Robustness of inferences to differential attrition (beta)

Quantifies what would need to be true in the attritted (missing) data to nullify an inference based on the observed data, using two complementary approaches: a nonparametric decomposition of the combined-sample effect, and a correlation-based framework adapted from Frank and Min (2007).

Usage

robust_attrition(
  std_err,
  ntreatob,
  ncontrolob,
  ntreattot,
  ncontroltot,
  yobt,
  yobc,
  syob,
  n_covariates,
  R2,
  R2xz = NULL,
  R2yz = NULL,
  alpha = 0.05,
  pad_frac = 0.15,
  verbose = FALSE,
  to_return = "print"
)

Arguments

std_err

standard error of the observed treatment effect.

ntreatob

number of observed treatment cases.

ncontrolob

number of observed control cases.

ntreattot

intended total treatment cases.

ncontroltot

intended total control cases.

yobt

observed treatment-group mean.

yobc

observed control-group mean.

syob

observed outcome standard deviation.

n_covariates

number of covariates in the observed-data model.

R2

unadjusted R-squared of the observed-data model (0 <= R2 < 1).

R2xz

optional predictor-covariate fit; computed internally from other inputs when NULL (default).

R2yz

optional outcome-covariate fit; computed internally when NULL (default).

alpha

significance level (default 0.05).

pad_frac

half-width of the plotting window for to_return = "plot", as a fraction of the observed pooled mean. Defaults to 0.15 (a +/- 15 values draw a longer line.

verbose

if TRUE, the printed output includes a short description of what each block computes and inline notes about internally-computed defaults. Defaults to FALSE.

to_return

one of "print" (default) to display output and return the result list invisibly, "raw_output" to return the full list, or "plot" to return a ggplot object showing the effect required in the missing data to nullify the inference across assumed missing-data means.

Value

A list with components inputs, nonpar, correlation_based, and derived. Returned invisibly when to_return = "print". When to_return = "plot", a ggplot object is returned instead.

Details

This is a beta (development) version. Calculations and output are under review.

Examples

if (FALSE) { # \dontrun{
robust_attrition(
  std_err = 2.19,
  ntreatob = 1817, ncontrolob = 1981,
  ntreattot = 2028, ncontroltot = 2311,
  yobt = 54.72, yobc = 49.90, syob = 29,
  n_covariates = 1, R2 = 0.01, R2xz = 0.479^2
)
} # }