4. Options


opt = 1415 glmopt (string1, value1, ... {, opt})
creates a list of options for GLM estimation or appends options to an existing list

Options for the algorithm and optional parameters should be collected in a list object. This allows us to set or to modify those options which are necessary. Almost all functions in the glm library allow options. It is possible to give the same list of options to different functions. For example,


  opt=glmopt("miter",20,"name","MyDisplay")

will set the maximal number of iteration to 20 and the name of the output display to MyDisplay. Now, one can call first 1420 glmest and then 1423 glmout with the list opt:

  l=glmest("bilo",x,y,opt)

  glmout("bilo",x,y,l.b,l.bv,l.stat,opt)

Both 1426 glmest and 1429 glmout only consider those optional parameters which are intended for them. Hence 1432 glmest will only care about miter and 1435 glmout will present a display with the title MyDisplay.


4.1 Setting Options

Principally, it is possible to define the list of optional parameters with the XploRe command list. However, it is recommended to use the 1444 glmopt tool to set the options. The first call of 1447 glmopt will create a list of options. To append a further component to opt, we have to repeat the name opt as the last argument of 1450 glmopt. For example,


  opt=glmopt("miter",20,"title","MyDataset")

  opt=glmopt("name","MyDisplay",opt)

creates the list opt with a component miter containing the value 20, a component title containing the string "MyDataset" and a component name containing the string "MyDisplay". The resulting list has three components as one can check with names(opt).

The next sections will explain which options can be used in GLM estimation.


4.2 Weights and Offsets

Prior weights and offsets can always be given as an optional parameter. The corresponding components of the list of optional parameters are

wx
weights, nx1 vector or scalar. Default is wx=1.
off
offset, nx1 vector or scalar. Default is off=1.
Neither of both parameters should contain missing or infinite values.


4.3 Control Parameters

There is a number of control parameters which modify the used algorithm:

cnv
convergence criterion. The iteration stops when the relative change of the coefficients vector b or the relative change in deviance is less than cnv. Default is cnv=0.0001. This parameter is ignored in noniterative estimation (model code "noid").
miter
maximal number of iterations. The iteration stops when this maximal number of iterations is reached. Default is miter=10. This parameter is ignored in noniterative estimation (model code "noid").
fscor
indicator for Fisher scoring (instead of Newton-Raphson optimization). fscor=1 means that the Fisher scoring is used. Default is fscor=0 for Newton-Raphson. This parameter is ignored for canonical link functions.
norepl
norepl=1 forces not to search for replications in x. Default is norepl=0, i.e. to search for replications.

The following parameters switch on/off information during the computation.

shf
shows how the iteration proceeds, if shf=1 set. Default is shf=0.
shm
shows how the model selection proceeds, if shm=1 set. Default is shm=0. This parameter is only recognized for model selection.

The model selection functions 1457 glmselect, 1460 glmforward and 1463 glmbackward also provide two extra parameters:

crit
single string, "aic" or "bic" for Akaike or Schwarz criterion to use in model selection.
fix
indicates which columns in x are held fixed in model selection.


4.4 Output Modification

Functions which provide graphical output (1468 glmout, 1471 glmplot, 1474 doglm) accept special options to change output.

nopic
suppresses output display in 1477 glmout or 1480 doglm, if nopic=1. Default is nopic=0.
xvars
string vector, containing variable names for the columns of x.
name
single string, name for output and prefix for output displays.
title
single string, title to be used in 1483 glmout or 1486 doglm.



Method and Data Technologies   MD*TECH Method and Data Technologies
  http://www.mdtech.de  mdtech@mdtech.de