cvParetoElnet: cvParetoElnet

In Diversity-ParetoOptimal/ParetoR: Estimates Pareto-Optimal solution for diversity hiring

cvParetoElnet

Description

Cross-validation procedure to choose alpha and lambda values for regularized tradeoff curve algorithm

Usage

cvParetoElnet(
  data,
  prop,
  sr,
  k = 5,
  rep = 1,
  Spac = 21,
  alpha.grid = 0,
  lambda.grid = 0
)

Arguments

data	Calibration data: a dataframe with first p columns as standardized predictor scores, the (p+1)th column as standardized job performance ratings, and the (p+2)th column (rightmost column) as a race dummy; all data should be numeric
prop	Proportion of minority applicants in total applicant pool (#minority applicants/#total applicants)
sr	Selection ratio
k	Number of folds, default is 5
rep	Number of repeats if doing repeated k-fold CV
Spac	Number of Pareto-optimal solutions, default is 21
alpha.grid	Vector of alpha values to try
lambda.grid	Vector of lambda values to try

Value

A list containing the following:

weights

Predictor weights obtained based on best alpha and lambda values

alpha

Best alpha parameter value

lambda

Best lambda parameter value

Examples

# Calibration data
# (1) Calibration data
# Format: Predictor_1, ..., Predictor_n, Job Performance Validity, Race dummy variable
# (e.g., 0-minority; 1-majority)
# All data should be numeric
# sample_data
# (3) Alpha values to try
alpha.grid <- seq(0, 1, length = 2)
# (4) Lambda values to try
lambda.grid <- 10^seq(1, -2, length = 4)
# (5) Proportion of minority applicants (prop) = (# of minority applicants)/(total # of applicants)
prop <- 1/4
# (6) Selection ratio (sr) = (# of selected applicants)/(total # of applicants)
sr <- 0.10
# (7) Spac = number of Pareto points
Spac <- 21
# Fit Regularized Pareto-optimal model with parameter selection via cross-validation
## May take a while
out <- cvParetoElnet(data = sample_data, prop = prop, sr = sr, Spac = Spac,
                     k = 5, rep = 1,
                     lambda.grid = lambda.grid, alpha.grid = alpha.grid)

Diversity-ParetoOptimal/ParetoR documentation built on Feb. 9, 2024, 1:06 a.m.