HOF: Fitting Huisman-Olff-Fresco
models with maximum likelihood
I have recommend HOF models as an alternative to analyse whether
species responses are significantly skewed along gradients (Journal
of Vegetation Science 8, 147-152; 1997: abstract).
I have now developed a program for Maximum Likelihood estimation
(Poisson error) for HOF models. Data input is from CANOCO-formatted
files, and output statistics use Deviance statistics (like GLIM).
As a sample, here is a figure (gif, 14K) of
diatom responses along alkalinity gradient with HOF models.
I have moved my development work into R statistical environment. A
brief introduction into fitting HOF models in R is available in a pdf document. Despite my earlier message on this
page, I have started to develop the stand-alone hof program again.
However, the upgraded versions are available for Linux only (I don't
have an access to a Windows machine with development tools).
- I have started to develop an R package called `gravy' (gradient
analysis of vegetation) which includes first version of HOF
fitting. The package is still very preliminary and incomplete, and it
is not for the faint at heart. You may check `gravy' among R packages
at my software listing page (and pick the
Download Linux binaries hof-2.3.8-1.i386.rpm (96K, 10
Jun 2002). The rpm files require Dislin graphics libraries to run;
these can be downloaded free through Dislin homepage. I cannot
make full source code available, since hof uses Numerical Recipes which cannot be
distributed in source code. If you can help me to replace these
routines (maximizing likelihood function with dfpmin.c and evaluating
F and Chi-squared statistics) I will release the source code as
well. The rpm files were built for Red Hat 7.1 and they are GPG
signed. You should be able to get my public key at http://www.pgp.net/, ID 4F4D8DA5.
- DOS/Windows: Download hof.zip (executable,
version 2.3, 16/10/98, 174K), read the manual
of version 1 or the readme file
of version 2.3.
- Citing my HOF software.