Confidence intervals for qgcompmulti coefficients
Source:R/confint.qgcompmulti.R
confint.qgcompmulti.RdReturns confidence intervals for the marginal structural model (MSM)
coefficients reported by a fitted qgcompmulti object. Wald, percentile,
and basic bootstrap intervals are supported for single-fit objects. For
ratio estimands, intervals are computed on the MSM fitting scale and then
transformed to the user-facing ratio scale.
Usage
# S3 method for class 'qgcompmulti'
confint(object, parm = NULL, level = 0.95, method = NULL, ...)Arguments
- object
A fitted
qgcompmultiobject.- parm
Optional specification of which coefficients to include. May be
NULLfor all coefficients, an integer vector of coefficient positions, or a character vector of coefficient names.- level
Confidence level for the returned intervals. Must be strictly between 0 and 1.
- method
Optional interval method.
NULLuses the fitted object's stored default interval method. Supported values are"wald","percentile", and"basic". The basic method is the reverse-percentile bootstrap interval.- ...
Not used.
Value
A numeric matrix with one row per selected coefficient and two
columns giving the lower and upper confidence limits on the display scale
recorded in object$analysis$estimand_scale.
Details
Wald confidence intervals are based on the MSM coefficient estimates and the
stored bootstrap covariance matrix returned by vcov.qgcompmulti().
Percentile intervals use empirical quantiles of the stored bootstrap
coefficient draws. Basic intervals use the reverse-percentile construction,
reflecting the bootstrap quantiles around the full-sample estimate.
All interval calculations are carried out on the MSM fitting scale. When the
active estimand is "odds_ratio" or "rate_ratio", the returned limits are
exponentiated for display. The rows of the returned matrix align with
coef.qgcompmulti(), but the numeric scale may differ for ratio estimands.
Pooled multiple-imputation objects intentionally remain Wald-only.