rwrmed: Regression-with-Residuals Analysis of Causal Mediation

In xiangzhou09/rwrmed: Regression-with-residuals (RWR) Analysis of Causal Mediation

Description

rwrmed is a function that implements the regression-with-residuals (RWR) approach to causal mediation, allowing for post-treatment confounding of the mediator-outcome relationship. Specifically, it fits user-specified mediator and outcome models with a set of residualized post-treatment confounders. It returns an object of class rwrmed, which can be used for effect decomposition via the decomp function.

Usage

rwrmed(treatment, pre_cov, zmodels, y_form, m_form, m_family = gaussian,
  weights, data)

Arguments

treatment	A character string indicating the name of the treatment variable, which can be either binary or continuous.
pre_cov	A character string vector indicating the names of the pretreatment covariates that are to be centered at their means. Thus factor-valued covariates should be pre-coded as dummy variables.
zmodels	A list of fitted lm or glm objects for the post-treatment confounders of the mediator-outcome relationship.
y_form	Formula for the outcome model.
m_form	Formula for the mediator model.
m_family	The family of the mediator model to be specified in glm.
weights	An optional vector of weights to be used in fitting the outcome and mediator models.
data	A data frame containing the variables in the model.

Value

An object of class rwrmed.

y_model	The fitted outcome model.
m_model	The fitted mediator model.
z_models	Fitted lm or glm models for the post-treatment confounders.
var_names	Names of the treatment variable, the mediator, the outcome, and the pretreatment and posttreatment covariates
data	The original data frame.
data_ed	The data frame with residualized pre- and post-treatment covariates
call	The matched call.

References

Wodtke, Geoffrey T. and Xiang Zhou. 2019. "Effect Decomposition in the Presence of Treatment-induced Confounding: A Regression-with-Residuals Approach."

Zhou, Xiang and Geoffrey T. Wodtke. 2019. "A Regression-with-Residuals Method for Estimating Controlled Direct Effects." Political Analysis.

VanderWeele, Tyler J, Stijn Vansteelandt and James M Robins. 2014. "Effect Decomposition in the Presence of an Exposure-induced Mediator-outcome Confounder." Epidemiology 25:300-306.

VanderWeele, Tyler J. 2014. "A Unification of Mediation and Interaction: a Four-Way Decomposition." Epidemiology 25(5):749-761.

See Also

decomp for effect decomposition based on a fitted rwrmed object.

Examples

## an example with a continuous mediator ##

# treatment and pre-treatment covariates
treatment <- "tone_eth"
pre_cov <- c("ppage", "female", "hs", "sc", "ba", "ppincimp")

# outcome and mediator models
y_form <- immigr ~ ppage + female + hs + sc + ba + ppincimp +
 tone_eth + emo + tone_eth * emo + p_harm
m_form <- emo ~ ppage + female + hs + sc + ba + ppincimp + tone_eth

# model for a post-treatment covariate
m1 <- lm(p_harm ~ ppage + female + hs + sc + ba + ppincimp + tone_eth,
 data = immigration)

# effect decomposition using RWR
fit1 <- rwrmed(treatment = treatment, pre_cov = pre_cov, zmodels = list(m1),
  y_form = y_form, m_form = m_form, data = immigration)

out1 <- decomp(fit1)

## an example with a binary mediator ##

# treatment and pre-treatment covariates
treatment <- "democ"
pre_cov <- c("ally", "trade", "h1", "i1", "p1", "e1", "r1", "male", "white", "age", "ed4")

# outcome and mediator models
y_form <- strike ~ ally + trade + h1 + i1 + p1 + e1 + r1 + male + white + age + ed4 + democ +
  immoral + democ * immoral + threatc + cost + successc
m_form <- immoral ~ ally + trade + h1 + i1 + p1 + e1 + r1 + male + white + age + ed4 + democ

# models for post-treatment covariates
m1 <- lm(threatc ~ ally + trade + h1 + i1 + p1 + e1 + r1 + male + white + age + ed4 + democ,
  data = peace)
m2 <- lm(cost ~ ally + trade + h1 + i1 + p1 + e1 + r1 + male + white + age + ed4 + democ,
  data = peace)
m3 <- lm(successc ~ ally + trade + h1 + i1 + p1 + e1 + r1 + male + white + age + ed4 + democ,
  data = peace)

# effect decomposition using RWR

fit2 <- rwrmed(treatment = treatment, pre_cov = pre_cov, zmodels = list(m1, m2, m3),
  y_form = y_form, m_form = m_form, m_family = binomial("logit"), data = peace)

summary(fit2$m_model)

out2 <- decomp(fit2)

xiangzhou09/rwrmed documentation built on May 27, 2019, 2:08 p.m.