR(1) error structure seems
appropriate.
A possible BMDP5V program for this would be:
/PROBLEM TITLE IS 'ONE-GROUPING, TWO TRIAL FACTORS'.
/INPUT VARIABLES ARE 5. FORMAT IS FREE.
/VARIABLE NAMES ARE TREAT,YAM1,YPM1,YAM2,YPM2.
/GROUPING CODE(TREAT)=1,2,3.
/DESIGN GROUPING=TREAT.
LEVEL=2,2.
REPEATED=DAY,AMPM.
DPNAME=Y.
DPVAR=YAM1,YPM1,YAM2,YPM2.
/MODEL Y='TREAT + DAY + DAY.TREAT + AMPM + AMPM.TREAT +
DAY.AMPM + DAY.AMPM.TREAT'.
/STRUCTURE TYPE=AR1.
/PRINT COVR. COVC.
/END
1 10 * 11 12
data
3 13 14 15 15
BMDP5V is described in the BMDP technical report #86, "BMDP 5V -
Unbalanced Repeated Measures Models with Structured Covariance
Matrices." A reference copy is available in the Consulting Room, 110
McVey Hall. A more detailed description of the algorithms appears in
Jennrich and Schluchter's paper in "Biometrics," 1986, pp. 805-820.
For more information or help with BMDP5V, contact Steve Thomson,
STEVE@UKCC, 120 McVey Hall, 257-2259.
-- Steve Thomson
*************************************************************************
SUGGESTIONS
1. Does anybody know how to unerase files for the Macintosh? Many
thanks in advance.
>>>It's sometimes possible to recover erased files, and there are
a number of commercially available utilities for this. The
MacZap package is one.
2. I'm using a hard-wired IBM PS2 model 30 and Kermit to call the
library hookup LS/2000. I want to read the library holdings into a
text file so that I can manipulate it with WordPerfect to make
bibliographies. Is this possible via UKnet? Will Kermit work, or do
I have to use a communications package like PC Modem? If I can use my
UKCC account, please give directions for making an XEDIT textfile out
of the library information which I can then download. Thanks.
>>>Try using Kermit's session logging. This copies characters that
appear on the screen during connect into a specified file on your PC.
Kermit-MS->log session filename.ext Kermit-MS->connect
Logon to the library system and call up what you want recorded. When
you've captured everything you need, escape back to kermit ms
Kermit-MS->close session
You can then bring the file into WordPerfect and modify as needed.
3. Many of the terminals in 111 McVey Hall do not work or have bad
keys. Teaching students how to use the computer systems is becoming
more and more difficult because of this. Would someone please take
the time to fix these terminals so that finding one to use isn't as
hard as finding a parking place?
>>>Thanks for bringing this to our attention. Much of the equipment
in 111 McVey is very old, and we're trying to secure funds for
upgrades. In the meantime, we're continually trying to keep up with
the necessary repairs. Whenever you find a broken terminal, please
report it to the Data Center immediately.
4. I heard some time ago there was an Amiga 2000 in the Micro Lab.
I recently went there to use that machine and was informed it was out
for repairs and that it had been out for repairs a long, long time.
What's going on? Amigas would be valuable additions to the Micro Lab
since there are so many Amiga owners on campus.
>>>We've had a lot of problems with the machine, with the company we
purchased it from, with the same company when it came time for
repairs, and in getting the money for the repairs. A local company is
now repairing the machine, and we hope it will be back with us soon.
5. An online ZIP CODE directory would be very handy.
>>>We agree. We're working on getting this information online, and
details will be published as soon as they're available.
6. Would it be possible to add an option to the calendar that would
let you see one week at a time and allow more information for each day
to show up on the screen without going to the detail?
>>>We'll add that to our list of possible future enhancements. Thanks
for the suggestion.
7. I recently obtained an account on UKCCXA. I used my profile
exec to define: SET PF1 RECALL, and it isn't working. I also can't
change my password. Is there any way to implement these features
which are used in almost all systems? Can make SAS accessible on
systems other than UKCC? I'm trying to use SAS Graphics.
>>>Use SET PF1 RETRIEVE. XA requires RETRIEVE; the other VM systems
accept both RETRIEVE and RECALL. Password changing will be available
after VM/XA SP 2 is available and the necessary programs have been
converted. We're working on it. We don't have a SAS license
for the IBM 3090.
8. It would be helpful if there was a command that could be issued
to find one's project numbers. Is this possible?
>>>The BALANCE command can give you the account number for CPU time.
It doesn't display this directly, but it has a STACK option that will
give it to you. To see what this does, use the STK command:
STK BALANCE ( STACK
This will execute BALANCE with the STACK option and display the
stacked data. See HELP BALANCE for more information. There is no way
to determine your disk account number at this time, but we'll add it
to our list of possible future enhancements.
9. We're connecting a TVI955 to an Apple ImageWriter II. What is
the procedure for printing locally using the PRIMEs, IBMs, etc.?
Does the print key on the TVI955 do anything? What is the
appropriate set-up?
>>>Your TeleVideo 955 manual should explain what, if any, local print
functions it supports. When logged on to VM and using CMS the APRINT
command may be able to print on an attached printer. It has support
for a TeleVideo 950 and that may work with a 955 also. See HELP
APRINT for more information.
10. The LWSAS procedure doesn't generate PostScript in the suggested
style as documented in Appendix C of the "PostScript Language
Reference Manual." If someone does LWSAS (file to have the output
file saved, then FTP the resulting file over to a Math Sciences
machine for output, the "%!" line at the beginning is missing. While
it isn't a big deal to add that line, it would be a great help if it
were there. As is documented, the Unix printer drivers do PostScript
stuff when they see that "%!" line at the beginning.
>>>Both LWSAS and LWPLOT now generate files that start off with
"%!PS-Adobe-2.0." Use LWSAS (CCxxx VIA *. The PostScript file will
go into your reader. You need a valid printer, even though it's not
used. Add the KEEP option if you want to keep the SAS metafile.
11. When I try to access the UKCC FTP server from another campus machine
the server allows the connection but doesn't establish a "working
directory". I'm unable to list the files on any of my CMS minidisks
or to move files back and forth between the machines. Efforts to
establish such a "working directory" result in an obscure error
message being sent by the UKCC FTP server. How can the working
directory on the remote machine be established? Is it not possible
to use the UKCC FTP server to move files onto a CMS minidisk (as
opposed to getting them from a minidisk)? Thanks.
>>>We use AUTOLINK to authorize links to most minidisks instead of
the standard VM password scheme. Since there is no password for the
minidisk, the FTP server can't link to it, regardless of what you
put on the CD command. For now, you can't FTP from a remote host
to UKCC when a private minidisk is involved. You can FTP when a
public disk is involved, such as PUBLIC 191. You can also log onto
UKCC and use FTP to the remote host to transfer files.
*************************************************************************
UKCC CLASSIFIEDS
Ads are free to UK students, faculty, and staff. If you'd like to place
a computing-related ad, contact the Editor at 257-2219, EDITOR@UKCC,
200 McVey Hall for details.
FOR SALE: NEC Multispeed Laptop computer, 640K, two 3.5" drives,
clock speed 9.54 Mhz or 4.77 Mhz selectable, weight approx. 5 kg.
Built-in calendar, detachable LCD screen, full keyboard with ten
function keys and number pad. A very nice computer and used only
for nine months during a sabbatical. $850.00. Call Newbery at
257-6746 or 252-6944.
*************************************************************************
UKCC SERVICE DIRECTORY
McVey
Service E-Mail Address Phone Hall
Vice President, Information Services
Eugene R. Williams DPS128@UKCC 257-3609
Director, University Computing Services
Dr. Douglas Hurley HURLEY@UKCC 257-2900 128
Director, Communications & Distributed Systems
Doyle Friskney DOYLE@UKCC 257-6225
Director, Computational Sciences
Dr. John Connolly CONNOLLY@UKCC 257-8737 324
Academic Consulting Services
Lavine Thrailkill UKC105@UKCC 257-2257 121
CMS Consulting
Bob Crovo CROVO@UKCC 257-2258 109
Complaints
Carol Lotz LOTZ@UKCC 257-2213 129
Consultant for Remote Sites
Wanda Dixon Spisak WANDA@UKCC 257-2206 115
Consulting
Consultant on Duty SUGGEST@UKCC 257-2249 110
Contingency Planning & Security
Jack L. Coffman UKA051@UKCC 257-2273 218
Database - IDMS
Rick Chlopan DBA003@UKCC 257-2211 230E
Data Center 257-2222 61
Data Entry
Frank McCormick OPFRANK@UKCC 257-2216 72
Disk Rental
Janet Hyatt HYATT@UKCC 257-2212 130
Larry Johnson JOHNSON@UKCC 257-2217 130
Facilities Operations
Joe Williams UKA048@UKCC 257-2231 122
Graphics Consultation
Bob Williamson ROBERTT@UKCC 257-2227 207
Information Center
Judy Kisil UKA041@UKCC 257-2241 222
Information Resources
Dr. Jon Hesseldenz UKA045@UKCC 257-3904 230D
Instructional Software
Wayne Beech WAYNE@UKCC 257-2238 100
Machine Room 257-2222 59
Management Information Systems
Forrest Hahn UKA006@UKCC 257-2260 123
Memos and Manuals
Consulting Room 257-2249 110
Micro Lab 257-2207 107
Network/Telecommunications
UKT101@UKCC 257-2229
New Accounts
Janet Hyatt HYATT@UKCC 257-2212 130
Larry Johnson JOHNSON@UKCC 257-2217 130
Numerical Analysis Consulting
Anne Leigh ANNE@UKCC 257-2205 109B
Optical Scanner - NCS
Chris Corman CHRIS@UKCC 257-2243 109
Bob Crovo CROVO@UKCC 257-2258 109
Passwords
Janet Hyatt HYATT@UKCC 257-2212 130
Larry Johnson JOHNSON@UKCC 257-2217 130
PRIME Information
Peggy Akridge PEGGY@UKCC 257-2237 100
Program Documentation/Libraries
Consulting Room 257-2249 110
Publications Office
Marguerite Floyd EDITOR@UKCC 257-2219 200
Refunds
Consulting Room 257-2249 110
SAS and SPSS Consulting
Steve Thomson STEVE@UKCC 257-2259 120
Lorinda Wang UKC333@UKCC 257-2204 109B
Statistical Consulting
Steve Thomson STEVE@UKCC 257-2259 120
Tapes to Borrow, Tape Storage
Data Center 257-2222 61
Tours of UKCC
Lavine Thrailkill UKC105@UKCC 257-2257 121
User Account Services
Janet Hyatt HYATT@UKCC 257-2212 130
Larry Johnson JOHNSON@UKCC 257-2217 130
Vectorization Consulting
Tom Faller TOMFAL@UKCC 257-2236 314
*************************************************************************
UNIVERSITY COMPUTING ADVISORY COMMITTEE
Douglas E. Hurley, Central Administration
H. Clay Owen, Central Administration
A.J. Hauselman, Community Colleges
James W. Phillips, Community Colleges
Raphael Finkel, Lexington Campus
Leonard K. Peters, Lexington Campus
N. Clare Detraz, Medical Center
David A. Nash, Medical Center
T. Earle Bowen, Ex Officio
Ben W. Carr, Ex Officio
Wimberly C. Royster, Ex Officio
Donald E. Sands, Ex Officio
Eugene R. Williams, Ex Officio
*************************************************************************