estimate_ATE_FIRC: Estimate the ATE using Random-Intercept, Random-Coefficient...

In lmiratrix/blkvar: ATE and Treatment Variation Estimation for Blocked and Multisite RCTs

Estimate the ATE using Random-Intercept, Random-Coefficient (RIRC) Model

Description

This method fits the FIRC model to estimate (1) ATE across sites and (2) cross site treatment variation.

Usage

estimate_ATE_FIRC(
  Yobs,
  Z,
  B,
  siteID = NULL,
  control_formula = NULL,
  data = NULL,
  include_testing = FALSE,
  REML = !include_testing,
  anova = FALSE,
  pool = FALSE,
  keep_EB_estimates = TRUE
)

Arguments

Yobs	Name of outcome variable (assumed to exist in data)
Z	vector of assignment indicators (1==treated)
B	Name of the block indicator.
siteID	If not null, name of siteID that has randomization blocks
control_formula	The control_formula argument must be of the form ~ X1 + X2 + ... + XN. (nothing on left hand side of ~)
data	Dataframe with all needed variables.
include_testing	Logical Include likelihood ratio test for cross-site treatment variation.
REML	Logical, Restricted maximum likelihood or maximum likelihood estimation. Default of TRUE if include_testing = FALSE and FALSE otherwise. Note that FIRC can fail badly in estimating cross-site variation when REML=FALSE.
anova	Use the anova() method to do the test for significance between the models. FALSE means do the modified chi-squared test.
pool	TRUE means tx and co have same reBual variance. FALSE gives seperate estimates for each (recommended, default).
keep_EB_estimates	TRUE means returned object has EB estimates. FALSE means do not keep them.

Details

Warning: If you want to test for cross-site variation you need REML = FALSE to allow for likelihood ratio testing. If you want to estimate this variation, REML = TRUE is recommended as the ML estimate tends to be wildly biased towards 0 when modest amounts of cross-site variation are present.

Acknowledgement: Unpooled version taken and adapted from Catherine's weiss.tau() method.

lmiratrix/blkvar documentation built on Nov. 18, 2024, 1:27 p.m.