mcb.compare {mcb}

Comparisons of Model Confidence Bounds for Different Variable selection Methods

Description

This function is a supplement of the function mcb. It is used to compare different variable selection methods and would return all the MUCs on same canvas. A good variable selection method’s MUC will tend to arch towards the upper left corner.

Usage

mcb.compare(x, y, B=200, lambdas=NA, methods=NA, level=0.95, seed=122)

Arguments

x

input matrix presenting independent variables as in mcb.

y

response vector as in mcb.

B

number of bootstrap replicates to perform; Default value is 200.

lambdas

A vector of penalty tuning parameters for each variable selection method. The default values are the optimal choices for each selection method computed automatically.

methods

a vector including all variable selection methods the user wants to test and compare. The default value is c ('aLasso', 'Lasso', 'SCAD', 'MCP', 'stepwise', 'LAD', 'SQRT')

level

user-defined confidence level as in mcb; Default value is 0.95.

seed

Default value is 122.

Value

The mcb.compare method returns an object of class “mcb.compare” An object of class "mcb.compare " is a list containing at least the following components:

mcb

a list containing the bootstrap coverage rate and the corresponding model confidence bound for all user-given variable selection methods.

mucplot

plot of the model uncertainty curves for all variable selection methods and could be used to choose the best method.

mcbframe

a list containing all the information about MCBs for all variable selection methods under all available bootstrap coverage rates.

References

Li,Y., Luo,Y., Ferrari,D., Hu,X. and Qin,Y. (2019) Model Confidence Bounds for Variable Selection. Biometrics, 75:392-403.

Examples

data(Diabetes) # load data
x <- Diabetes[,c('S1', 'S2', 'S3', 'S4', 'S5')]
y <- Diabetes[,c('Y')]
x <- data.matrix(x)
y <- data.matrix(y)
result <- mcb.compare(x=x, y=y)
# plot of the model uncertainty curves for all variable selection methods
result$mucplot
# a list containing the bootstrap coverage rate and mcb which based on Lasso
result$mcb$Lasso
# a dataframe containing all the information about MCBs which based on Lasso
result$mcbframe$Lasso