Konfound robustness for a coefficient (native konfound) — morie_sensitivity

Native konfound::pkonfound (two tails, null 0): the threshold is \(t^* se\) with \(t^*\) on \(n - n_{cov} - 2\) df; the percent bias to invalidate is \(100 (1 - t^* se / \hat\beta)\) (to sustain when not significant) and RIR the corresponding number of cases to replace; the impact threshold of a confounding variable is \((r - r^*) / (1 \pm |r^*|)\) with \(r = t / \sqrt{t^2 + df}\). Pairs with morie_sensitivity_omitted_var_bias. Mirrors konfound in the Python arm.

Usage

morie_sensitivity_konfound(
  estimate,
  se,
  n,
  n_covariates = 0L,
  alpha = 0.05,
  ...
)

Arguments

estimate: Treatment-coefficient estimate.
se: Standard error of estimate.
n: Number of observations.
n_covariates: Number of covariates in the model (excluding the intercept and the treatment). Default 0.
alpha: Significance level. Default 0.05.
...: Unused.

Value

A list of class morie_sensitivity_konfound with percent_bias_to_invalidate, rir, impact_threshold_confounder and beta_threshold.

References

Frank, K. A., Maroulis, S. J., Duong, M. Q., & Kelcey, B. M. (2013). What would it take to change an inference? Educational Evaluation and Policy Analysis, 35(4), 437–460.

Examples

str(morie_sensitivity_konfound(0.5, 0.15, 200), max.level = 1)
#> List of 8
#>  $ estimate                   : num 0.5
#>  $ se                         : num 0.15
#>  $ n                          : num 200
#>  $ percent_bias_to_invalidate : num 40.8
#>  $ rir                        : num 82
#>  $ impact_threshold_confounder: num 0.107
#>  $ beta_threshold             : num 0.296
#>  $ method                     : chr "pkonfound (native)"
#>  - attr(*, "class")= chr [1:2] "morie_sensitivity_konfound" "list"