forsearch (version 6.4.0)

forsearch_glm: Create Statistics of Forward Search in a Generalized Linear Model Database

Description

Prepares summary statistics at each stage of forward search for subsequent plotting. Forward search is conducted in three steps: Step 1 to identify minimal set of observations to estimate unknown parameters, and Step 2 to add one observation at each stage such that observations in the set are best fitting at that stage. A preliminary step (Step 0) contains code for pre-processing of the data.

Usage

forsearch_glm(initial.sample=1000, response.cols, indep.cols, family,  
   formula=NULL, binomialrhs=NULL, formula.cont.rhs, data, 
   estimate.phi = TRUE, wiggle=1, skip.step1=NULL, unblinded=TRUE, 
   begin.diagnose=100, verbose=TRUE)

Value

LIST

Rows in stage

Observation numbers of rows included at each stage

Family

Family and link

Number of model parameters

Number of fixed effect parameters

Fixed parameter estimates

Matrix of parameter estimates at each stage

Residual deviance

Vector of deviances

Null deviance

Vector of null deviances

PhiHat

Vector of values of phi parameter

Deviance residuals and augments

Deviance residuals with indication of whether each is included in fit

AIC

Vector of AIC values

Leverage

Matrix of leverage of each observation at each stage

Call

Call to this function

Arguments

initial.sample

Number of random sets of observations in Step 1 of forward search

response.cols

Vector of column numbers (1 or 2) of responses and nonresponses (if binomial)

indep.cols

Column number(s) of independent variables

family

Error distribution and link

formula

Formula relating response to independent variables. Required except for family=binomial

binomialrhs

Quoted character.Right-hand side of formula. Required for family=binomial

formula.cont.rhs

Quoted character.Right-hand side of formula, omitting factor variables. Required for all families

data

Name of database

estimate.phi

TRUE causes phi to be estimated; FALSE causes phi to be set = 1

wiggle

Number multiplier to minimize arbitrary exchange of observations in step 2. Default is 1. Used only if independent variabls are all factors.

skip.step1

NULL, or vector of observation numbers to include at end of Step 1

unblinded

TRUE allows print of formula of analysis function

begin.diagnose

Numeric. Indicates where in code to begin printing diagnostics. 0 prints all; 100 prints none

verbose

TRUE causes function identifier to display before and after run

Author

William R. Fairweather

Details

Step 2 is determined by the results of Step 1, which itself is random. So, it is possible to reproduce the entire run by using the skip.step1 argument. Inner subgroups are produced by presence of categorical variables. Current version assumes independent variables are all continuous.

References

Atkinson, A and M Riani. Robust Diagnostic Regression Analysis, Springer, New York, 2000.

Examples

if (FALSE) {
# Train deaths (Atkinson and Riani, 2000) with Rolling Stock as a factor
Observation<-1:67
Month<-c(9,8,3,1,10,6,7,1,8,4,3,3,12,11,10,9,9,4,3,12,12,10,7,2,12,2,12,12,12,
    11,3,10,4,2,12,12,9,11,1,10,8,6,1,10,6,12,8,4,9,6,12,10,7,2,5,12,5,5,4,3,1,
    9,11,9,7,3,2)
Year<-c(97,96,96,95,94,94,91,91,90,89,89,89,88,88,87,86,86,86,86,84,84,84,84,84,
    83,83,82,81,81,80,80,79,79,79,78,78,77,76,76,75,75,75,75,74,74,73,73,73,72,
    72,71,71,71,71,70,69,69,69,69,69,69,68,67,67,67,67,67)
RollingStock<-c(2,2,3,2,1,1,1,1,2,3,1,1,1,2,1,2,1,3,2,2,1,2,2,3,1,2,1,1,2,3,1,
    1,1,1,1,1,1,3,3,2,3,1,2,3,1,1,1,3,3,1,3,3,1,1,1,2,1,1,2,1,1,1,1,1,1,1,1)
RollingStock <- as.factor(RollingStock)    
Traffic<-c(0.436,0.424,0.424,0.426,0.419,0.419,0.439,0.439,0.431,0.436,0.436,
    0.436,0.443,0.443,0.397,0.414,0.414,0.414,0.414,0.389,0.389,0.389,0.389,
    0.389,0.401,0.401,0.372,0.417,0.417,0.43,0.43,0.426,0.426,0.426,0.43,0.43,
    0.425,0.426,0.426,0.436,0.436,0.436,0.436,0.452,0.452,0.433,0.433,0.433,
    0.431,0.431,0.444,0.444,0.444,0.444,0.452,0.447,0.447,0.447,0.447,0.447,
    0.447,0.449,0.459,0.459,0.459,0.459,0.459)
Deaths<-c(7,1,1,1,5,2,4,2,1,1,2,5,35,1,4,1,2,1,1,3,1,3,13,2,1,1,1,4,1,2,1,5,7,
    1,1,3,2,1,2,1,2,6,1,1,1,10,5,1,1,6,3,1,2,1,2,1,1,6,2,2,4,2,49,1,7,5,9)
train2022 <- data.frame(Observation, Year, RollingStock, Traffic, Deaths)
test02<-forsearch_glm(initial.sample = 100, response.cols = 5, 
    indep.cols = 2:4, formula=Deaths~Year + RollingStock + Traffic,
    formula.cont.rhs="Year + Traffic", 
    family = poisson("log"), data = train2022, 
    estimate.phi = TRUE, skip.step1 = NULL, 
    unblinded = TRUE, begin.diagnose=100)
}