undi: Test for Unjustified Disparate Impact

Description Usage Arguments Details Value

Within specified range of sensitivity parameters, find the ones that achieve minimum/maximum sensitivity results

optimsens(pol, range_q = c(0, 1), range_dp = c(0, log(2)), range_d0 = c(0,
  log(2)), range_d1 = c(0, log(2)), base_group = NULL,
  minority_groups = NULL, range_q_ratio = NULL, allow_sgv = FALSE,
  controls = NULL, fit_fn = "logit_coef", optim_control = list(),
  include_benchmark = FALSE, verbose = TRUE, debug = FALSE)

`pol`	object of class `policy`
`range_q`	2D vector specifying min/max value of p(u = 1 \| x)
`range_dp`	2D vector specifying min/max value of change in log-odds of treat = 1 if u = 1
`range_d0`	2D vector specifying min/max value of change in log-odds of response = 1 if treat = 0 and u = 1
`range_d1`	2D vector specifying min/max value of change in log-odds of response = 1 of treat = 1 and u = 1
`base_group`	(Optional) single group that acts as the pivot/base; by default, if the grouping variable is a factor, set to the first level, otherwise set to the first of sorted unique values
`minority_groups`	(Optional) groups to compare to the base group; by default, set to every unique value other than the base group
`range_q_ratio`	(Optional) 2D vector. If set, the minority and base values of q will not be allowed to vary independently, but instead will be constrained to vary by the given range of log odds. ie q_minority = inv.logit(logit(q_base) + u), where u is in range_q_ratio
`allow_sgv`	logical; whether to allow for subgroup validity; i.e., if `TRUE`, the delta parameters (`dp`, `d0`, `d1`) will be allowed to vary between base/minority groups, but if `FALSE`, a single value for each delta parameter will be used for each base/minority pair
`controls`	vector of legitimate controls to use; the ones specified within the policy object will be used if not specified
`fit_fn`	string indicating the rad estimation model/procedure used. `_coef` methods use models without interaction between risk and group, and return the coeficient on group membership. `_avg` methods will fit more flexible models (possibly with interactions), and compute average ratios across the population. (TODO: better documentation is expected)
`optim_control`	list of control parameters passed to `optim`
`include_benchmark`	logical; whether to include the two extreme benchmark test results (default: FALSE)
`verbose`	whether or not to print debug messages (0 = none, 1 = results only, 2 = everything)
`debug`	logical flag, if TRUE, returns a list of results and the expanded data frame used to fit model

If any of the range_ arguments are set to a single value (instead of a 2D vector), the corresponding paramter will be fixed and not explored for the optimization

a list-type object of class optimsens with the following elements

`results`	`tidy` dataframe of second-stage model coefficients after searching for min/max values across specified sensitivy parameter ranges, independently for each minority group
`optim`	nested data frame where the `$optim` column contains the optimization results
`base_case`	result from `compute_rad` on base policy with specified groups and controls
`base_group`	base group used in analysis