Confintfit

Confintfit is a statistical concept and, in some software ecosystems, a function or procedure that produces confidence intervals for quantities derived from a fitted model. It is used to quantify the uncertainty around parameter estimates or predicted values generated by models such as linear regression, generalized linear models, nonlinear models, or survival models. The aim is to provide a plausible range within which the true quantity is expected to lie with a specified level of confidence.

Methods used to construct confintfit intervals vary and may include:

- Profile likelihood: intervals derived from the likelihood function, often providing accurate coverage for parameters in nonlinear or constrained problems.

- Delta method: intervals obtained from approximating the distribution of a function of estimated parameters using a Taylor expansion.

- Bootstrap: resampling-based intervals that reflect empirical variability, including percentile or bias-corrected and accelerated (BCa) approaches.

Input and output

Confintfit typically operates on a fitted model object, with options to specify which parameters or predicted quantities to target, the desired confidence level (e.g., 95%), and, for predictions, a set of newdata points. The function returns a table or data structure containing point estimates (for parameters or predictions) along with lower and upper bounds of the confidence interval for each quantity.

Usage considerations

The exact behavior of confintfit depends on the software package. It may be named confintfit or exposed as a method of a model object. Profile-based intervals can be more accurate in small samples but are computationally intensive; bootstrap intervals depend on the resampling scheme and number of replications. Users should be mindful of model correctness and the interval interpretation corresponding to the chosen method.

See also: confint, predict, profile, bootstrap, confidence interval.