PQMethod Additional Program Files
Note: The following programs are contained also within the PQMethod 2.20 package, without documentation, however.
pqm2htm.zip (2012-01-04)
- Two command-line tools for 'converting' PQMethod
Q-Sort files to HTML:
PQM2HTM ('PQMethod to HTM') creates
a single HTML document containing all Q-sorts, PQM2mHTM
('PQMethod to multiple HTML files') outputs separate
HTML files for every Q-sort, together with all frame
files required.
- Installation: After downloading extract
both exe-files into a directory in your path
(e.g., into C:\PQMETHOD or into C:\WINDOWS)
- Usage (with the 'Lipset' example): Enter
(at the DOS command prompt):
- NB: you need not enter the full path if
you first cd to the project files' location.
Moreover, instead of the project name ('LIPSET')
you can as well specify one of the project files
(e.g., LIPSET.STA) in case you wish to make use
of windows' open file with or drag-and-drop
functions.
Note: The statements in the .sta file
need not be shortened, you can display sorts with long
statements also.
- horst55.zip
- Contains horst55.exe, a DOS program for Centroid factor
analysis with iterative solutions for communalities;
adapted from Horst (1965). Reads the
<study_name>.cor correlation matrix file as
produced by PQMethod (QPCA or QCENT), and outputs a
<study_name>.unr
unrotated factor matrix file which
in turn can be used as input for PQMethod. Fortran source
code included. The protocol with the iterated series of
communality estimates, factor loadings, and residual correlation matrix
is output to the file <study_name>.lis.
Horst, P. (1965) Factor Analysis of Data Matrices. Holt,
Rinehart and Winston.
Using "HORST 5.5" as an Alternative to PQMethod's Builtin Centroid Extraction Method (see Brown, 1980)
In its default setting, horst55
typically extracts two factors only. That is because Horst suggested a
stopping criterion that is rather strict, given the typically small
number of cases, m, i.e., statement numbers, in Q studies. It demands that the average squared residual correlation must be greater than 1/m.
So, for a study with 30 statements, this figure is .03, which does not
seem so small at first glance. However, un-squared it corresponds to r =
.18, and it refers to the average not the maximal value in the
residual correlation matrix. At any rate, you can experiment by
modifying the default parameters in the first header line of the .cor file (three numbers with three digits each, padded blanks or zeroes):
1-3: Maximal numbers of factors to extract (up to 8): Default is 0 for max. 8 factors.
4-6: n, number of sorts. Do not change.
7-9: m,
number of statements. If you change that number to, for instance, 100,
the stopping criterion would be changed to an average residual r2 < .01 (corresponds to an r
= .10). Setting the number to 999 practically overrules this criterion
for using the max. number of factors as the only criterion instead.
If you use the 'Lipset' example data for trying out, and change the first line in the original lipset.cor from
0 9 33 Lipset Study (PQMETHOD testfile; cf. Brown, 1980, pp. 183ff.)
to
3 9 50 Lipset Study (PQMETHOD testfile; cf. Brown, 1980, pp. 183ff.),
horst55 will output three unrotated centroid factors in lipset.unr. You need not mind the program message 'Communality estimates did NOT converge after 30 Iterations.' If you inspect the lipset.lis you
will see that the differences between 29th and 30th iterations are
really very small (the limit set in the program is .001).
Before you can continue with PQMethod for rotating factors and
running QANALYZE, you must not forget to correct the 'faked' number of
sorts in the first line of lipset.unr (from 50 back to 33).
Please note that different from the implementation of Brown's (1980) Centroid method
in PQMethod, the iterative process of inserting as starting values
communality estimates found after extracting all factors, factors are
not invariant over solutions with increasing numbers of factors
extracted.
Back to the QMethod Page
Peter Schmolck
<p41bsmk@unibw-muenchen.de> 2012-March-09