estimateEffects: Stabilized IPTW Estimators for Causal Effects Assuming...

In BarkleyBG/stabilizedinterference: Stabilized IPW Estimators for Causal Effects with Partial Interference

Description

This fits some of the IPW estimators introduced in Liu, Hudgens, and Becker-Dreps (2016) Biometrika. These estimators estimate causal effects in the presence of partial interference, with estimates of the asymptotic variance from standard M-estimation theory.

Usage

estimateEffects(data, formula, alphas, weight_type = c("HT", "Hajek1",
  "Hajek2", "HT_TV")[3], model_method = c("glm", "glmer", "oracle")[2],
  model_options = list(nAGQ = 5, family = "binomial"), ...)

Arguments

data	the dataframe. Will be coerced from "tbl_df" to data.frame.
formula	Multi-part formula: Outcome | Treatment ~ model_predictors | cluster_ID. This will be coerced to object of type Formula. When using model_method = "glmer", then the random intercept term is supplied in the style preferred by lme4's merMod: e.g., Outcome | Treatment ~ model_predictors + ( 1 | cluster_ID ) | cluster_ID.
alphas	the range of allocations or policies from 0 to 1. When model_method == "oracle" then model_options must be a list with named numeric vectors fixefs and var_comp. See prepareOracle. For mixed effect model, note that that random intercept's term in the modeling formula (e.g., ( 1 | cluster_ID ) ) must be omitted from formula. Arguments that can be passed through ... include: integrate_alphas. Not yet supported. verbose. Set to TRUE for more verbose messaging. Default FALSE. contrast_type. Not yet supported. keep_components. Set to TRUE for more verbose output. Default FALSE. target_grids. User can supply target estimands with makeTargetGrids. deriv_control. User can supply the deriv_control argument to m_estimate with setup_deriv_control.
weight_type	Estimators as presented in Liu, Hudgens, and Becker-Dreps (2016) Biometrika. Select "HT" for unstabilized weights. Select "Hajek1" or "Hajek2" for stabilized weights. Select "HT_TV" for the estimators presented in Tchetgen Tchetgen and VanderWeele (2012) SMMR and Perez-Heydrich et al. (2014) Biometrics, which in general target estimands different from those in Liu, Hudgens, and Becker-Dreps (2016) Biometrika.
model_method	"glm" for logistic, "glmer" for logistic with single random intercept. "oracle" supported; see interference for syntax.
model_options	passed to glmer or perhaps glm(in future).
...	additional args. See details.

Details

Note that these estimators estimate different causal estimands than those in Tchetgen Tchetgen and VanderWeele (2012) SMMR that were applied in Perez-Heydrich et al. (2014) Biometrics and implemented with interference by Saul and Hudgens (2017) JSS.

BarkleyBG/stabilizedinterference documentation built on May 23, 2019, 8:37 a.m.