variable_analysis: Determine Significance Level of Individual Variables
In Laksafoss/ICPSurv: Invariant Causal Prediction for Survival Data

Description Usage Arguments Details Value See Also Examples

View source: R/ICP-variable_analysis.R

The variable_analysis function takes an ICP object and outputs evidence for each individual variable being a direct cause.

1	variable_analysis(x, gof = 0.1)

`x`	An ICP object
`gof`	If no set of variables (including the empty set) leads to a p-value larger than the goodness-of-fit cutoff `gof`, the whole model will be rejected. If the model is correct, this will happen with a probability of gof and this option protects again making statements when the model is obviously not suitable for the data.

The function variable_analysis takes an ICP object and a minimum goodness-of-fit cutoff gof and outputs the evidence for each individual variable being a direct cause. To conduct a variable analysis the inputted ICP object should be the result of a call to the ICP function with option fullAnalysis = TRUE. This ensures that variable_analysis has access to the p-values for all the hypothses (H_0,S) (see ICP for an explanation of the hypotheses).

If the needed gof level is reached the significance is calculated for each variable Xⁱ by

p_i = max{ p-value for H_0,S ∣ Xⁱ is not part of X^S }.

The gof parameter protects against making statements when the model is obviously not suitable for the data. If no model reaches the threshold gof significance level, i.e. the p-values for (H_0,S) are all smaller then gof, we report that there is no evidence for individual variables, as there is no evidence for an invariant set.

The variable_analysis function is also used internally in the ICP function itself if ICP is called with level specified and fullAnalysis = TRUE.

variable_analysis returns a table with class "variable_analysis" where each row reports the significance of the corresponding variable.

The variable_analysis function is also used internally in the ICP function itself if ICP is called with fullAnalysis = TRUE.

model_analysis is another function for summarizing ICP objects.

n <- 100
E <- sample(5L, n, replace = TRUE)
X <- data.frame(X1 = rnorm(n, E, 1), X2 = rnorm(n, 3, 1))
Y <- rnorm(n, X$X1, 1)

obj <- ICP(Y, X, E, level = 0.05, fullAnalysis = TRUE)
variable_analysis(obj, gof = 0.1)

# If we make extreem requirements for gof
# all "p-values" will be taken to be 1,
# i.e. no variable is a significant causal predictor
variable_analysis(obj, gof = 0.9)