This is the stepp model for data arising from the following Generalized Linear Models: 1. gaussian with identity link, 2. binomial with logit link, and 3. Poisson with log link.
One can specify additional covariates in the model using model.matrix in R.
1.1
class
Objects from the Class
Objects can be created by calls of the form new("stmodelGLM", ...) or by
the construction function stmodel.GLM.
Slots
coltrt:: Object of class "numeric"
the treatment variable
colY:: Object of class "numeric"
a vector containing the outcome
trts:: Object of class "numeric"
a vector containing the codes for the 2 treatment groups, first and second treatment groups, respectively
MM:: Object of class "ANY"
a model matrix for extra adjustment covariates; default is NULL
currently, stepp can only support covariates with numeric values, no logical values or factors allowed
these values need to be converted to numeric with some encoding schemes
glm:: Object of class "character"
the glm to be used for analysis: "gaussian", "binomial", "poisson"
link:: Object of class "character"
the link function; reserved for future use
Extends
Class "stmodel", directly.
Methods
estimate: signature(.Object = "stmodelGLM"):
estimate the effect in absolute and relative scale of the overall and each subpopulation
print: signature(.Object = "stmodelGLM"):
print the estimate, covariance matrices and statistics
test: signature(.Object = "stmodelGLM"):
perform the permutation tests or GEE and obtain various statistics
Returns
The new method returns the stmodelGLM object.
The estimate method returns a list with the following fields:
model: the stepp model - "GLMGe" - Gaussian model, "GLMBe" - binomial model, and "GLMPe" - Poisson model
sObs1: a vector of effect estimates of all subpopulations based on the first treatment
sSE1: a vector of standard errors of effect estimates of all subpopulations based on the first treatment
oObs1: effect estimate of the entire population based on the first treatment
oSE1: the standard error of the effect estimate of the entire population based on the first treatment
sObs2: a vector of effect estimates of all subpopulations based on the first treatment
sSE2: a vector of standard errors of effect estimates of all subpopulations based on the first treatment
oObs2: effect estimate of the entire population based on the first treatment
oSE2: the standard error of the effect estimate of the entire population based on the first treatment
sglmw: Wald's statistics for the effect estimate differences between the two treatments
RD: a vector of effect estimates difference of all subpopulations between the two treatments
RDSE: a vector of the standard error effect estimates difference of all subpopulations between the two treatments
ORD: overall difference of effect estimates of the entire population between the two treatments
ORDSE: the standard error of the overall difference of the effect estimates of the entire population between the two treatments
logR: a vector of log ratio of effect estimates of all subpopulations between the two treatments
logRSE: a vector of standard error of log ratio of effect estimates of all subpopulations between the two treatments
ologR: log ratio of effect estimates of the entire population between the two treatments
ologRSE: the standard error of the log ratio of effect estimates of the entire population between the two treatments
sglmlogrw: Wald's statistics for the log ratio of effect estimates between the two treatments
The test method returns a list with the following fields: - model: the stepp model - "GLMt"
sigma: the covariance matrix for subpopulations based on effect differences
hasigma: the homogeneous association covariance matrix for subpopulations based on effect differences
HRsigma: the covariance matrix for the subpopulations based on hazard ratio
haHRsigma: the homogeneous association covariance matrix for subpopulations based on hazard ratio
pvalue: the supremum pvalue based on effect difference
chi2pvalue: the chisquare pvalue based on effect difference
hapvalue: the homogeneous association pvalue based on effect difference
HRpvalue: the supremum pvalue based on hazard ratio
haHRpvalue: the homogeneous association pvalue based on hazard ratio