,
Mark Lunt [aut],
Marcus Beasley [aut],
Gary J Macfarlane [aut],
John McBeth [aut]| Title: | Bayesian Causal Mediation Analysis using 'Stan' |
|---|---|
| Description: | Performs parametric mediation analysis using the Bayesian g-formula approach for binary and continuous outcomes. The methodology is based on Comment (2018) <doi:10.5281/zenodo.1285275> and a demonstration of its application can be found at Yimer et al. (2022) <doi:10.48550/arXiv.2210.08499>. |
| Authors: | Belay Birlie Yimer [aut, cre] ,
Mark Lunt [aut],
Marcus Beasley [aut],
Gary J Macfarlane [aut],
John McBeth [aut] |
| Maintainer: | Belay Birlie Yimer <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 0.0.4 |
| Built: | 2025-02-23 05:18:15 UTC |
| Source: | https://github.com/belayb/bayesgmed |
A DESCRIPTION OF THE PACKAGE
Stan Development Team (2022). RStan: the R interface to Stan. R package version 2.21.5. https://mc-stan.org
Estimates various quantities for causal mediation analysis using 'Stan'.
bayesgmed( outcome, mediator, treat, covariates = NULL, dist.y = "continuous", dist.m = "continuous", link.y = "identity", link.m = "identity", data, priors = NULL, ... )bayesgmed( outcome, mediator, treat, covariates = NULL, dist.y = "continuous", dist.m = "continuous", link.y = "identity", link.m = "identity", data, priors = NULL, ... )
outcome |
a character string indicating the name of the outcome variable. |
mediator |
a character string indicating the name of the mediator variable. |
treat |
a character string indicating the name of the treatment variable. The treatment variable is considered binary and should be coded as 0 for control and 1 for treated. |
covariates |
a character vector indicating the name of the confounding variables. |
dist.y |
a character string indicating the family distribution of the outcome. E.g., dist.y = "bernoulli" will fit a logistic regression for the outcome. |
dist.m |
a character string indicating the family distribution of the mediator, E.g., dist.m = "bernoulli" will fit a logistic regression model for the mediator |
link.y |
a character string indicating the link function to be used for the outcome model. |
link.m |
a character string indicating the link function to be used for the mediator model. |
data |
A |
priors |
A list of named values for the prior scale parameters. See details. |
... |
Other optional parameters passed to |
This is the main function for estimating causal mediation effects from several types of data within the Bayesian framework. We followed the potential outcome framework for effects definition and the package uses the 'rstan' utility functions for exploring the posterior distribution.
Users may pass a list of named values for the priors argument. The values will be used to define the scale parameter of the respective prior distributions. This list may specify some or all of the following parameters: priors <- list( scale_m = 2.5diag(P_m) scale_y = 2.5diag(P_y), location_m = rep(0, P_m) location_y = rep(0, P_y), scale_sd_y = 2.5, scale_sd_m = 2.5) where P_m is the number of regression parameters (including the intercept) in the mediator model and P_y is the number of regression parameters in the outcome model.
An object of 'S4' class 'stanfit', with all its available methods.
Belay Birlie Yimer [email protected]
McCandless, L.C. and J.M. Somers, Bayesian sensitivity analysis for unmeasured confounding in causal mediation analysis. Statistical Methods in Medical Research, 2019. (28)(2): p. 515-531.
Comment, L., Coull, B. A., Zigler, C., and Valeri, L. (2019). Bayesian data fusion for unmeasured confounding. arXiv preprint arXiv:1902.10613.
## Run example using the example_data data(example_data) fit1 <- bayesgmed(outcome = "Y", mediator = "M", treat = "A", covariates = c("Z1", "Z2"), dist.y = "binary", dist.m = "binary", link.y = "logit", link.m = "logit", data = example_data) bayesgmed_summary(fit1) # With priors P <- 3 # number of covariates plus the intercept term priors <- list(scale_m = 2.5*diag(P+1), scale_y = 2.5*diag(P+2), location_m = rep(0, P+1), location_y = rep(0, P+2)) fit1 <- bayesgmed(outcome = "Y", mediator = "M", treat = "A", covariates = c("Z1", "Z2"), dist.y = "binary", dist.m = "binary", link.y = "logit", link.m = "logit", priors = priors, data = example_data) bayesgmed_summary(fit1)## Run example using the example_data data(example_data) fit1 <- bayesgmed(outcome = "Y", mediator = "M", treat = "A", covariates = c("Z1", "Z2"), dist.y = "binary", dist.m = "binary", link.y = "logit", link.m = "logit", data = example_data) bayesgmed_summary(fit1) # With priors P <- 3 # number of covariates plus the intercept term priors <- list(scale_m = 2.5*diag(P+1), scale_y = 2.5*diag(P+2), location_m = rep(0, P+1), location_y = rep(0, P+2)) fit1 <- bayesgmed(outcome = "Y", mediator = "M", treat = "A", covariates = c("Z1", "Z2"), dist.y = "binary", dist.m = "binary", link.y = "logit", link.m = "logit", priors = priors, data = example_data) bayesgmed_summary(fit1)
'bayesgmed_sens' is used to conduct sensitivity analysis for unmeasured confounders in mediation analysis.
bayesgmed_sens( outcome, mediator, treat, covariates = NULL, dist.y = "continuous", dist.m = "continuous", link.y = "identity", link.m = "identity", data, priors = NULL, ... )bayesgmed_sens( outcome, mediator, treat, covariates = NULL, dist.y = "continuous", dist.m = "continuous", link.y = "identity", link.m = "identity", data, priors = NULL, ... )
outcome |
a character string indicating the name of the outcome variable. |
mediator |
a character string indicating the name of the mediator variable. |
treat |
a character string indicating the name of the treatment variable. The treatment variable is considered binary and should be coded as 0 for control and 1 for treated. |
covariates |
a character vector indicating the name of the confounding variables. |
dist.y |
a character string indicating the family distribution of the outcome. E.g., dist.y = "bernoulli" will fit a logistic regression the outcome. |
dist.m |
a character string indicating the family distribution of the mediator E.g., dist.m = "bernoulli" will fit a logistic regression the mediator |
link.y |
a character string indicating the link function to be used for the outcome model. |
link.m |
a character string indicating the link function to be used for the mediator model. |
data |
A |
priors |
A list of named values for the prior scale parameters. See details. |
... |
Other optional parameters passed to |
Perform a sensitivity analysis for unmeasured confounding following the approach proposed by McCandless LC and Somers JM (2019). This is done by incorporating uncertainty about unmeasured confounding in the outcome and mediator model through a prior distribution. One can control the size of unmeasured confounding by varying the location and scale parameters of the prior distribution values for the bias parameter (i.e., gamma). See the Vignette for more details.
Users may pass a list of named values for the priors argument. The values will be used to define
the scale parameter of the respective prior distributions. This list may specify some or all of the
following parameters:
priors <- list(
scale_m = 2.5diag(P_m) scale_y = 2.5diag(P_y),
llocation_m = rep(0, P_m) location_y = rep(0, P_y),
location_gamma = rep(0, 4), scale_gamma = 0.1*diag(4),
scale_sd_y = 2.5, scale_sd_m = 2.5)
where P_m is the number of regression parameters (including the intercept) in the mediator model and
P_y is the number of regression parameters in the outcome model. Note that there are 4 bias parameters
i.e., gamma).
An object of 'S4' class 'stanfit', with all its available methods.
Belay Birlie Yimer [email protected]
McCandless, L.C. and J.M. Somers, Bayesian sensitivity analysis for unmeasured confounding in causal mediation analysis. Statistical Methods in Medical Research, 2019. (28)(2): p. 515-531.
Comment, L., Coull, B. A., Zigler, C., and Valeri, L. (2019). Bayesian data fusion for unmeasured confounding. arXiv preprint arXiv:1902.10613.
## Run example using the example_data data(example_data) # priors including for unmeasured single confounder P <- 3 # number of covariates plus the intercept term priors <- list(scale_m = 2.5*diag(P+1), scale_y = 2.5*diag(P+2), location_m = rep(0, P+1), location_y = rep(0, P+2), location_gamma = rep(0,4), scale_gamma = 0.5*diag(4)) fit1 <- bayesgmed_sens(outcome = "Y", mediator = "M", treat = "A", covariates = c("Z1", "Z2"), dist.y = "binary", dist.m = "binary", link.y = "logit", link.m = "logit", priors = priors, data = example_data) bayesgmed_summary(fit1)## Run example using the example_data data(example_data) # priors including for unmeasured single confounder P <- 3 # number of covariates plus the intercept term priors <- list(scale_m = 2.5*diag(P+1), scale_y = 2.5*diag(P+2), location_m = rep(0, P+1), location_y = rep(0, P+2), location_gamma = rep(0,4), scale_gamma = 0.5*diag(4)) fit1 <- bayesgmed_sens(outcome = "Y", mediator = "M", treat = "A", covariates = c("Z1", "Z2"), dist.y = "binary", dist.m = "binary", link.y = "logit", link.m = "logit", priors = priors, data = example_data) bayesgmed_summary(fit1)
Print a summary of the estimated causal mediation model
bayesgmed_summary( model = NULL, level = 0.95, pars = c("NDE_control", "NDE_treated", "NIE_control", "NIE_treated", "ANDE", "ANIE", "TE"), digits = 3 )bayesgmed_summary( model = NULL, level = 0.95, pars = c("NDE_control", "NDE_treated", "NIE_control", "NIE_treated", "ANDE", "ANIE", "TE"), digits = 3 )
model |
A |
level |
The "confidence" level that defines the limits of the credible intervals (default is .95, i.e. 95% CIs). |
pars |
The parameters to summarize (defaults to main causal effect estimands similar to the |
digits |
The number of decimal points to display in the output (default is 3). |
After estimating a model with bayesgmed(), use bayesgmed_summary(fit) to display the estimated results,
where fit is an object containing the fitted model. By default, bayesgmed_summary() only displays a subset of
the estimated parameters:
- NDE_control: direct effect estimate when the exposure level is set to the control value.
- NDE_treated: direct effect estimate when the exposure level is set to the treated value.
- NIE_control: mediated effect estimate when the exposure level is set to the control value.
- NIE_treated: mediated effect estimate when the exposure level is set to the treated value.
- ANDE: average direct effect of X on Y.
- ANIE: average indirect effect of X on Y.
- TE: the total effect of A on Y.
To display all estimated parameters where all chains merged, set pars = NULL. This will print all parameters defined
in the model definitions, including the most important ones:
- alphaZ[]: parameter estimate of the confounders (i.e., X -> Y) relationship, listed in the order they are specified
in the covariates argument of bayesgmed(). alpha[1] is the intercept.
- alphaM: parameter estimate of the M -> Y relationship.
- alphaA: parameter estimate of the A -> Y relationship.
- betaZ: parameter estimate of the confounders (i.e., X -> M) relationship, listed in the order they are specified
in the covariates argument of bayesgmed(). beta[1] is the intercept
To learn more about the additional parameters, refer to the Stan code
(cat(get_stancode(fit))).
A data.frame summarizing the estimated causal mediation model, including the following columns:
Parameter: the name of the parameter.
Mean: the mean of the parameter's posterior distribution.
Median: the median of the parameter's posterior distribution.
SE: the standard deviation of the parameter's posterior distribution.
ci_lwr: the lower limit of the credible interval.
ci_upr: the upper limit of the credible interval.
n_eff: the number of efficient samples.
Rhat: a value of 1.00 suggests model convergence.
Belay B. Yimer [email protected]
The data set contains a binary outcome variable Y, a binary exposure A, a binary mediator M, and two confounding variables (Z1 and Z2) for 300 individuals.
example_dataexample_data
example_dataA data frame with 300 rows and 5 columns:
confounders
binary exposure
binary mediator
binary outcome
...