Generalized Linear/Nonlinear Models - Advanced Tab
Select the Advanced tab of the Generalized Linear/Nonlinear Models Startup Panel to access options to specify the desired method of analysis (see also, Generalized Linear/Nonlinear Models Index). In order to perform a generalized linear/nonlinear model analysis, a data file must be selected at this point.
This tab provides a variety of common regression analyses and experimental designs (see also the Introductory Overview section of the GLM module), and provides access to the three different user interfaces available in the STATISTICA Generalized Linear/Nonlinear Models module (in the Specification method list; these types of user interfaces are also available in STATISTICA General Linear Models, General Regression Models, General Partial Least Squares Models, General Discriminant Analysis Models, and ANOVA). This tab also provides options to select the appropriate Distribution function (for the response variable), and Link function. Refer to the Introductory Overview for additional details.
- Type of analysis
- Select the type of design that you want to perform from the Type of analysis list. Note that all types of designs can be specified via the General custom designs option. For more information on a particular
type of analysis, click on the links below. The links will display the discussion of the respective types of research designs in the context of the General Linear Model (GLM) module. You can analyze most of the types of designs discussed in the context of GLM with the GLZ module, except that 1)
within subject (repeated-measures) designs are not available, 2)
mixed-model ANOVA and ANCOVA designs for random effects are not available, and 3)
multivariate analyses are not allowed (i.e., only a single dependent (response) variable can be analyzed). More detailed discussions of these designs can also be found in the
Types of Analyses topic of GLM.
- Specification method
- The Specification method list presents the three alternative user-interfaces available in GLZ (as well as GLM, GRM, PLS, GDA, and ANOVA). You can choose among the three different user interfaces in the Specification method list only when General custom design is selected as the Type of analysis. When any other Type of analysis is selected, you must use the Quick specs dialog. See Specification methods in GLZ for further details.
- Distribution, Link functions
- Select the Distribution and Link functions for the dependent (response) variable in the respective lists. The link function in the generalized linear model specifies a nonlinear transformation of the predicted values so that the distribution of predicted values is one of several special members of the exponential family of distributions (Normal, Poisson, Gamma, Binomial, Multinomial, Ordinal multinomial, Tweedie, Negative Binomial, Beta). The link function is therefore used to model responses when a dependent variable is assumed to be nonlinearly related to the predictors. Various link functions (see McCullagh and Nelder, 1989) are commonly used, depending on the assumed distribution of the dependent variable (y) values. When you select a distribution in the Distribution list, the available choices in the Link functions list will change accordingly.
- Normal, Poisson, Gamma, Inverse normal, Tweedie, and Negative Binomial distributions
- When you select any of these Distributions, you will have the following choices in the Link functions list:
Log link: f(z) = log(z)
Power link: f(z) = za, for a given a
Identity link: f(z) = z
- Binomial, and Ordinal multinomial distributions
- When you select the Binomial, and Ordinal multinomial Distributions, you will have the following choices in the Link functions list:
Logit link: f(z) = log(z/(1-z))
Probit link: f(z) = invnorm(z) where invnorm is the inverse of the standard normal cumulative distribution function (see Distributions and their functions).
Log-Log link: f(z) = -log(-log(z))
C(omplementary)-Log-Log link: f(z) = log(-log(1-z))
- Multinomial distribution
- When you select the Multinomial Distribution, you will have the following choice in the Link functions list:
Generalized Logit link: f(z1|z2, ..., zc) = log(z1/(1-z1-...-zc)) where the model has c+1 categories.
For more information, refer to the Introductory Overview or the GLZ - Index.