Package 'causens'

Bayesian parametric sensitivity analysis for causal inference

Description

This function runs a Bayesian sensitivity analysis for causal inference using JAGS or Stan as a backend. For now, only JAGS is supported.

Usage

bayesian_causens(
  trt_model,
  outcome_model,
  U_model,
  data,
  beta_uy = ~dunif(-2, 2),
  alpha_uz = ~dunif(-2, 2),
  backend = "jags",
  output_trace = FALSE,
  ...
)

Arguments

trt_model	The treatment model object as a formula.
outcome_model	The outcome model object as a formula.
U_model	The unmeasured confounder model object as a formula.
data	A data frame containing the exposure, outcome, and confounder variables.
beta_uy	Prior distribution for the effect of the missing confounder U on the outcome Y.
alpha_uz	Prior distribution for the effect of the missing confounder U on the treatment assignment mechanism Z.
backend	The backend to use for the sensitivity analysis. Currently only "jags" is supported.
output_trace	Whether to output the full trace of the MCMC sampler.
...	Additional arguments to be passed to the backend.

Value

A list of posterior samples for the causal effect of the exposure variable on the outcome, as well as the confounder-adjusted causal effect.

---

Monte Carlo sensitivity analysis for causal effects

Description

This function performs a Monte Carlo sensitivity analysis for causal effects.

Usage

causens_monte_carlo(outcome, exposure, confounders, data, ...)

Arguments

outcome	The name of the outcome variable in the data frame.
exposure	The name of the exposure variable in the data frame.
confounders	The name of the confounders in the data frame.
data	A data frame containing the exposure, outcome, and confounder variables.
...	Additional arguments to be passed to the function.

Value

The estimated causal effect.

---

Bayesian Estimation of ATE Subject to Unmeasured Confounding

Description

This function provides an estimate of the Average Treatment Effect (ATE) using Bayesian modelling.

Usage

causens_sf(
  trt_model,
  outcome,
  data,
  bootstrap = FALSE,
  B = 1000,
  seed = 123,
  ...
)

Arguments

trt_model	A model formula specifying the treatment model.
outcome	The name of the outcome variable.
data	A data frame containing the exposure, outcome, and confounder variables.
bootstrap	A logical indicating whether to perform bootstrap estimation of the 95% confidence interval.
B	If the bootstrap argument is TRUE, the number of bootstrap samples to generate.
seed	An integer to set the random seed for reproducibility.
...	Additional arguments to be passed to the sensitivity function.

Value

A point estimate of the corrected ATE.

---

Create an JAGS model for Bayesian sensitivity analysis

Description

Creates a JAGS model available as a string, or .txt file, where priors are initialized to be uninformative by default.

Usage

create_jags_model(binary_outcome, beta_uy, alpha_uz)

Arguments

binary_outcome	Boolean indicating whether the outcome is binary.
beta_uy	Prior distribution for the effect of the missing confounder U on the outcome Y.
alpha_uz	Prior distribution for the effect of the missing confounder U on the treatment assignment mechanism Z. No inputs are given to this function (for now) since data-related information is provided in jags.model() during model initialization.

---

Generate data with a binary unmeasured confounder and binary outcome

Description

Generate data with a binary unmeasured confounder and binary outcome

Usage

gData_U_binary_Y_binary(
  ymodel = "linear",
  N = 500,
  alpha_uz = 0.2,
  beta_uy = 0.5,
  treatment_effects = 1,
  seed = 123
)

Arguments

ymodel	A string indicating the functional form of the outcome model.
N	The number of observations to be generated.
alpha_uz	Unmeasured confounder coefficient in the propensity score model.
beta_uy	Unmeasured confounder coefficient in the outcome model.
treatment_effects	The treatment effect.
seed	The seed for the random number generator.

Value

A data frame with the simulated dataset with U binary, Y binary.

---

Generate data with a binary unmeasured confounder and continuous outcome

Description

Generate data with a binary unmeasured confounder and continuous outcome

Usage

gData_U_binary_Y_cont(
  ymodel = "linear",
  N = 500,
  alpha_uz = 0.2,
  beta_uy = 0.5,
  treatment_effects = 1,
  seed = 123
)

Arguments

ymodel	A string indicating the functional form of the outcome model.
N	The number of observations to be generated.
alpha_uz	Unmeasured confounder coefficient in the propensity score model.
beta_uy	Unmeasured confounder coefficient in the outcome model.
treatment_effects	The treatment effect.
seed	The seed for the random number generator.

Value

A data frame with the simulated dataset with U binary, Y continuous.

Examples

fulldata <- gData_U_binary_Y_cont(
  ymodel = "linear",
  N = 500,
  alpha_uz = 0.2,
  beta_uy = 0.5,
  treatment_effects = 1,
  seed = 123
)
table(fulldata$Z)

---

Generate data with a continuous unmeasured confounder and a binary outcome

Description

Generate data with a continuous unmeasured confounder and a binary outcome

Usage

gData_U_cont_Y_binary(
  ymodel = "linear",
  N = 500,
  alpha_uz = 0.2,
  beta_uy = 0.5,
  treatment_effects = 1,
  seed = 123
)

Arguments

ymodel	A string indicating the functional form of the outcome model.
N	The number of observations to be generated.
alpha_uz	Unmeasured confounder coefficient in the propensity score model.
beta_uy	Unmeasured confounder coefficient in the outcome model.
treatment_effects	The treatment effect.
seed	The seed for the random number generator.

Value

A data frame with the simulated dataset with U continuous, Y binary.

---

Generate data with a continuous unmeasured confounder and continuous outcome

Description

Generate data with a continuous unmeasured confounder and continuous outcome

Usage

gData_U_cont_Y_cont(
  ymodel = "linear",
  N = 500,
  alpha_uz = 0.2,
  beta_uy = 0.5,
  treatment_effects = 1,
  seed = 123
)

Arguments

ymodel	A string indicating the functional form of the outcome model.
N	The number of observations to be generated.
alpha_uz	Unmeasured confounder coefficient in the propensity score model.
beta_uy	Unmeasured confounder coefficient in the outcome model.
treatment_effects	The treatment effect.
seed	The seed for the random number generator.

Value

A data frame with the simulated dataset with U continuous, Y continuous.

---

Plot ATE with respect to sensitivity function value when it is constant, i.e. c(1, e) = c1 and c(0, e) = c0.

Description

This function plots 1) the ATE as a function of the sensitivity function value when it is constant and 2) its associate 95 interval obtained via bootstrapping, if desired. The latter process can be take a few seconds to minutes.

Usage

plot_causens(
  trt_model,
  data,
  outcome,
  c1_upper = 0.5,
  c1_lower = 0,
  r = 1,
  by = 0.01
)

Arguments

trt_model	The treatment model object as a formula or fitted glm.
data	A data frame containing the variables of interest.
outcome	The name of the outcome variable.
c1_upper	The upper bound for the sensitivity function value.
c1_lower	The lower bound for the sensitivity function value.
r	The ratio between c1 and c0.
by	The increment for the sensitivity function value.

Value

A plot of the ATE as a function of c1 values.

---

Process model formula

Description

This helper function takes in a model in its 'formula' format and returns a list comprising of various information, including the fitted model as an R object.

Usage

process_model_formula(model, data)

Arguments

model	The model object as a formula or fitted glm.
data	A data frame used for fitting the model.

Value

A list containing the fitted model, model formula, response variable name, and confounder names.

---

Calculate sensitivity of treatment effect estimate to unmeasured confounding

Description

This function calculates the sensitivity of a treatment effect estimate to unmeasured confounding, as described in Rosenbaum (2002). The sensitivity is defined as the maximum strength of association between the unmeasured confounder and the treatment assignment that would be needed to explain away the observed treatment effect estimate. This function assumes that the treatment assignment is binary and that the outcome is continuous.

Usage

sf(z, e, form = "constant", c1 = 0, c0 = 0, s1 = 0, s0 = 0)

Arguments

z	Treatment assignment (binary: 0 or 1)
e	Propensity score value (numeric)
form	Form of the sensitivity function (character: "constant" or "linear")
c1	Value of the sensitivity function when z = 1 (numeric)
c0	Value of the sensitivity function when z = 0 (numeric)
s1	Slope of the sensitivity function when z = 1 (numeric)
s0	Slope of the sensitivity function when z = 0 (numeric)

Value

Sensitivity of treatment effect estimate to unmeasured confounding (numeric)

sf(z, e, form)

---

Generate data with unmeasured confounder

Description

Generate data with unmeasured confounder

Usage

simulate_data(
  ymodel = "linear",
  N = 500,
  u_type = "binary",
  y_type = "continuous",
  seed = 123,
  alpha_uz = 0.2,
  beta_uy = 0.5,
  treatment_effects = 1,
  informative_u = FALSE
)

Arguments

ymodel	A string indicating the functional form of the outcome model.
N	The number of observations to be generated.
u_type	A string indicating the type of the unmeasured confounder: "binary" or "continuous".
y_type	A string indicating the type of the outcome: "binary" or "continuous".
seed	The seed for the random number generator.
alpha_uz	Unmeasured confounder coefficient in the propensity score model.
beta_uy	Unmeasured confounder coefficient in the outcome model.
treatment_effects	The treatment effect.
informative_u	A boolean indicating whether the unmeasured confounder is driven by covariates.

Value

A data frame with the simulated dataset.

---

Summarize the results of a causal sensitivity analysis via Bayesian modelling of an unmeasured confounder.

Description

Summarize the results of a causal sensitivity analysis via Bayesian modelling of an unmeasured confounder.

Usage

## S3 method for class 'bayesian_causens'
summary(object, ...)

Arguments

object	An object of class bayesian_causens.
...	Additional arguments to be passed to summary.

Value

A summary of results from a Bayesian causal sensitivity analysis.

---

Summarize the results of a causal sensitivity analysis via sensitivity function.

Description

Summarize the results of a causal sensitivity analysis via sensitivity function.

Usage

## S3 method for class 'causens_sf'
summary(object, ...)

Arguments

object	An object of class causens_sf.
...	Additional arguments to be passed to summary.

Value

A summary of the results of the causal sensitivity analysis.

---

Summarize the results of a causal sensitivity analysis via the Monte Carlo method.

Description

Summarize the results of a causal sensitivity analysis via the Monte Carlo method.

Usage

## S3 method for class 'monte_carlo_causens'
summary(object, ...)

Arguments

object	An object of class causens_monte_carlo.
...	Additional arguments to be passed to summary.

Value

A summary of results from a Monte Carlo causal sensitivity analysis.

---