plausible_predictor_test: Internal Test for plausible predictors
In Laksafoss/ICPSurv: Invariant Causal Prediction for Survival Data

Description Usage Arguments Details Value See Also

View source: R/ICP-plausible_predictor_test.R

plausible_predictor_test is a generic function used to test whether a set of predictors might be plausible causal predictors.

plausible_predictor_test(method, Y, X, ...)

## Default S3 method:
plausible_predictor_test(method, Y, X, ...)

## S3 method for class 'TimeVar'
plausible_predictor_test(method, Y, X, level, Bonferroni = TRUE, ...)

## S3 method for class 'EnvirRel'
plausible_predictor_test(method, Y, X, E, ...)

## S3 method for class 'CR'
plausible_predictor_test(
  method,
  Y,
  X,
  E,
  level,
  fullAnalysis,
  Bonferroni = TRUE,
  ...
)

`method`	a method object describing the model class to use in the fitting procedure and the testing method. When used in the `ICP` function the method object is inherited from the `ICP` function.
`Y`	a vector or `Surv` object describing the target variable. The `Y` will be passed to the `fit_model` function for fitting, and it is therefor important that the class of `Y` is understood by the regression method used in `fit_model`. So if `method` specifies that a cox regression is to be used, then `Y` must be a `Surv` object.
`X`	a matrix, vector or data frame describing the covariates.
`...`	additional arguments to be passed to lower level functions.
`level`	the alpha level of testing, but it is only relevant if the method is iterative and `fullAnalysis` is `FALSE`.
`Bonferroni`	if `TRUE` Bonferroni correcting will be used if relevant.
`E`	a vector describing the environmants
`fullAnalysis`	if `TRUE` a p-value must be retured. If `FALSE` the method is 'allowed' to only return 0/1 for hypothesis rejected/accepted.

The plausible_predictor_test is meant to be used in the wrapper function ICP, but can also be used on its own as a simple hypothesis testing function. The method object method, which is created by the method_obj function dictates the plausible_predictor_test method. As this function is generic which means that new testing methods can easily be added for new classes.

For the plausible_predictor_test to work correctly in the ICP wrapper it must be a test of the null hypothesis

H_{0,S}: S is an invariant set w.r.t. (X,Y)

where an invariant set is defined as a set of indicies S such that

(Y^e | X^e_S) = (Y^f | X^f_S)

in distribution for all environments e,f. If the data is time dependent we formulate an analog invariance statement for all time points t and s. For more detailes and discussion of this hypothesis and invariance see the references.

As the plausible_predictor_test is a hypothesis test it must return a p value, that is a number between 0 and 1.

Returns a p-value, i.e. a number between 0 and 1, to be used in the ICP function.

ICP for the full wrapper function.

Laksafoss/ICPSurv documentation built on Feb. 26, 2020, 11:32 a.m.