# Raw General Linear Model Code and Sample Input and Output

We provide here a raw implementation of the ideas discussed in Verrill (2001), "Rolling Your Own: Linear Model Hypothesis Testing and Power Calculations via the Singular Value Decomposition," Statistical Computing and Graphics Newsletter, 12, No. 1, 15-18. This paper is also available here in both LaTeX and postscript form. Warning as of 1/31/05: Some of our users have had problems downloading files via ftp from our site. If you click on one of the ftp links above and the ftp fails, you have two options. You can ftp from a DOS or UNIX command line (click here for details) or you can e-mail Steve Verrill at sverrill@fs.fed.us and ask him to e-mail the file).

## General linear model power calculations on the Web

The power calculation portion of the program can be run here.

## General linear model hypothesis testing on the Web

The hypothesis testing portion of the program can be run here.

## General linear model estimation on the Web

The estimation portion of the program can be run here.

## The source code

The main routine is in the file glm.f. The necessary LINPACK routines are in a file called linpack.f. These codes and sample input and output files are available in compressed tar, Windows 95 zip, and Windows 98 and later zip form. Alternatively, the codes and sample input and output files are available individually: The "91" files are used to work through the example associated with Table 9.1 in Milliken and Johnson's Analysis of Messy Data, 1992, Chapman and Hall, London. The ".x." files contain design matrices and data values. The ".runf." files contain the input that a user needs to type in to run the glm program (the glm program can also be run directly from these files). The ".res." files contain the results produced by runs on the associated ".x." and ".runf." files. "means" indicates what Milliken and Johnson call a "means" parametrization of the model. "eff" indicates what Milliken and Johnson call an "effects" or "classical" parametrization of the model. "pow" indicates a power calculation rather than an hypothesis test.

The glm program was created via the command line

f77 -g -o glm glm.f linpack.f dcdflib.a

where dcdflib.a is a static version of the DCDFLIB library of distribution functions. A user would need to create a version of dcdflib.a on their own machine for glm to work.

Here is a link to the DCDFLIB material.

[Forest Service] [Forest Products Lab] [FPL Statistics Group]