| Type: | Package |
| Title: | Model Averaged Double Robust Estimation |
| Version: | 1.0.0 |
| Author: | Matthew Cefalu |
| Maintainer: | Matthew Cefalu <Matthew_Cefalu@rand.org> |
| Description: | Estimates average treatment effects using model average double robust (MA-DR) estimation. The MA-DR estimator is defined as weighted average of double robust estimators, where each double robust estimator corresponds to a specific choice of the outcome model and the propensity score model. The MA-DR estimator extend the desirable double robustness property by achieving consistency under the much weaker assumption that either the true propensity score model or the true outcome model be within a specified, possibly large, class of models. |
| License: | GPL-3 |
| LazyData: | TRUE |
| RoxygenNote: | 5.0.1 |
| NeedsCompilation: | no |
| Packaged: | 2016-09-03 02:42:56 UTC; mcefalu |
| Repository: | CRAN |
| Date/Publication: | 2016-09-05 20:23:43 |
Calculate model probabilities for the outcome models using a pseudo-MC3 algorithm
Description
This function uses a pseudo-MC3 algorithm to search the outcome model space.
Usage
OM.MA(Y, X, U, W = NULL, M = 1000, alpha = NULL, binary = F)
Arguments
Y |
vector of the outcome |
X |
vector of the treatment (0/1) |
U |
matrix of covariates to be considered for inclusion/exclusion |
W |
matrix of covariates that will be included in all models (optional) |
M |
the number of MCMC iteration |
alpha |
vector of inclusion indicators (which columns of U) to start MCMC algorithm (optional) |
binary |
indicator if the outcome is binary (optional) |
Value
A list. The list contains the following named components:
dict |
a list that contains the BIC, predicted values, and estimated treatment effect from each outcome model |
alpha |
the last model visited by the algorithm |
out.table |
a matrix that contains the BIC and estimated treatment effect from each outcome model |
Enumerates all possible outcome models (linear terms only)
Description
This function enumerates and fits all possible outcome models
Usage
OM.MA.enumerate(Y, X, U, W = NULL)
Arguments
Y |
vector of the outcome |
X |
vector of the treatment indicator (0/1) |
U |
matrix of covariates to be considered for inclusion/exclusion |
W |
matrix of covariates that will be included in all models (optional) |
Value
A list. The listcontains the following named components:
dict |
a list that contains the BIC, predicted values, and estimated treatment effect from each outcome model |
out.table |
a matrix that contains the BIC and estimated treatment effect from each outcome model |
Calculate model probabilities for the propensity score model using a pseudo-MC3 algorithm
Description
This function uses a pseudo-MC3 algorithm to search the propensity score model space.
Usage
PS.MA(X, U, W = NULL, M = 1000, alpha = NULL, master.index = NULL,
master.dict = list())
Arguments
X |
vector of the treatment (0/1) |
U |
matrix of covariates to be considered for inclusion/exclusion |
W |
matrix of covariates that will be included in all models (optional) |
M |
the number of MCMC iteration |
alpha |
vector of inclusion indicators (which columns of U) to start MCMC algorithm (optional) |
master.index |
indexes which columns of U should be considered for inclusion in the propensity score model (optional) |
master.dict |
list containing information from previous propensity score model fits (optional) |
Value
A list. The list contains the following named components:
dict |
a list that contains the BIC and estimated propensity scores from propensity score models |
alpha |
the last model visited by the algorithm |
out.table |
a matrix that contains the BIC from each propensity score model |
Enumerates all possible propensity score models (linear terms only)
Description
This function enumerates and fits all possible propensity score models
Usage
PS.MA.enumerate(X, U, W = NULL)
Arguments
X |
vector of the treatment indicator (0/1) |
U |
matrix of covariates to be considered for inclusion/exclusion |
W |
matrix of covariates that will be included in all models (optional) |
Value
A list. The list contains the following named components:
dict |
a list that contains the BIC and estimated propensity scores from propensity score models |
out.table |
a matrix that contains the BIC from each propensity score model |
Worker function that fits propensity score models
Description
This function fits propensity score models and saves necessary information
Usage
add.to.dictionary(X, U, W, alpha)
Arguments
X |
vector of the treatment (0/1) |
U |
matrix of covariates to be considered for inclusion/exclusion |
W |
matrix of covariates that will be included in all models (optional) |
alpha |
vector of inclusion indicators (which columns of U) to included in the propensity score model |
Value
A list. The list contains the following named components:
out |
a list that contains the BIC and estimated propensity scores from propensity score models |
Worker function that fits outcome models
Description
This function fits outcome models and saves necessary information
Usage
add.to.dictionary.outcome(Y, X, U, W, alpha, binary = F)
Arguments
Y |
vector of the outcome |
X |
vector of the treatment (0/1) |
U |
matrix of covariates to be considered for inclusion/exclusion |
W |
matrix of covariates that will be included in all models (optional) |
alpha |
vector of inclusion indicators (which columns of U) to included in the propensity score model |
binary |
indicates if the outcome is binary |
Value
A list. The list contains the following named components:
out |
a list that contains the BIC, predicted values, and estimated treatment effect from each outcome model |
Convert BIC to model probabilities
Description
This function transforms BIC to model probabilities
Usage
bic.to.prob(bic)
Arguments
bic |
vector of BICs |
Value
A vector of model probabilities of the same dimension of bic
Expit (inverse logit) function
Description
This function transforms the input using the expit function
Usage
expit(x)
Arguments
x |
vector of values to apply the expit function |
Value
A vector of the same dimension of x
Calculate model averaged double robust estimate
Description
This function estimates a model averaged double robust estimate.
Usage
madr(Y, X, U, W = NULL, M = 1000, cut = 0.95, enumerate = F,
tau = NULL, two.stage = NULL)
Arguments
Y |
vector of the outcome |
X |
vector of the treatment (0/1) |
U |
matrix of covariates to be considered for inclusion/exclusion |
W |
matrix of covariates that will be included in all models (optional) |
M |
the number of MCMC iteration |
cut |
cumulative probability of models to be retained for improved computational efficiency (1 retains all visited models) |
enumerate |
indicator if all possible models should be enumerated (default: FALSE) |
tau |
scalar value for the prior model dependence (1 is an independent prior; defaults to 0) |
two.stage |
indicator if the two-stage procedure for calculating the model weights should be used (defaults to TRUE) |
Value
A list. The list contains the following named components:
madr |
the model averaged double robust estimate |
weight.ps |
a vector that contains the inclusion probability of each covariate in the propensity score model |
weight.om |
a vector that contains the inclusion probability of each covariate in the outcome model |
Examples
set.seed(122)
## generate data
n = 100 # number of observations
k = 4 # number of covariates
U = matrix(rnorm(n*k),n,k)
colnames(U) = paste0("U",1:k)
A = rbinom(n,1,expit(-1+.5*rowSums(U)))
Y = rnorm(n,1+A+.25*rowSums(U))
## A is confounded -- true effect is 1
lm(Y~A)
## fit ma-dr -- can enumerate models if k isnt too big
res = madr(Y=Y,X=A,U=U,enumerate=TRUE,tau=1,two.stage=FALSE) # independent prior
res
res = madr(Y=Y,X=A,U=U,enumerate=TRUE,tau=0,two.stage=TRUE) # tau=0 and using two-stage weights
res
## no need to refit madr each time when enumerating -- use summarize and specify different taus
summary(res,tau=1,two.stage=FALSE) # independent prior
summary(res,tau=0,two.stage=FALSE)
summary(res,tau=0,two.stage=TRUE) # two-stage procedure for calculating weights
## use mcmc instead of enumerating (the default)
madr(Y=Y,X=A,U=U,M=1000,cut=1) #should approximate tau=0 and two.stage=TRUE
Model averaged double robust estimate with enumeration of all possible models (linear terms only)
Description
This function enumerates all possible models and estimates a model averaged double robust estimate
Usage
madr.enumerate(Y, X, U, W = NULL, tau = 1, two.stage = F)
Arguments
Y |
vector of the outcome |
X |
vector of the treatment indicator (0/1) |
U |
matrix of covariates to be considered for inclusion/exclusion |
W |
matrix of covariates that will be included in all models (optional) |
tau |
scalar value for the prior model dependence (1 is an independent prior) |
two.stage |
indicator if the two-stage procedure for calculating the model weights should be used |
Value
A object of class madr.enumerate. The object contains the following named components:
out |
a matrix that contains the BIC and estimated treatment from each outcome model |
ps |
a matrix that contains the BIC from each propensity score model |
dr |
a matrix that contains the model-specific double robust estimates |
U.names |
the column names of U |
Calculate model averaged double robust estimate using a pseudo-MC3 algorithm
Description
This function uses a pseudo-MC3 algorithm to search the model space, then estimate a model averaged double robust estimate using the two-stage procedure for estimating model weights with tau=0.
Usage
madr.mcmc(Y, X, U, W = NULL, M = 1000, cut = 0.95)
Arguments
Y |
vector of the outcome |
X |
vector of the treatment (0/1) |
U |
matrix of covariates to be considered for inclusion/exclusion |
W |
matrix of covariates that will be included in all models (optional) |
M |
the number of MCMC iteration |
cut |
cumulative probability of models to be retained for improved computational efficiency (1 retains all visited models) |
Value
A list. The list contains the following named components:
madr |
the model averaged double robust estimate |
weight.ps |
a vector that contains the inclusion probability of each covariate in the propensity score model |
weight.om |
a vector that contains the inclusion probability of each covariate in the outcome model |
Print function for madr.enumerate class
Description
This function prints results from madr.enumerate class
Usage
## S3 method for class 'madr.enumerate'
print(x, ...)
Arguments
x |
madr.enumerate object |
... |
ignored |
Print function for madr.mcmc class
Description
This function prints results from madr.mcmc class
Usage
## S3 method for class 'madr.mcmc'
print(x, ...)
Arguments
x |
madr.mcmc object |
... |
ignored |
Print function for summary.madr.enumerate class
Description
This function prints results from summary.madr.enumerate class
Usage
## S3 method for class 'summary.madr.enumerate'
print(x, ...)
Arguments
x |
summary.madr.enumerate object |
... |
ignored |
Provides model averaged double robust estimate for different values of tau
Description
This function estimates model averaged double robust estimate for different values of tau using a madr.enumerate object
Usage
## S3 method for class 'madr.enumerate'
summary(object, tau = NULL, two.stage = NULL, ...)
Arguments
object |
madr.enumerate object |
tau |
scalar value for the prior model dependence (1 is an independent prior; defaults to value used in madr.enumerate) |
two.stage |
indicator if the two-stage procedure for calculating the model weights should be used (defaults to value used in madr.enumerate) |
... |
ignored |
Value
A list. The list contains the following named components:
madr |
the model averaged double robust estimate |
weight.ps |
a vector that contains the inclusion probability of each covariate in the propensity score model |
weight.om |
a vector that contains the inclusion probability of each covariate in the outcome model |
tau |
value of tau used in estimation |
two.stage |
indicator if the two-stage procedure for calculating the model weights was used |