persuasio

persuasio estimates and bounds persuasion effects in instrumental variable settings with binary outcomes. You provide the outcome, the treatment, and the instrument, tell persuasio which estimand you want (average or local persuasion rate), and it takes care of the bounds and inference. Based on Jun and Lee (2023) https://doi.org/10.1086/724114.

Installation

You can install the development version of persuasio from GitHub with:

# install.packages("pak")
pak::pak("xinrui33/persuasio")

Quick Example

library(persuasio)
## basic example code

# Average persuasion rate (APR): normal inference
persuasio(
  est     = "apr",
  varlist = c("voteddem_all", "readsome", "post"),
  data    = GKB,
  level   = 0.80,
  method  = "normal"
)
#> 
#> Average persuasion rate for binary outcomes, binary treatments and binary instruments
#> 
#> Outcome:    voteddem_all
#> Treatment:  readsome
#> Instrument: post
#> Model:      no_interaction
#> Method:     normal
#> Observations: 701
#> 
#> Estimates:
#>  Lower Bound Upper Bound CI Lower CI Upper
#>       0.0707      0.6343   0.0288   0.6611
#> 
#> Confidence level: 80%

# Local persuasion rate (LPR): bootstrap inference
persuasio(
  est     = "lpr",
  varlist = c("voteddem_all", "readsome", "post"),
  data    = GKB,
  level   = 0.80,
  method  = "bootstrap",
  nboot   = 1000
)
#> 
#> Local persuasion rate for binary outcomes, binary treatments and binary instruments 
#> 
#> Outcome:    voteddem_all
#> Treatment:  readsome
#> Instrument: post
#> Model:      no_interaction
#> Method:     bootstrap
#> Observations: 701
#> 
#> Estimates:
#>     LPR CI Lower CI Upper
#>  0.8067   0.0409   1.8671
#> 
#> Confidence level: 80%
#> Bootstrap replications: 1000

Learn more

See vignette("persuasio") for a full walkthrough including covariates, model specifications, and the relationship between estimands.