Regression and generalized linear models


GenStat provides directives for carrying out linear and nonlinear regression, also generalized linear, generalized additive and generalized nonlinear models. They are designed to allow easy comparison between models, and comparison between groups of data (specified as factors). The directives for nonlinear regression can also be used for general optimization. There are three preliminary directives for defining the form of model to be fitted, of which the MODEL directive must always be given first:


MODEL
defines the response variate(s) and the type of model to be fitted

TERMS
specifies a maximal model, containing all terms to be used in subsequent regression models

RCYCLE
controls iterative fitting of generalized linear models, generalized additive models and nonlinear models, and specifies parameters and bounds for nonlinear models


Separate directives carry out the fitting of the various types of model:


FIT
fits a linear model, a generalized linear model, a generalized additive model, or a generalized nonlinear model

FITCURVE
fits a standard nonlinear regression model

FITNONLINEAR
fits a user-defined nonlinear regression model or optimizes a scalar function


Further directives are provided to allow sequential modification of the set of explanatory variables:


ADD
adds extra terms to any type of regression model

DROP
drops terms from any type of regression model

SWITCH
adds terms to, or drops them from, any type of regression model

TRY
displays results of single-term changes to a linear or generalized linear model

STEP
selects terms to include in or exclude from a linear or generalized linear model


The results of fitting the models can be displayed or stored in data structures:


RDISPLAY
displays the fit of any type of regression model

RKEEP
stores the results from any type of regression model

RKESTIMATES
saves estimates and other information about individual terms in a regression analysis

PREDICT
forms predictions from a linear or generalized linear model

RFUNCTION
estimates functions of parameters of a regression model


Procedure in the Library relevant to regression analysis include:


RCHECK
checks the fit of a regression model

RGRAPH
draws a graph to display the fit of a regression model

RDESTIMATES
plots one- or two-way tables of regression estimates

RPERMTEST
does random permutation and exact tests for regression or generalized-linear-model analyses

RPOWER
calculates the power (probability of detection) for regression models

RCOMPARISONS
calculates comparison contrasts amongst the levels of a factor classifying a table of regression means

RTCOMPARISONS
calculates comparison contrasts within a multi-way table of means

RWALD
calculates Wald and F tests for dropping terms from a regression

SED2ESE
calculates effective standard errors that give good approximate sed's

SEDLSI
calculates least significant intervals

LSIPLOT
plots least significant intervals

BRDISPLAY
displays a regression tree

BREGRESSION
constructs a regression tree

BRPREDICT
makes predictions using a regression tree

BRVALUES
forms values for nodes of a regression tree

DILUTION
calculates Most Probable Numbers from dilution series data

EXTRABINOMIAL
fits models to overdispersed proportions

FIELLER
calculates effective doses or relative potencies

FITINDIVIDUALLY
fits regression models one term at a time (useful for obtaining an accumulated analysis of deviance table containing the contributions of individual terms in a generalized linear model)

FITMULTINOMIAL
fits generalized linear models with multinomial distribution

GEE
fits models to longitudinal data by generalized estimating equations

GLM
analyses non-standard generalized linear models

GLMM
fits a generalized linear mixed model

HGANALYSE
analyses data using a hierarchical generalized linear model (HGLM) or a double hierarchical generalized linear model (DHGLM)

HGDISPLAY
displays results from an HGLM or DHGLM

HGDRANDOMMODEL
adds random terms into the dispersion models of an HGLM, so that the whole model becomes a DHGLM

HGFIXEDMODEL
defines the fixed model for an HGLM or DHGLM

HGGRAPH
draws a graph to display the fit of an HGLM or DHGLM analysis

HGKEEP
saves information from an HGLM or DHGLM analysis

HGNONLINEAR
defines nonlinear parameters for the fixed model of an HGLM

HGPLOT
produces model-checking plots for an HGLM or DHGLM

HGPREDICT
forms predictions from an HGLM or DHGLM analysis

HGRANDOMMODEL
defines the random model for an HGLM

HGSTATUS
displays the current HGLM model definitions

HGWALD
prints or saves Wald tests for fixed terms in an HGLM

IFUNCTION
estimates implicit and/or explicit functions of parameters

MAREGRESSION
does regressions for single-channel microarray data

MINIMIZE
finds the minimum of a function calculated by a procedure

MIN1DIMENSION
finds the minimum of a function in one dimension

NLAR1
fits curves with an AR1 or a power-distance correlation model

PAIRTEST
performs t-tests for pairwise differences

PPAIR
displays results of t-tests for pairwise differences in compact diagrams

PROBITANALYSIS
fits probit models allowing for natural mortality and immunity

R0INFLATED
fits zero-inflated regression models to count data with excess zeros

R0KEEP
saves information from models fitted by R0INFLATED

RAR1
fits regressions with an AR1 or a power-distance correlation model

RCIRCULAR
does circular regression of mean direction for an angular response

RJOINT
does modified joint regression analysis for variety-by-environment data

RQLINEAR
fits and plots quantile regressions for linear models

RQSMOOTH
fits and plots quantile regressions for loess or spline models

RLFUNCTIONAL
fits a linear functional relationship model

RMGLM
fits a model where different units follow different generalized linear models

RNEGBINOMIAL
fits a negative binomial GLM estimating the aggregation parameter

RNONNEGATIVE
fits a generalized linear model with nonnegativity constraints (synonym FITNONNEGATIVE)

RPAIR
gives t-tests for all pairwise differences of means from linear or generalized linear models

RPARALLEL
carries out analysis of parallelism for nonlinear functions (synonym FITPARALLEL)

RQUADRATIC
fits a quadratic surface and estimates its stationary point

RSCHNUTE
fits a general four-parameter growth model to a non-decreasing response variate (synonym FITSCHNUTE)

RSCREEN
performs screening tests for generalized or multivariate linear models

RSEARCH
searches through models for a regression or generalized linear model (with methods including all-subsets, forward and backward stepwise regression)

R2LINES
fits two-straight-line (broken-stick) models to data

SIMPLEX
searches for the minimum of a function using the Nelder-Mead algorithm

SVGLM
fits generalized linear models to survey data

WADLEY
fits models for Wadley's problem, allowing alternative links and errors

XOCATEGORIES
performs analyses of categorical data from crossover trials

YTRANSFORM
estimates the parameter lambda of a single parameter transformation