| Type: | Package |
| Title: | Causal Inference with Super Learner and Deep Neural Networks |
| Version: | 0.0.107 |
| Maintainer: | Nguyen K. Huynh <khoinguyen.huynh@r.hit-u.ac.jp> |
| Description: | Functions for deep learning estimation of Conditional Average Treatment Effects (CATEs) from meta-learner models and Population Average Treatment Effects on the Treated (PATT) in settings with treatment noncompliance using reticulate, TensorFlow and Keras3. Functions in the package also implements the conformal prediction framework that enables computation and illustration of conformal prediction (CP) intervals for estimated individual treatment effects (ITEs) from meta-learner models. Additional functions in the package permit users to estimate the meta-learner CATEs and the PATT in settings with treatment noncompliance using weighted ensemble learning via the super learner approach and R neural networks. |
| License: | GPL-3 |
| Encoding: | UTF-8 |
| LazyData: | true |
| Imports: | ROCR, caret, neuralnet, SuperLearner, ggplot2, tidyr, magrittr, reticulate, keras3, Hmisc |
| Suggests: | testthat (≥ 3.0.0), dplyr, class, xgboost, randomForest, glmnet, ranger, gam, e1071, gbm, tensorflow |
| RoxygenNote: | 7.3.2 |
| Depends: | R (≥ 4.1.0) |
| URL: | https://github.com/hknd23/DeepLearningCausal |
| BugReports: | https://github.com/hknd23/DeepLearningCausal/issues |
| Config/testthat/edition: | 3 |
| NeedsCompilation: | no |
| Packaged: | 2025-10-30 07:58:32 UTC; nguye |
| Author: | Nguyen K. Huynh 
[aut, cre],
Bumba Mukherjee
[aut],
Yang Yang [aut] |
| Repository: | CRAN |
| Date/Publication: | 2025-10-30 10:30:02 UTC |
Build Keras model
Description
Specify model for deep learning
Usage
build_model(
hidden.layer,
input_shape,
output_units = 1,
output_activation = "sigmoid",
hidden_activation = "relu",
dropout_rate = NULL
)
Arguments
|
vector of integers for number of hidden units in each hidden layer | |
input_shape |
integer for number of input features |
output_units |
integer for number of output units, default is 1 |
output_activation |
string for output layer activation function, default is "sigmoid" |
|
string for hidden layer activation function, default is "relu" | |
dropout_rate |
double or vector for proportion of hidden layer to drop out. |
Value
Keras model object
Check for required CRAN packages and prompt installation if missing.
Description
Check for required CRAN packages and prompt installation if missing.
Usage
check_cran_deps()
Value
Invisibly returns TRUE if all required packages are installed.
Check for required Python modules and prompt installation if missing.
Description
Check for required Python modules and prompt installation if missing.
Usage
check_python_modules()
Value
Invisibly returns TRUE if all required modules are available.
Train complier model using ensemble methods
Description
Train model using group exposed to treatment with compliance as binary outcome variable and covariates.
Usage
complier_mod(
exp.data,
complier.formula,
treat.var,
ID = NULL,
SL.learners = c("SL.glmnet", "SL.xgboost", "SL.ranger", "SL.nnet", "SL.glm")
)
Arguments
exp.data |
list object of experimental data. |
complier.formula |
formula to fit compliance model (c ~ x) using complier variable and covariates |
treat.var |
string specifying the binary treatment variable |
ID |
string for name of identifier variable. |
SL.learners |
vector of strings for ML classifier algorithms. Defaults to extreme gradient boosting, elastic net regression, random forest, and neural nets. |
Value
model object of trained model.
Complier model prediction
Description
Predict Compliance from control group in experimental data
Usage
complier_predict(complier.mod, exp.data, treat.var, compl.var)
Arguments
complier.mod |
output from trained ensemble superlearner model |
exp.data |
|
treat.var |
string specifying the binary treatment variable |
compl.var |
string specifying binary complier variable |
Value
data.frame object with true compliers, predicted compliers in the
control group, and all compliers (actual + predicted).
conformal_plot
Description
Visualizes the distribution of estimated individual treatment effects (ITEs)
along with their corresponding conformal prediction intervals.
The function randomly samples a proportion of observations from a fitted
metalearner_ensemble or metalearner_deeplearning object and
plots the conformal intervals as vertical ranges around the point estimates.
This allows users to visually assess the uncertainty and variation in
estimated treatment effects.
Usage
conformal_plot(
x,
...,
seed = 1234,
prop = 0.3,
binary.outcome = FALSE,
x.labels = TRUE,
x.title = "Observations",
color = "steelblue",
break.by = 0.5
)
Arguments
x |
A fitted model object of class |
... |
Additional arguments (currently unused). |
seed |
Random seed for reproductibility of subsampling. Default is |
prop |
Proportion of observations to randomly sample for plotting.
Must be between 0 and 1. Default is |
binary.outcome |
Logical; if |
x.labels |
Logical; if |
x.title |
Character string specifying the x-axis title.
Default is |
color |
Color of the conformal intervals and points.
Default is |
break.by |
Numeric value determining the spacing between y-axis breaks.
Default is |
Details
The function extracts the estimated ITEs (CATEs) and conformal intervals
(ITE_lower, ITE_upper) from the model output, samples a subset
of rows, and generates a ggplot2 visualization.
Each vertical line represents the conformal prediction interval for one observation’s
treatment effect estimate.
The conformal intervals are typically obtained from weighted split-conformal inference,
using propensity overlap weights to adjust interval width.
Value
A ggplot object showing sampled individual treatment effects
with their weighted conformal prediction intervals.
Train complier model using deep neural learning through Tensorflow
Description
Train model using group exposed to treatment with compliance as binary outcome variable and covariates.
Usage
deep_complier_mod(
complier.formula,
exp.data,
treat.var,
algorithm = "adam",
hidden.layer = c(2, 2),
hidden_activation = "relu",
ID = NULL,
epoch = 10,
verbose = 1,
batch_size = 32,
validation_split = NULL,
patience = NULL,
dropout_rate = NULL
)
Arguments
complier.formula |
formula to fit compliance model (c ~ x) using complier variable and covariates |
exp.data |
list object of experimental data. |
treat.var |
string specifying the binary treatment variable |
algorithm |
string for name of optimizer algorithm. Set to adam. other optimization algorithms available are sgd, rprop, adagrad. |
|
vector specifying the hidden layers and the number of neurons in each layer. | |
|
string or vector for activation function used for hidden layers. Defaults to "relu". | |
ID |
string for name of identifier variable. |
epoch |
integer for number of epochs |
verbose |
1 to display model training information and learning curve plot. 0 to suppress messages and plots. |
batch_size |
integer for batch size to split the training set. Defaults to 32. |
validation_split |
double for proportion of training data to be split for validation. |
patience |
integer for number of epochs with no improvement after which training will be stopped. |
dropout_rate |
double or vector for proportion of hidden layer to drop out. |
Value
deep.complier.mod model object
Complier model prediction
Description
Predict Compliance from control group in experimental data
Usage
deep_predict(
deep.complier.mod,
complier.formula,
exp.data,
treat.var,
compl.var
)
Arguments
deep.complier.mod |
model object from |
complier.formula |
formula to fit compliance model (c ~ x) using complier variable and covariates |
exp.data |
|
treat.var |
string specifying the binary treatment variable |
compl.var |
string specifying binary complier variable |
Value
data.frame object with true compliers, predicted compliers in the
control group, and all compliers (actual + predicted).
Response model from experimental data using deep neural learning through Tensorflow
Description
Train response model (response variable as outcome and covariates) from all compliers (actual + predicted) in experimental data using Tensorflow.
Usage
deep_response_model(
response.formula,
exp.data,
exp.compliers,
compl.var,
algorithm = "adam",
hidden.layer = c(2, 2),
hidden_activation = "relu",
epoch = 10,
verbose = 1,
batch_size = 32,
output_units = 1,
validation_split = NULL,
patience = NULL,
output_activation = "linear",
loss = "mean_squared_error",
metrics = "mean_squared_error",
dropout_rate = NULL
)
Arguments
response.formula |
formula specifying the response variable and covariates. |
exp.data |
experimental dataset. |
exp.compliers |
|
compl.var |
string specifying binary complier variable |
algorithm |
string for optimizer algorithm in response model. |
|
vector specifying hidden layers and the number of neurons in each hidden layer | |
|
string or vector for activation functions in hidden layers. | |
epoch |
integer for number of epochs |
verbose |
1 to display model training information and learning curve plot. 0 to suppress messages and plots. |
batch_size |
batch size to split training data. |
output_units |
integer for units in output layer. Defaults to 1 for continuous and binary outcome variables. In case of multinomial outcome variable, value should be set to the number of categories. |
validation_split |
double for the proportion of test data to be split as validation in response model. |
patience |
integer for number of epochs with no improvement after which training will be stopped. |
output_activation |
string for activation function in output layer. "linear" is recommended for continuous outcome variables, and "sigmoid" for binary outcome variables |
loss |
string for loss function. "mean_squared_error" recommended for linear models, "binary_crossentropy" for binary models. |
metrics |
string for metrics. "mean_squared_error" recommended for linear models, "binary_accuracy" for binary models. |
dropout_rate |
double or vector for proportion of hidden layer to drop out in response model. |
Value
model object of trained response model.
Survey Experiment of Support for Populist Policy
Description
Shortened version of survey response data that incorporates a vignette survey experiment. The vignette describes an international crisis between country A and B. After reading this vignette, respondents are randomly assigned to the control group or to one of two treatments: policy prescription to said crisis by strong (populist) leader and centrist (non-populist) leader. The respondents are then asked whether they are willing to support the policy decision to fight a war against country A, which is the dependent variable.
Usage
data(exp_data)
Format
exp_data
A data frame with 257 rows and 12 columns:
- female
Gender.
- age
Age of participant.
- income
Monthly household income.
- religion
Religious denomination
- practicing_religion
Importance of religion in life.
- education
Educational level of participant.
- political_ideology
Political ideology of participant.
- employment
Employment status of participant.
- marital_status
Marital status of participant.
- job_loss
Concern about job loss.
- strong_leader
Binary treatment measure of leader type.
- support_war
Binary outcome measure for willingness to fight war.
#' ...
Source
Yadav and Mukherjee (2024)
Survey Experiment of Support for Populist Policy
Description
Extended experiment data with 514 observations
Usage
data(exp_data_full)
Format
exp_data_full
A data frame with 514 rows and 12 columns:
- female
Gender.
- age
Age of participant.
- income
Monthly household income.
- religion
Religious denomination
- practicing_religion
Importance of religion in life.
- education
Educational level of participant.
- political_ideology
Political ideology of participant.
- employment
Employment status of participant.
- marital_status
Marital status of participant.
- job_loss
Concern about job loss.
- strong_leader
Binary treatment measure of leader type.
- support_war
Binary outcome measure for willingness to fight war.
#' ...
Source
Yadav and Mukherjee (2024)
Create list for experimental data
Description
create list object of experimental data for easy data processing
Usage
expcall(
response.formula,
treat.var,
compl.var,
exp.data,
weights = NULL,
cluster = NULL,
ID = NULL
)
Arguments
response.formula |
formula for response equation of binary outcome variable and covariates |
treat.var |
string for binary treatment variable |
compl.var |
string for complier variable |
exp.data |
|
weights |
observation weights |
cluster |
clustering variable |
ID |
identifier variable |
Value
list of processed dataset
hte_plot
Description
Produces plot to illustrate sub-group Heterogeneous Treatment Effects (HTE)
of estimated CATEs from metalearner_ensemble and
metalearner_neural, as well as PATT-C from pattc_ensemble
and pattc_neural.
Usage
hte_plot(
x,
...,
boot = TRUE,
n_boot = 1000,
cut_points = NULL,
custom_labels = NULL,
zero_int = TRUE,
selected_vars = NULL
)
Arguments
x |
estimated model from |
... |
Additional arguments |
boot |
logical for using bootstraps to estimate confidence intervals. |
n_boot |
number of bootstrap iterations. Only used with boot = TRUE. |
cut_points |
numeric vector for cut-off points to generate subgroups from covariates. If left blank a vector generated from median values will be used. |
custom_labels |
character vector for the names of subgroups. |
zero_int |
logical for vertical line at 0 x intercept. |
selected_vars |
vector for names of covariates to use for subgroups. |
Value
ggplot object illustrating subgroup HTE and 95% confidence
intervals.
Examples
# load dataset
set.seed(123456)
xlearner_nn <- metalearner_neural(cov.formula = support_war ~ age +
income + employed + job_loss,
data = exp_data,
treat.var = "strong_leader",
meta.learner.type = "X.Learner",
stepmax = 2e+9,
nfolds = 5,
algorithm = "rprop+",
hidden.layer = c(3),
linear.output = FALSE,
binary.preds = FALSE)
hte_plot(xlearner_nn)
hte_plot(xlearner_nn,
selected_vars = c("age", "income"),
cut_points = c(33, 3),
custom_labels = c("Age <= 33", "Age > 33", "Income <= 3", "Income > 3"),
n_boot = 500)
metalearner_deeplearning
Description
metalearner_deeplearning implements the meta learners for estimating
CATEs using Deep Neural Networks through Tensorflow.
Deep Learning Estimation of CATEs from four meta-learner models (S,T,X and R-learner)
using TensorFlow and Keras3
Usage
metalearner_deeplearning(
data = NULL,
train.data = NULL,
test.data = NULL,
cov.formula,
treat.var,
meta.learner.type,
nfolds = 5,
algorithm = "adam",
hidden.layer = c(2, 2),
hidden_activation = "relu",
output_activation = "linear",
output_units = 1,
loss = "mean_squared_error",
metrics = "mean_squared_error",
epoch = 10,
verbose = 1,
batch_size = 32,
validation_split = NULL,
patience = NULL,
dropout_rate = NULL,
conformal = FALSE,
alpha = 0.1,
calib_frac = 0.5,
prob_bound = TRUE,
seed = 1234
)
Arguments
data |
data.frame object of data. If a single dataset is specified, then the model will use cross-validation to train the meta-learners and estimate CATEs. Users can also specify the arguments (defined below) to separately train meta-learners on their training data and estimate CATEs with their test data. |
train.data |
data.frame object of training data for Train/Test mode. Argument must be specified to separately train the meta-learners on the training data. |
test.data |
data.frame object of test data for Train/Test mode. Argument must be specified to estimate CATEs on the test data. |
cov.formula |
formula description of the model y ~ x(list of covariates). Permits users to specify covariates in the meta-learner model of interest. This includes the outcome variables and the confounders. |
treat.var |
string for name of Treatment variable. Users can specify the treatment variable in their data by employing the treat.var argument. |
meta.learner.type |
string of "S.Learner", "T.Learner", "X.Learner", or "R.Learner". Employed to specify any of the following four meta-learner models for estimation via deep learning: S,T,X or R-Learner. |
nfolds |
integer for number of folds for Meta Learners. When a single dataset is specified, then users employ cross-validation to train the meta-learners and estimate CATEs. For a single dataset, users specify nfolds to define the number of folds to split data for cross-validation. |
algorithm |
string for optimization algorithm.
For optimizers available see |
|
permits users to specify the number of hidden layers in the model and the number of neurons in each hidden layer. | |
|
string or vector for name of activation function for hidden layers of model. Defaults to "relu" which means that users can specify a single value to use one activation function for each hidden layer. While "relu" is a popular choice for hidden layers, users can also use "softmax" which converts a vector of values into a probability distribution and "tanh" that maps input to a value between -1 and 1. | |
output_activation |
string for name of activation function for output layer of model.
"linear" is recommended for continuous outcome variables, and "sigmoid" for binary outcome variables.
For activation functions available see |
output_units |
integer for units in output layer. Defaults to 1 for continuous and binary outcome variables. In case of multinomial outcome variable, set to the number of categories. |
loss |
string for loss function "mean_squared_error" recommended for linear models, "binary_crossentropy" for binary models. |
metrics |
string for metrics in response model. "mean_squared_error" recommended for linear models, "binary_accuracy" for binary models. |
epoch |
interger for number of epochs. epoch denotes one complete pass through the entire training dataset. Model processes each training example once during an epoch. |
verbose |
integer specifying the verbosity level during training. 1 for full information and learning curve plots. 0 to suppress messages and plots. |
batch_size |
integer for batch size to split training data. batch size refers to the number of training samples processed before the model's parameters are updated. Batch size is a vital hyperparameter that affects both training speed and model performance. It is crucial for computational efficiency. |
validation_split |
double for proportion of training data to split for validation. validation split involves partitioning data into training and validation sets to build and tune model. |
patience |
integer for number of epochs with no improvement to wait before stopping training. patience stops training of neural network if model's performance on validation data stops improving. |
dropout_rate |
double or vector for proportion of hidden layer to drop out. dropout rate is hyperparameter for preventing a model from overfitting the training data. |
conformal |
logical for whether to compute conformal prediction intervals conformal prediction intervals provide measure of uncertainty for ITEs. |
alpha |
proportion for conformal prediction intervals alpha proportion refers to significance level that guarantees desired coverage probability for ITEs |
calib_frac |
fraction of training data to use for calibration in conformal inference |
prob_bound |
logical for whether to bound conformal intervals within [-1,1] for classification models |
seed |
random seed |
Value
metalearner_deeplearning object with CATEs
Examples
## Not run:
#check for python and required modules
python_ready()
data("exp_data")
s_deeplearning <- metalearner_deeplearning(data = exp_data,
cov.formula = support_war ~ age + female + income + education
+ employed + married + hindu + job_loss,
treat.var = "strong_leader", meta.learner.type = "S.Learner",
nfolds = 5, algorithm = "adam",
hidden.layer = c(2,2), hidden_activation = "relu",
output_activation = "sigmoid", output_units = 1,
loss = "binary_crossentropy", metrics = "accuracy",
epoch = 10, verbose = 1, batch_size = 32,
validation_split = NULL, patience = NULL,
dropout_rate = NULL, conformal= FALSE, seed=1234)
## End(Not run)
## Not run:
#check for python and required modules
python_ready()
data("exp_data")
t_deeplearning <- metalearner_deeplearning(data = exp_data,
cov.formula = support_war ~ age + female + income + education
+ employed + married + hindu + job_loss,
treat.var = "strong_leader", meta.learner.type = "T.Learner",
nfolds = 5, algorithm = "adam",
hidden.layer = c(2,2), hidden_activation = "relu",
output_activation = "sigmoid", output_units = 1,
loss = "binary_crossentropy", metrics = "accuracy",
epoch = 10, verbose = 1, batch_size = 32,
validation_split = NULL, patience = NULL,
dropout_rate = NULL, conformal= TRUE,
alpha = 0.1,calib_frac = 0.5, prob_bound = TRUE, seed = 1234)
## End(Not run)
metalearner_ensemble
Description
metalearner_ensemble implements the S-learner, T-learner, and X-learner for
weighted ensemble learning estimation of CATEs using super learner. The super learner in
this case includes the following machine learning algorithms:
extreme gradient boosting, glmnet (elastic net regression), random forest and
neural nets.
Usage
metalearner_ensemble(
data = NULL,
train.data = NULL,
test.data = NULL,
cov.formula,
treat.var,
meta.learner.type,
SL.learners = c("SL.glmnet", "SL.xgboost", "SL.nnet"),
nfolds = 5,
family = gaussian(),
binary.preds = FALSE,
conformal = FALSE,
alpha = 0.1,
calib_frac = 0.5,
seed = 1234
)
Arguments
data |
|
train.data |
|
test.data |
|
cov.formula |
formula description of the model y ~ x(list of covariates) permits users to incorporate outcome variable and confounders in model. |
treat.var |
string for the name of treatment variable. |
meta.learner.type |
string specifying is the S-learner and
|
SL.learners |
vector for super learner ensemble that includes extreme gradient boosting, glmnet, random forest, and neural nets. |
nfolds |
number of folds for cross-validation. Currently supports up to 5 folds. |
family |
gaussian() or binomial() family for outcome variable. 5 folds. |
binary.preds |
logical for whether outcome predictions should be binary |
conformal |
logical for whether to compute conformal prediction intervals |
alpha |
proportion for conformal prediction intervals |
calib_frac |
fraction of training data to use for calibration in conformal inference |
seed |
random seed |
Value
metalearner_ensemble of predicted outcome values and CATEs
estimated by the meta learners for each observation.
Examples
# load dataset
data(exp_data)
#load SuperLearner package
library(SuperLearner)
# estimate CATEs with S Learner
set.seed(123456)
slearner <- metalearner_ensemble(cov.formula = support_war ~ age +
income + employed + job_loss,
data = exp_data,
treat.var = "strong_leader",
meta.learner.type = "S.Learner",
SL.learners = c("SL.glm"),
nfolds = 5,
binary.preds = FALSE,
)
print(slearner)
# estimate CATEs with T Learner
set.seed(123456)
tlearner <- metalearner_ensemble(cov.formula = support_war ~ age + income +
employed + job_loss,
data = exp_data,
treat.var = "strong_leader",
meta.learner.type = "T.Learner",
SL.learners = c("SL.xgboost",
"SL.nnet"),
nfolds = 5,
binary.preds = FALSE,
)
print(tlearner)
# estimate CATEs with X Learner
set.seed(123456)
xlearner <- metalearner_ensemble(cov.formula = support_war ~ age + income +
employed + job_loss,
test.data = exp_data,
train.data = exp_data,
treat.var = "strong_leader",
meta.learner.type = "X.Learner",
SL.learners = c("SL.glmnet","SL.xgboost",
"SL.nnet"),
binary.preds = TRUE)
print(xlearner)
metalearner_neural
Description
metalearner_neural implements the S-learner and T-learner for estimating
CATE using Deep Neural Networks. The Resilient back propagation (Rprop)
algorithm is used for training neural networks.
Usage
metalearner_neural(
data,
cov.formula,
treat.var,
meta.learner.type,
stepmax = 1e+05,
nfolds = 5,
algorithm = "rprop+",
hidden.layer = c(4, 2),
act.fct = "logistic",
err.fct = "sse",
linear.output = TRUE,
binary.preds = FALSE
)
Arguments
data |
|
cov.formula |
formula description of the model y ~ x(list of covariates). |
treat.var |
string for the name of treatment variable. |
meta.learner.type |
string specifying is the S-learner and
|
stepmax |
maximum number of steps for training model. |
nfolds |
number of folds for cross-validation. Currently supports up to 5 folds. |
algorithm |
a string for the algorithm for the neural network.
Default set to |
|
vector of integers specifying layers and number of neurons. | |
act.fct |
"logistic" or "tanh" for activation function to be used in the neural network. |
err.fct |
"ce" for cross-entropy or "sse" for sum of squared errors as error function. |
linear.output |
logical specifying regression (TRUE) or classification (FALSE) model. |
binary.preds |
logical specifying predicted outcome variable will take binary values or proportions. |
Value
metalearner_neural of predicted outcome values and CATEs estimated by the meta
learners for each observation.
Examples
# load dataset
data(exp_data)
# estimate CATEs with S Learner
set.seed(123456)
slearner_nn <- metalearner_neural(cov.formula = support_war ~ age + income +
employed + job_loss,
data = exp_data,
treat.var = "strong_leader",
meta.learner.type = "S.Learner",
stepmax = 2e+9,
nfolds = 5,
algorithm = "rprop+",
hidden.layer = c(1),
linear.output = FALSE,
binary.preds = FALSE)
print(slearner_nn)
# load dataset
set.seed(123456)
# estimate CATEs with T Learner
tlearner_nn <- metalearner_neural(cov.formula = support_war ~ age +
income +
employed + job_loss,
data = exp_data,
treat.var = "strong_leader",
meta.learner.type = "T.Learner",
stepmax = 1e+9,
nfolds = 5,
algorithm = "rprop+",
hidden.layer = c(2,1),
linear.output = FALSE,
binary.preds = FALSE)
print(tlearner_nn)
# load dataset
set.seed(123456)
# estimate CATEs with X Learner
xlearner_nn <- metalearner_neural(cov.formula = support_war ~ age +
income +
employed + job_loss,
data = exp_data,
treat.var = "strong_leader",
meta.learner.type = "X.Learner",
stepmax = 2e+9,
nfolds = 5,
algorithm = "rprop+",
hidden.layer = c(3),
linear.output = FALSE,
binary.preds = FALSE)
print(xlearner_nn)
Train compliance model using neural networks
Description
Train model using group exposed to treatment with compliance as binary outcome variable and covariates.
Usage
neuralnet_complier_mod(
complier.formula,
exp.data,
treat.var,
algorithm = "rprop+",
hidden.layer = c(4, 2),
act.fct = "logistic",
ID = NULL,
stepmax = 1e+08
)
Arguments
complier.formula |
formula for complier variable as outcome and covariates (c ~ x) |
exp.data |
|
treat.var |
string for treatment variable. |
algorithm |
string for algorithm for training neural networks.
Default set to the Resilient back propagation with weight backtracking
(rprop+). Other algorithms include backprop', rprop-', 'sag', or 'slr'
(see |
|
vector for specifying hidden layers and number of neurons. | |
act.fct |
"logistic" or "tanh activation function. |
ID |
string for identifier variable |
stepmax |
maximum number of steps. |
Value
trained complier model object
Assess Population Data counterfactuals
Description
Create counterfactual datasets in the population for compliers and
noncompliers. Then predict potential outcomes using trained model from
neuralnet_response_model.
Usage
neuralnet_pattc_counterfactuals(
pop.data,
neuralnet.response.mod,
ID = NULL,
cluster = NULL,
binary.preds = FALSE
)
Arguments
pop.data |
population data. |
neuralnet.response.mod |
trained model from.
|
ID |
string for identifier variable. |
cluster |
string for clustering variable (currently unused). |
binary.preds |
logical specifying predicted outcome variable will take binary values or proportions. |
Value
data.frame of predicted outcomes of response variable from
counterfactuals.
Predicting Compliance from experimental data
Description
Predicting Compliance from control group experimental data
Usage
neuralnet_predict(neuralnet.complier.mod, exp.data, treat.var, compl.var)
Arguments
neuralnet.complier.mod |
results from |
exp.data |
|
treat.var |
string for treatment variable |
compl.var |
string for compliance variable |
Value
data.frame object with true compliers, predicted compliers in the
control group, and all compliers (actual + predicted).
Modeling Responses from experimental data Using Deep NN
Description
Model Responses from all compliers (actual + predicted) in experimental data using neural network.
Usage
neuralnet_response_model(
response.formula,
exp.data,
neuralnet.compliers,
compl.var,
algorithm = "rprop+",
hidden.layer = c(4, 2),
act.fct = "logistic",
err.fct = "sse",
linear.output = TRUE,
stepmax = 1e+08
)
Arguments
response.formula |
formula for response variable and covariates (y ~ x) |
exp.data |
|
neuralnet.compliers |
|
compl.var |
string of compliance variable |
algorithm |
neural network algorithm, default set to |
|
vector specifying hidden layers and number of neurons. | |
act.fct |
"logistic" or "tanh activation function. |
err.fct |
"sse" for sum of squared errors or "ce" for cross-entropy. |
linear.output |
logical for whether output (outcome variable) is linear or not. |
stepmax |
maximum number of steps for training model. |
Value
trained response model object
Assess Population Data counterfactuals
Description
Create counterfactual datasets in the population for compliers and noncompliers. Then predict potential outcomes from counterfactuals.
Usage
pattc_counterfactuals(
pop.data,
response.mod,
ID = NULL,
cluster = NULL,
binary.preds = FALSE
)
Arguments
pop.data |
population dataset |
response.mod |
trained model from |
ID |
string fir identifier variable |
cluster |
string for clustering variable |
binary.preds |
logical specifying whether predicted outcomes are proportions or binary (0-1). |
Value
data.frame object of predicted outcomes of counterfactual groups.
Deep PATT-C
Description
This function implements the Deep PATT-C method for estimating the Population Average Treatment Effect on the Treated Compliers (PATT-C) using deep learning models using keras and Tensorflow. It consists of training a deep learning model to predict compliance among treated individuals, predicting compliance in the experimental data, training a response model among predicted compliers, and estimating counterfactual outcomes in the population data.
Usage
pattc_deeplearning(
response.formula,
compl.var,
treat.var,
exp.data,
pop.data,
compl.algorithm = "adam",
response.algorithm = "adam",
compl.hidden.layer = c(4, 2),
response.hidden.layer = c(4, 2),
compl.hidden_activation = "relu",
response.hidden_activation = "relu",
response.output_activation = "linear",
response.output_units = 1,
response.loss = "mean_squared_error",
response.metrics = "mean_absolute_error",
ID = NULL,
weights = NULL,
cluster = NULL,
compl.epoch = 10,
response.epoch = 10,
compl.validation_split = NULL,
response.validation_split = NULL,
compl.patience = NULL,
response.patience = NULL,
compl.dropout_rate = NULL,
response.dropout_rate = NULL,
verbose = 1,
batch_size = 32,
nboot = 1000,
seed = 1234
)
Arguments
response.formula |
formula specifying the response variable and covariates. |
compl.var |
string specifying the name of the compliance variable. |
treat.var |
string specifying the name of the treatment variable. |
exp.data |
data frame containing the experimental data. |
pop.data |
data frame containing the population data. |
compl.algorithm |
string for name of optimizer algorithm for complier model. For optimizers available see |
response.algorithm |
string for name of optimizer algorithm for response model. For optimizers available see |
|
vector specifying the hidden layers in the complier model and the number of neurons in each hidden layer. | |
|
vector specifying the hidden layers in the response model and the number of neurons in each hidden layer. | |
|
string or vector for name of activation function for hidden layers complier model. Defaults to "relu" (Rectified Linear Unit) | |
|
string or vector for name of activation function for hidden layers complier model. Defaults to "relu" (Rectified Linear Unit) | |
response.output_activation |
string for name of activation function for output layer of response model.
"linear" is recommended for continuous outcome variables, and "sigmoid" for binary outcome variables.
For activation functions available see |
response.output_units |
integer for units in output layer. Defaults to 1 for continuous and binary outcome variables. In case of multinomial outcome variable, set to the number of categories. |
response.loss |
string for loss function in response model. "mean_squared_error" recommended for linear models, "binary_crossentropy" for binary models. |
response.metrics |
string for metrics in response model. "mean_squared_error" recommended for linear models, "binary_accuracy" for binary models. |
ID |
optional string specifying the name of the identifier variable. |
weights |
optional string specifying the name of the weights variable. |
cluster |
optional string specifying the name of the clustering variable. |
compl.epoch |
Integer for the number of epochs for complier model. |
response.epoch |
integer for the number of epochs for response model. |
compl.validation_split |
double for the proportion of test data to be split as validation in complier model. Defaults to 0.2. |
response.validation_split |
double for the proportion of test data to be split as validation in response model. Defaults to 0.2. |
compl.patience |
integer for number of epochs with no improvement after which training will be stopped in complier model. |
response.patience |
integer for number of epochs with no improvement after which training will be stopped in response model. |
compl.dropout_rate |
double or vector for proportion of hidden layer to drop out in complier model. |
response.dropout_rate |
double or vector for proportion of hidden layer to drop out in response model. |
verbose |
integer specifying the verbosity level during training. Defaults to 1. |
batch_size |
integer specifying the batch size for training the deep learning models. Default is 32. |
nboot |
integer specifying the number of bootstrap samples if bootstrap is TRUE. Default is 1000. |
seed |
random seed |
Value
pattc_deeplearning object containing the fitted models, predictions, counterfactuals, and PATT-C estimate.
Examples
## Not run:
#check for python and required modules
python_ready()
data("exp_data")
data("pop_data")
set.seed(1243)
deeppattc <- pattc_deeplearning(response.formula = support_war ~ age + female +
income + education + employed + married + hindu + job_loss,
exp.data = exp_data, pop.data = pop_data,
treat.var = "strong_leader", compl.var = "compliance",
compl.algorithm = "adam", response.algorithm = "adam",
compl.hidden.layer = c(4,2), response.hidden.layer = c(4,2),
compl.hidden_activation = "relu", response.hidden_activation = "relu",
response.output_activation = "sigmoid", response.output_units = 1,
response.loss = "binary_crossentropy", response.metrics = "accuracy",
compl.epoch = 50, response.epoch = 80,
verbose = 1, batch_size = 32,
compl.validation_split = 0.2, response.validation_split = 0.2,
compl.dropout_rate = 0.1, response.dropout_rate = 0.1,
compl.patience = 20, response.patience = 20,
nboot = 1000, seed = 1234)
## End(Not run)
Assess Population Data counterfactuals
Description
Create counterfactual datasets in the population for compliers and noncompliers.
Usage
pattc_deeplearning_counterfactuals(
pop.data,
response.mod,
response.formula,
ID = NULL,
cluster = NULL
)
Arguments
pop.data |
population dataset |
response.mod |
trained model from |
response.formula |
formula specifying the response variable and covariates. |
ID |
string fir identifier variable |
cluster |
string for clustering variable |
Value
data.frame object of predicted outcomes of counterfactual groups.
PATT-C SL Ensemble
Description
pattc_ensemble estimates the Population Average Treatment Effect
of the Treated from experimental data with noncompliers
using the super learner ensemble that includes extreme gradient boosting,
glmnet (elastic net regression), random forest and neural nets.
Usage
pattc_ensemble(
response.formula,
exp.data,
pop.data,
treat.var,
compl.var,
compl.SL.learners = c("SL.glmnet", "SL.xgboost", "SL.ranger", "SL.nnet", "SL.glm"),
response.SL.learners = c("SL.glmnet", "SL.xgboost", "SL.ranger", "SL.nnet", "SL.glm"),
response.family = gaussian(),
ID = NULL,
cluster = NULL,
binary.preds = FALSE,
bootstrap = FALSE,
nboot = 1000
)
Arguments
response.formula |
formula for the effects of covariates on outcome variable (y ~ x). |
exp.data |
|
pop.data |
|
treat.var |
string for binary treatment variable. |
compl.var |
string for binary compliance variable. |
compl.SL.learners |
vector of names of ML algorithms used for compliance model. |
response.SL.learners |
vector of names of ML algorithms used for response model. |
response.family |
gaussian() or binomial() for response model. |
ID |
string for name of identifier. (currently not used) |
cluster |
string for name of cluster variable. (currently not used) |
binary.preds |
logical specifying predicted outcome variable will take binary values or proportions. |
bootstrap |
logical for bootstrapped PATT-C. |
nboot |
number of bootstrapped samples. Only used with
|
Value
pattc_ensemble object of results of t test as PATTC estimate.
Examples
# load datasets
data(exp_data_full) # full experimental data
data(exp_data) #experimental data
data(pop_data) #population data
#attach SuperLearner (model will not recognize learner if package is not loaded)
library(SuperLearner)
set.seed(123456)
#specify models and estimate PATTC
pattc_boot <- pattc_ensemble(response.formula = support_war ~ age + income +
education + employed + job_loss,
exp.data = exp_data_full,
pop.data = pop_data,
treat.var = "strong_leader",
compl.var = "compliance",
compl.SL.learners = c("SL.glm", "SL.nnet"),
response.SL.learners = c("SL.glm", "SL.nnet"),
response.family = binomial(),
ID = NULL,
cluster = NULL,
binary.preds = FALSE,
bootstrap = TRUE,
nboot = 1000)
print(pattc_boot)
Estimate PATT_C using Deep NN
Description
estimates the Population Average Treatment Effect of the Treated from experimental data with noncompliers using Deep Neural Networks.
Usage
pattc_neural(
response.formula,
exp.data,
pop.data,
treat.var,
compl.var,
compl.algorithm = "rprop+",
response.algorithm = "rprop+",
compl.hidden.layer = c(4, 2),
response.hidden.layer = c(4, 2),
compl.act.fct = "logistic",
response.err.fct = "sse",
response.act.fct = "logistic",
linear.output = TRUE,
compl.stepmax = 1e+08,
response.stepmax = 1e+08,
ID = NULL,
cluster = NULL,
binary.preds = FALSE,
bootstrap = FALSE,
nboot = 1000
)
Arguments
response.formula |
formula of response variable as outcome and covariates (y ~ x) |
exp.data |
|
pop.data |
|
treat.var |
string for treatment variable. |
compl.var |
string for compliance variable |
compl.algorithm |
string for algorithm to train neural network for
compliance model. Default set to |
response.algorithm |
string for algorithm to train neural network for
response model. Default set to |
|
vector for specifying hidden layers and number of neurons in complier model. | |
|
vector for specifying hidden layers and number of neurons in response model. | |
compl.act.fct |
"logistic" or "tanh" activation function for complier model. |
response.err.fct |
"sse" for sum of squared errors or "ce" for cross-entropy for response model. |
response.act.fct |
"logistic" or "tanh" activation function for response model. |
linear.output |
logical for whether output (outcome variable) is linear or not for response model. |
compl.stepmax |
maximum number of steps for complier model |
response.stepmax |
maximum number of steps for response model |
ID |
string for identifier variable |
cluster |
string for cluster variable. |
binary.preds |
logical specifying predicted outcome variable will take binary values or proportions. |
bootstrap |
logical for bootstrapped PATT-C. |
nboot |
number of bootstrapped samples |
Value
pattc_neural class object of results of t test as PATTC estimate.
Examples
# load datasets
data(exp_data) #experimental data
data(pop_data) #population data
# specify models and estimate PATTC
set.seed(123456)
pattc_neural_boot <- pattc_neural(response.formula = support_war ~ age + female +
income + education + employed + married +
hindu + job_loss,
exp.data = exp_data,
pop.data = pop_data,
treat.var = "strong_leader",
compl.var = "compliance",
compl.algorithm = "rprop+",
response.algorithm = "rprop+",
compl.hidden.layer = c(2),
response.hidden.layer = c(2),
compl.stepmax = 1e+09,
response.stepmax = 1e+09,
ID = NULL,
cluster = NULL,
binary.preds = FALSE,
bootstrap = TRUE,
nboot = 1000)
plot.metalearner_ensemble
Description
Uses plot() to generate histogram of distribution of CATEs or predicted
outcomes from metalearner_ensemble
Usage
## S3 method for class 'metalearner_ensemble'
plot(x, ..., conf_level = 0.95, type = "CATEs")
Arguments
x |
|
... |
Additional arguments |
conf_level |
numeric value for confidence level. Defaults to 0.95. |
type |
"CATEs" or "predict" |
Value
ggplot object
plot.metalearner_neural
Description
Uses plot() to generate histogram of distribution of CATEs or predicted
outcomes from metalearner_neural
Usage
## S3 method for class 'metalearner_neural'
plot(x, ..., conf_level = 0.95, type = "CATEs")
Arguments
x |
|
... |
Additional arguments |
conf_level |
numeric value for confidence level. Defaults to 0.95. |
type |
"CATEs" or "predict". |
Value
ggplot object.
plot.pattc_deeplearning
Description
Uses plot() to generate histogram of distribution of predicted
outcomes from pattc_deeplearning
Usage
## S3 method for class 'pattc_deeplearning'
plot(x, ...)
Arguments
x |
|
... |
Additional arguments |
Value
ggplot object
plot.pattc_ensemble
Description
Uses plot() to generate histogram of distribution of CATEs or predicted
outcomes from pattc_ensemble
Usage
## S3 method for class 'pattc_ensemble'
plot(x, ...)
Arguments
x |
|
... |
Additional arguments |
Value
ggplot object
plot.pattc_neural
Description
Uses plot() to generate histogram of distribution of CATEs or predicted
outcomes from pattc_neural
Usage
## S3 method for class 'pattc_neural'
plot(x, ...)
Arguments
x |
|
... |
Additional arguments |
Value
ggplot object
World Value Survey India Sample
Description
World Value Survey (WVS) Data for India in 2022. The variables drawn from the said WVS India data match the covariates from the India survey experiment sample.
Usage
data(pop_data)
Format
pop_data
A data frame with 846 rows and 13 columns:
- female
Respondent’s Sex.
- age
Age of respondent.
- income
income group of Household.
- religion
Religious denomination
- practicing_religion
Importance of religion in respondent’s life.
- education
Educational level of respondent.
- political_ideology
Political ideology of respondent.
- employment
Employment status and full-time employee.
- marital_status
Marital status of respondent.
- job_loss
Concern about job loss.
- support_war
Binary (Yes/No) outcome measure for willingness to fight war.
- strong_leader
Binary measure of preference for strong leader.
...
Source
Haerpfer, C., Inglehart, R., Moreno, A., Welzel, C., Kizilova, K., Diez-Medrano J., M. Lagos, P. Norris, E. Ponarin & B. Puranen et al. (eds.). 2020. World Values Survey: Round Seven – Country-Pooled Datafile. Madrid, Spain & Vienna, Austria: JD Systems Institute & WVSA Secretariat. <doi.org/10.14281/18241.1>
World Value Survey India Sample
Description
Extended World Value Survey (WVS) Data for India in 1995, 2001, 2006, 2012, and 2022.
Usage
data(pop_data_full)
Format
pop_data_full
A data frame with 11,813 rows and 13 columns:
- female
Respondent’s Sex.
- age
Age of respondent.
- income
income group of Household.
- religion
Religious denomination
- practicing_religion
Importance of religion in respondent’s life.
- education
Educational level of respondent.
- political_ideology
Political ideology of respondent.
- employment
Employment status and full-time employee.
- marital_status
Marital status of respondent.
- job_loss
Concern about job loss.
- support_war
Binary (Yes/No) outcome measure for willingness to fight war.
- strong_leader
Binary measure of preference for strong leader.
...
Source
Haerpfer, C., Inglehart, R., Moreno, A., Welzel, C., Kizilova, K., Diez-Medrano J., M. Lagos, P. Norris, E. Ponarin & B. Puranen et al. (eds.). 2020. World Values Survey: Round Seven – Country-Pooled Datafile. Madrid, Spain & Vienna, Austria: JD Systems Institute & WVSA Secretariat. <doi.org/10.14281/18241.1>
Create list for population data
Description
create list object of population data for easy data processing
Usage
popcall(
response.formula,
compl.var,
treat.var,
pop.data,
weights = NULL,
cluster = NULL,
ID = NULL,
patt = TRUE
)
Arguments
response.formula |
formula for response equation of binary outcome variable and covariates |
compl.var |
string for complier variable |
treat.var |
string for treatment variable |
pop.data |
|
weights |
observation weights |
cluster |
clustering variable |
ID |
identifier variable |
patt |
logical for patt, subsetting population treated observations |
Value
list of processed dataset
print.metalearner_deeplearning
Description
Print method for metalearner_deeplearning
Usage
## S3 method for class 'metalearner_deeplearning'
print(x, ...)
Arguments
x |
|
... |
additional parameter |
Value
list of model results
print.metalearner_ensemble
Description
Print method for metalearner_ensemble
Usage
## S3 method for class 'metalearner_ensemble'
print(x, ...)
Arguments
x |
|
... |
additional parameter |
Value
list of model results
print.metalearner_neural
Description
Print method for metalearner_neural
Usage
## S3 method for class 'metalearner_neural'
print(x, ...)
Arguments
x |
|
... |
additional parameter |
Value
list of model results
print.pattc_deeplearning
Description
Print method for pattc_deeplearning
Usage
## S3 method for class 'pattc_deeplearning'
print(x, ...)
Arguments
x |
|
... |
additional arguments |
Value
list of model results
print.pattc_ensemble
Description
Print method for pattc_ensemble
Usage
## S3 method for class 'pattc_ensemble'
print(x, ...)
Arguments
x |
|
... |
additional parameter |
Value
list of model results
print.pattc_neural
Description
Print method for pattc_neural
Usage
## S3 method for class 'pattc_neural'
print(x, ...)
Arguments
x |
|
... |
additional parameter |
Value
list of model results
Check for Python module availability and install if missing.
Description
Call this to manually set up Python and dependencies. The function checks if Python is available via the reticulate package, and if not, it creates a virtual environment and installs the specified Python modules.
Usage
python_ready(
modules = c("keras", "tensorflow", "numpy"),
envname = "r-reticulate"
)
Arguments
modules |
Character vector of Python modules to check for and install if missing. |
envname |
Name of the virtual environment to use or create. Defaults to "r-reticulate". |
Value
Invisibly returns TRUE if setup is complete.
Examples
## Not run:
python_ready(modules = c("keras", "tensorflow", "numpy"),
envname = "r-reticulate")
## End(Not run)
Response model from experimental data using SL ensemble
Description
Train response model (response variable as outcome and covariates) from all compliers (actual + predicted) in experimental data using SL ensemble.
Usage
response_model(
response.formula,
exp.data,
compl.var,
exp.compliers,
family = gaussian(),
ID = NULL,
SL.learners = c("SL.glmnet", "SL.xgboost", "SL.ranger", "SL.nnet", "SL.glm")
)
Arguments
response.formula |
formula to fit the response model (y ~ x) using binary outcome variable and covariates |
exp.data |
experimental dataset. |
compl.var |
string specifying binary complier variable |
exp.compliers |
|
family |
gaussian() or binomial(). |
ID |
string for identifier variable. |
SL.learners |
vector of names of ML algorithms used for ensemble model. |
Value
trained response model.