Programs: BYCOMP.SAS & MACOMP.SAS

Authors: Peter Ott & Fred Hovey
Created: June 3, 1997
Versions: 1.0

Here are two SAS programs that perform Compositional Analysis as
described in the work by Aebischer et. al. (Ecology. 1993. 74(5): 1313-1325).

It is important that the user fully understand the above paper, and has some
knowledge of SAS before proceeding. We realize that both programs are not
optimal in terms of efficiency, but they do run well even on
older versions of SAS, as long as the SAS/STAT module is installed.
The programs are also straightforward to modify/customize
(e.g. incorporating design effects, changing the style of input, etc.).
They've been tested on several different platforms including Windows NT 3.51, Windows 3.11 and
UNIX, in both batch and interactive modes. Additional error monitoring has
*NOT* been incorporated into the code.

The two programs do differ markedly in speed and memory requirements.

BYCOMP.SAS uses a SAS 'by' statement to cycle through the randomization procedure.
This allows the program to run fast but also demands a lot of memory (16 MBYTES
is a *minimum* amount to run 999 simulations). A very large file (rwilks.txt) can
be generated and so disk space may also be a concern. Therefore, it is best
suited for users with (access to) beefy machines.

MACOMP.SAS uses a macro loop to cycle through the randomization scheme. Although
this makes the program very slow, it takes up only a small amount of memory.
It is best suited for users with less powerful (or unstable) machines. Once
familiarized with this program, we recommend running BYCOMP.SAS to increase
its speed. For example, when 999 randomizations were performed on an old SUN Sparc
workstation, MACOMP.SAS took 40 minutes (using the XWindows GUI) to complete,
while BYCOMP.SAS finished in only 5 minutes!

The two programs are similar in most other respects.

Please be aware that two files are generated. The file 'rwilks.txt' is the output
(Wilk's lambda) of the randomization(s). The file 'owilks.txt' is the output
(Wilk's lambda) based on the observed data. These two files are parsed and the
statistics are compared to calculate a pvalue. As it stands, the two files are
placed in the SAS working directory. You may change the location for these
files by appending a path to the filename statements (the stuff in quotes)
near the beginning of the code. Do *NOT* change the file references "rwilks"
and "owilks".

The user must supply the data as an external text (space delimited) file in the
following form: A one line header labeling the columns must be present, which
consist of observed data (proportions) ordered as "used habitat 1", "used habitat
2", ... , "used habitat D", "available habitat 1", "available habitat 2", ... ,
"available habitat D". Habitat types that are available but not utilized by an
animal should have been replaced by a small number (e.g. 0.01%). True 'missing
values' (i.e. when the habitat type is neither available nor used) are not (*yet*)
permissible - this situation can be remedied by removing animals with missing
values or collapsing the number of habitat types. Please see the provided file
"radvsmcp.txt" for an example; this file depicts pheasant radio locations vs. MCP
home range information and is taken directly from Aebischer et. al. (1993).

The ordering of the data-columns is retained throughout the program, including
the display of the final output - keep this in mind when interpreting the results.
Note that it is common for the available proportions to be equal for every animal;
modifying the code to efficiently read such data involves changing the input
statement in the "raw" data step accordingly AND removing the "/2" from the first
"numb" data step.

The path in the third filename statement (near the beginning of the code) will
likely need to be modified to include a different location. Do *NOT* change the
file reference "use_avl".

The only other item that the user may need to change is the number of simulations
(which is one of the first lines in the code). Currently, it is set to 999.

We hope these programs are useful to you. Please give credit to their authors if
you do use them. We take *NO* responsibility for their misuse or any damage they
may cause (e.g. melted hard disks, fried cerebral cortices) in their current or
alternative form.

Cheers,

Peter Ott
peter.ott@gems7.gov.bc.ca
Biometrician
Research Branch
B.C. Forest Service
Victoria, B.C.

Fred Hovey
fhovey@galaxy.gov.bc.ca
Wildlife Habitat Analyst
Research Branch
B.C. Forest Service
Revelstoke, B.C.

P.S. For those who are interested in conducting a retrospective power analysis
after the smoke clears, Michael Friendly has an excellent program (that does
require the SAS/IML module) called MPOWER.SAS; it can be found at the URL:
< http://hotspur.psych.yorku.ca/SCS/sasmac/mpower.html >.
A recommended reference on this topic is: Stevens, J. P. 1980. Power of the
multivariate analysis of variance tests. Psychological Bulletin. 88(3): 728-737.