calc_pppd: Partial prior predictive distribution

In carlislerainey/separation: Using informative priors to deal with separation in logistic regression models

Description

calc_pppd returns the partial prior predictive distribution for a logistic regression.

Usage

calc_pppd(formula, data, prior_sims, sep_var_name, treat_one_low = FALSE,
  X_pred_list = NULL, prior_label = NULL)

Arguments

formula	A formula for a logistic regression model.
data	A data frame.
prior_sims	Simulations from the desired prior distribution.
sep_var_name	The name of the separating variable.
treat_one_low	Treat sep_var_at as the low value when computing QIs. Defaults to TRUE. If FALSE, the sep_var_at is treated as the high value.
X_pred_list	A named list of values at which to set variables. The function set_at_median() facilitates creating this list.
prior_label	The name of the prior used.

Details

Researchers can use this function to convert simulations from the prior distribution to simulations from the partial prior predictive distribution See Rainey (2014) for the details.

References

Rainey, Carlisle. 2014. "Dealing with Separation in Logistic Regression Model." Working paper. Available at http://crain.co/papers/separation.pdf.

Examples

# load data from Barrilleaux and Rainey (2014)
data(politics_and_need)

# prior simulations
normal1 <- rnorm(10000, sd = 1)

# formula
f <- oppose_expansion ~ gop_governor + percent_favorable_aca + percent_uninsured

# convert prior simulations of the coefficient to simulations of the
#   quantitie of interest
pppd <- calc_pppd(f, data = politics_and_need, prior_sims = normal1,
                  sep_var_name = "gop_governor", prior_label = "N(0, 1)")

# plot and print summaries of the PPPD
print(pppd)

par(mfrow = c(1, 1))
plot(pppd)  # predicted probability is the default
plot(pppd, qi_name = "rr")  # risk-ratio
plot(pppd, qi_name = "fd")  # first-difference

carlislerainey/separation documentation built on May 13, 2019, 12:45 p.m.