Skip to contents

|bias| <= c ||s||_2 by Cauchy-Schwarz, attained at gamma = c s / ||s||, and the conservative interval is estimate +/- (c ||s|| + z_{1-alpha/2} se).

Usage

Misspecbd(estimate, sensitivity, c, se, conf = 0.95)

Arguments

estimate

Point estimate under the baseline model.

sensitivity

Derivative of the estimate w.r.t. the perturbation.

c

Radius of the misspecification neighbourhood.

se

Standard error under the baseline model.

conf

Nominal confidence level.

Value

List with bias, lower, upper, halfwidth, worstgamma, normsens, z, c.

References

Bias bound by Cauchy-Schwarz; the bias-aware interval is Armstrong and Kolesar (2021), Quantitative Economics 12(1), 77-108, Sect. 2. It is sometimes attributed to Andrews and Kasy (2019), which is about publication bias and is not the source.

Examples

Misspecbd(estimate = 0.5, sensitivity = c(0.1, 0.2), c = 0.3, se = 0.1)
#> $bias
#> [1] 0.06708204
#> 
#> $lower
#> [1] 0.2369216
#> 
#> $upper
#> [1] 0.7630784
#> 
#> $halfwidth
#> [1] 0.2630784
#> 
#> $worstgamma
#> [1] 0.1341641 0.2683282
#> 
#> $normsens
#> [1] 0.2236068
#> 
#> $z
#> [1] 1.959964
#> 
#> $c
#> [1] 0.3
#> 
#> $method
#> [1] "Conservative bias-aware interval under local misspecification"
#> 
#> attr(,"class")
#> [1] "morie_rich_result" "list"