Prev Up Next
Go backward to 3.1 The Main Window Items
Go up to 3 glmlab Reference
Go forward to 3.3 The Main Window Buttons

3.2 The Main Window Edit Areas

3.2.1 The Response Area

The user types the name of the response variable (usually designated as y) in this area. Most legal MATLAB commands can be used (like [1 2 3]' or randn(100,1)) though glmlab could complain if the commands become too complicated.

The binomial distribution requires two columns in the Response area; one for the observed counts and the other for the sample size. If only one column is given, it is assumed to be probabilities if all its elements are between 0 and 1, otherwise glmlab issues a warning. See Section 5.5. No model is able to be fitted until a valid string is entered into the Response area.

3.2.2 The Covariates Area

The user types the name of the covariates (usually designated as X) in this area. It is possible to use most legal MATLAB commands (like [1 2 3]' [3 0 1]' or randn(100,4)) though glmlab could complain if they become too complicated. In particular, glmlab will become confused if a matrix is entered, or if some covariates are listed by variable name and others using legal MATLAB commands. glmlab requires that each column of input into the Covariates window has its own MATLAB variable name (or command). For example, valid strings are:

   [1 2 3]',[32 92 21]'
   Wt, Ln
where Wt and Ln are valid MATLAB variables with one column each. Invalid entries are:
   [1 2 3;3 4 5]
  magic(8)
  Msrments(:,1)
  Msrments
where Msrments is a matrix.


 It is best to define vector variables in the MATLAB workspace and enter the variable names in the glmlab window. It is best not to use matrices.

Strings of variables can be separated by commas or spaces. glmlab tries to sort out the string if it contains other characters, or double commas and so on, but it is certainly not foolproof. The constant term is included in fitting models by default, but can be controlled through the Options menu.

3.2.3 The Prior Weights Area

If prior weights are to be used, the variable name is entered here. Prior weights are used, for example, in weighted regression, log-linear models where there are structural zeroes, or when using the binomial distribution (see Section 5.5). If the variable consists of all zeroes, an error message is given. Valid MATLAB vectors are allowed.

3.2.4 The Offset Area

An offset is a variable with a known coefficient. The offset variable name is entered here. Valid MATLAB vectors are allowed.


Peter Dunn

Prev Up Next