MacComp Readme

I've received (and tabled below) feedback from a number of users
regarding attempted execution of MacComp. This programs's author has
informed me that those running the program on machines with 68xxx
processors make sure to turn off 32 bit processing in the control panel.
The program will not run with it turned on.

Model System Ver. Execute?
--------------------------------------
Centris 650 - Y
Quad/Cent 660 7.5 N
Mac IIse 6.03 Y
Mac IIfx 7.5 N
PowerMac ?? 7.5 N
Quadra 630 - Y

RD

Quick Instructions for MacComp 0.90 (Composition Analysis)

John P. Carroll
Department of Biological and Environmental Sciences
California University of Pennsylvania
California, PA 15419

1. The disk contains 3 MacComp files. MacComp 0.90 is the main
program and all subprograms can be accessed from there. File 1
contains the data entry routines. File 2 contains the
randomization routines. Both File 1 and Files 2 can be launched
directly or accessed through MacComp. The only other files that
are needed are QB10.bl or QB10.cl which contain the runtime
libraries, and MathStatLib which contains routines necessary to
run the MANOVA program. **Warning-MathStatLib is copyrighted by
Clear Lake Research. Clear Lake recently gave me permission to
distribute routines needed to run my program.

2. The disk contains 2 sample data files. Testraw is a sample
raw data set in BASIC format containing proportion data on 7
different habitats used by 16 animals. Testlog contains the
logratio difference data derived from Testraw. It contains 42
columns of paired logratio differences, 21 for comparisons between
study area availability and amount of each habitat in home ranges,
and 21 comparisons between amount of habitat in each home range
and proportions individual radio-locations in each habitat for
each animal.

3. Data format has to be done in the form of percentages or
proportions in each habitat (2-8 habitats) for each animal. The
first habitat should be one that is widely used by all animals.
Data is entered with the Study Area (availability for first
analysis) proportions first, followed by the Home Range
(utilization for first analysis and availability for second
analysis) proportions, followed by the Radio-location (utilization
for second analysis) proportions as shown in the following
example.
Study Area Home Range Radio-locations
Anim Hab1 Hab2 Hab3 Hab1 Hab2 Hab3 Hab1 Hab2 Hab3
1 .5 .4 .1 .3 .4 .3 .3 .4 .3
2 .5 .4 .1 .4 .2 .4 .7 .3 .0
3 .5 .4 .1 .2 .0 .8 .9 .1

4. Data entry is most efficiently done using a spreadsheet
program and entering the data in the above format. Once the data
is entered on the spreadsheet SELECT only the data cells
containing the proportion data and COPY. I have not yet put in a
habitat naming function. The output is referenced by the letters
A B C...... in the same order as the original data matrix. Launch
MacComp, select DATA MANAGEMENT from the main menu and follow the
given instructions. You will then have the data in a raw data
file (for example the file Testraw) that can be analyzed by the
other program functions.

5. Before the actual statistical analysis can be completed the
raw data must be converted to logratio difference data. This is
done by selecting RUN LOGRATIOS from the main menu. You can then
save that data in a logratio difference file (for example
Testlog). Zero values are handled in one of 2 ways. When there
is a zero in the "availability" part of the analysis then zeros
are treated as missing data and when a zero occurs in the "use"
part then it is treated as a real value. Because zero values
cannot be log transformed a dummy value is automatically
substituted. The program prompts the user for that value and a
"rule of thumb" for its value is 1/10 of the smallest measured
value in the study (the program displays the smallest proportion
for you--if it is 0.10 the you could select a dummy value of
0.01). Examples of a zero for use and availability is found in
the previous example. For Animal 3, there is no Habitat 2 in the
home range. The habitat is available in the study area so it is a
use zero for the first analysis. When analyzing the radio-
telemetry locations Habitat 2 is not available for animal 3 since
there is none in the home range, therefore it is a utilization
zero.

6. The first step in the analysis if to select the RUN MATRIX
option from the main menu. You will be asked to supply the name
of a logratio difference data set (e.g., TESTLOG). This option
creates a matrix of all habitat pairs, including mean values,
standard errors, normality tests, and one-group t-tests. The
habitats are ranked according to relative selection for both
levels of comparison. However, the statistical significance of
this information is valueless until the MANOVA analysis is
completed. The reason for running this option first is to
determine if some habitats contain too many missing values. You
may want to decide to drop some insignificant habitats or pool
closely related habitats and then rerun the data.

7. The second step in the anlysis is to run a MANOVA by
selecting the WILK'S option from the main menu. You will be
asked to provide the names of the raw data (e.g., TESTRAW) and
logratio difference data. This is a Wilk's L analysis with a X2
test of its significance. For this analysis only a single habitat
is used as the denominator for each Wilk's analysis. If you have
no missing values then use of any of the habitats as the
denominator has no effect on your output and the simple X2 can be
used. However, if you have missing values then your L value will
be affected by the denominator habitat. The program handles this
for you by first replacing missing values with the mean value for
that habitat pair and then by using all possible habitats as the
denominator and calculating a separate Wilk's for each. A mean L
value is then calculated by weighting each L according to its
number of nonmissing values. The mean L cannot be tested using
the simple X2, instead you must then use randomization (see next
section) to calculate statistical significance. If these values
are statistically significant, testing the hypothesis that overall
habitat use if different from availability, then the matrix output
that you previously calculated can be interpreted properly.

8. Randomization is a type of statistical design where instead
of running your standard parametric statistics with their
necessary assumptions of normality, you create expected
distributions for your data and compare your results to that
distribution. This process eliminates the need to transform that
data. The process of creating the expected distribution of values
is a tedious one. The program does it by creating and running 999
expected values for your data. Then it compares your observed
value to this distribution. For the matrix data the randomization
process provides statistical significance values that substitute
for the one-group t-tests. For the Wilk's L analysis the program
calculates 999 expected values for L and compares your observed
value to that distribution. To run the randomization programs
select the RANDOMIZATION option from the main menu. Both
randomization programs can take quite a long time to run. The
most important variable in affecting run time is the number of
habitats. For example, with 7 habitats, the matrix analysis on a
Mac II with the math coprocessor takes about 10 minutes and the
Wilk's analysis takes about 2 hours. However, for 5 habitats the
run times are about 3 minutes and 30 minutes, respectively.

9. You can output logratio difference data to statistical
packages if you desire. This is done using the DATA MANAGEMENT
routine from the main menu. If you select the EXPORT function
then you can select either all of your logratio differences or
just those needed to run additional statistics using various other
independent variables.

10. There are several known limitations to this version of the
program. I have not completed some of the output save functions
as of yet. There is a save Wilk's option in the RUN LOGRATIOS
routine, but it does not do anything and is not needed. Finally,
for the Wilk's analysis, I have not completed the routine for
calculating mean values for 8 habitats. Otherwise, the program
has been independently checked with several sample data sets and
no other "bugs" appear to be present.

If you have any questions or need assistance please call me or
write to the following address:

Dr. John P. Carroll
Dept. of Biological and Environmental Sciences
California University of Pennsylvania
California, PA 15419

Phone: 412-938-4215