cstarme {cstar} R Documentation

## Compute cstar for simulated marginal effects

### Description

cstarme computes the minimum actionable effect size of a simulated marginal effect under a kinked linear loss function with user-specified degree of loss.

### Usage

cstarme(sims, r)


### Arguments

 sims A vector of simulated marginal effects. See Details. r The degree of loss aversion; must be a non-negative number. This parameter maps to gamma in Esarey and Danneman (2014).

### Details

sims can be any vector of simulated marginal effects. For example, the change in predicted probability of an outcome as we change the level of a predictor in a logistic regression model.

### Value

A vector of expected values for the utility of acting on the evidence encapsulated by the simulated marginal effects, given the researcher's stated level of loss aversion.

### References

Esarey and Danneman (2014). A Quantitative Method for Substantive Robustness Assessment. Political Science Research and Methods.

### Examples

# create some logit data
x <- rnorm(50)
xb <- .5 + 2*x
pry <- exp(xb) / (1 + exp(xb))
y <- rbinom(50, 1, pry)

plot(x, y)

# run logistic regression

# extract variance-covariance matrix
VCV <- vcov(mod)

# simulate intercept and B1 from multivariate normal
require(MASS)
simulated_betas <- mvrnorm(n=50, mu=coefficients(mod), Sigma=VCV)

# calculate pr(y=1) for each simulated pair of (intercept, B1);
# do so at x=0 and x=2
pry_x0 <- apply(simulated_betas, 1, function(x){
exp(x + 0*x) / (1 + exp(x + 0*x))
})
pry_x2 <- apply(simulated_betas, 1, function(x){
exp(x + 2*x) / (1 + exp(x + 2*x))
})

# compute the simulated change in predicted probability
simulated_marginal_effects <- pry_x2 - pry_x0

# estimate the expected utility of accepting evidence
cstarme(simulated_marginal_effects, 2)


[Package cstar version 1.0 Index]