| Type: | Package |
| Title: | Bayesian Post-Estimation on the Mean Scale |
| Version: | 0.2.1 |
| URL: | https://github.com/dalenbe2/bayesMeanScale |
| BugReports: | https://github.com/dalenbe2/bayesMeanScale/issues |
| Description: | Computes Bayesian posterior distributions of predictions, marginal effects, and differences of marginal effects for various generalized linear models. Importantly, the posteriors are on the mean (response) scale, allowing for more natural interpretation than summaries on the link scale. Also, predictions and marginal effects of the count probabilities for Poisson and negative binomial models can be computed. |
| License: | GPL (≥ 3) |
| Encoding: | UTF-8 |
| Imports: | bayestestR (≥ 0.13.2), data.table (≥ 1.15.2), magrittr (≥ 2.0.3), posterior (≥ 1.5.0) |
| Depends: | R (≥ 3.5.0) |
| Suggests: | flextable (≥ 0.9.5), knitr (≥ 1.45), MASS (≥ 7.3-60.2), rmarkdown (≥ 2.26), rstanarm (≥ 2.32.1), testthat (≥ 3.0.0) |
| Config/testthat/edition: | 3 |
| VignetteBuilder: | knitr |
| NeedsCompilation: | no |
| Packaged: | 2025-01-03 14:59:31 UTC; dalenbed |
| Author: | David M. Dalenberg [aut, cre] |
| Maintainer: | David M. Dalenberg <dalenbe2@gmail.com> |
| Repository: | CRAN |
| Date/Publication: | 2025-01-08 13:30:02 UTC |
Bayesian marginal effects on count probabilities.
Description
Computes Bayesian marginal effects on count probabilities. Models must have a family of "poisson"" or "neg_binomial_2" and be fit using the "rstanarm" package. Marginal effects can be averaged over the values of the covariates in the data (average marginal effects), or the covariates can be held at their means (marginal effects at the means). Also, "at" values can be specified to fix covariates at particular values. The marginal effects can be specified in terms of discrete changes or instantaneous rates of change.
Usage
bayesCountMargEffF(model,
counts,
marginal_effect,
start_value,
end_value,
n_draws = 2000,
ci = .95,
hdi_interval = TRUE,
centrality = 'mean',
digits = 4,
at = NULL,
at_means = FALSE,
h = .0001)
Arguments
model |
A model object of class "stanreg." |
counts |
Vector of counts to get the predicted probabilities for. |
marginal_effect |
A character vector of terms that you want to estimate the marginal effect for. |
start_value |
The start value(s) for the marginal effect(s). Must be a list when specifying multiple values. Specify "instantaneous" when you want to calculate an approximate instantaneous rate of change. |
end_value |
The end value(s) for the marginal effect(s). Must be a list when specifying multiple values. Specify "instantaneous" when you want to calculate an approximate instantaneous rate of change. |
n_draws |
The number of draws to take from the joint posterior distribution. |
ci |
The level for the credible intervals. |
hdi_interval |
If TRUE, the default, computes the highest density credible interval. If FALSE, computes the equal-tailed interval. |
centrality |
Centrality measure for the posterior distribution. Options are "mean" or "median". |
digits |
The number of digits to report in the summary table. |
at |
Optional list of covariate values to estimate the marginal effect at. |
at_means |
If FALSE, the default, the marginal effects are averaged across the rows of the data. If TRUE, the covariate values that are not specified in the "at" argument are held at their means. |
h |
Step size for calculating the approximate instantaneous rate of change. |
Details
Marginal effects on count probabilities give you useful information to complement post estimation summaries on the mean scale.
Value
A list of class "bayesmeanscale_marg" with the following components:
diffTable |
summary table of the marginal effects |
diffDraws |
posterior draws of the marginal effects |
Author(s)
David Dalenberg
References
Long, J. Scott and Jeremy Freese. 2001. "Predicted Probabilities for Count Models." Stata Journal 1(1): 51-57.
Examples
if(require(rstanarm)){
crabs <- read.table("https://users.stat.ufl.edu/~aa/cat/data/Crabs.dat",
header = TRUE)
poissonModel <- stan_glm(sat ~ weight + width,
data = crabs,
family = poisson,
refresh = 0,
iter = 500)
# marginal effect holding covariates at their means #
bayesCountMargEffF(poissonModel,
counts = 1,
marginal_effect = 'width',
start_value = 25,
end_value = 20,
at_means = TRUE,
n_draws = 500)
}
Bayesian predictions of count probabilities.
Description
Computes Bayesian predictions for count probabilities. Models must have a family of "poisson"" or "neg_binomial_2" and be fit using the "rstanarm" package. Predictions can be averaged over the values of the covariates in the data (average marginal predictions), or the covariates can be held at their means (marginal predictions at the means). Also, "at" values must be specified to fix at least one covariate at particular values.
Usage
bayesCountPredsF(model,
counts,
at,
n_draws = 2000,
ci = .95,
hdi_interval = TRUE,
centrality = 'mean',
digits = 4,
at_means = FALSE,
data_slice = 'full')
Arguments
model |
A model object of class "stanreg." |
counts |
Vector of counts to get the predicted probabilities for. |
at |
List of covariate values to estimate the predictions at. |
n_draws |
The number of draws to take from the joint posterior distribution. |
ci |
The level for the credible intervals. |
hdi_interval |
If TRUE, the default, computes the highest density credible interval. If FALSE, computes the equal-tailed interval. |
centrality |
Centrality measure for the posterior distribution. Options are "mean" or "median". |
digits |
The number of digits to report in the summary table. |
at_means |
If FALSE, the default, the predictions are averaged across the rows of the model data for each unique combination of "at" values. If TRUE, the covariate values that are not specified in the "at" argument are held at their means. |
data_slice |
The number of rows of data to average over for the predictions. Defaults to all rows. This can be useful for very large data sets. |
Details
Predicted count probabilities give you useful information to complement post estimation summaries of the mean.
Value
A list of class "bayesmeanscale_pred" with the following components:
predTable |
summary table of the predictions |
predDraws |
posterior draws of the predictions |
Author(s)
David Dalenberg
References
Long, J. Scott and Jeremy Freese. 2001. "Predicted Probabilities for Count Models." Stata Journal 1(1): 51-57.
Examples
if(require(rstanarm)){
crabs <- read.table("https://users.stat.ufl.edu/~aa/cat/data/Crabs.dat",
header = TRUE)
poissonModel <- stan_glm(sat ~ weight + width,
data = crabs,
family = poisson,
refresh = 0,
iter = 500)
bayesCountPredsF(poissonModel,
counts = c(0,1),
at = list(weight=c(2,3)),
n_draws = 500)
}
Comparison of Bayesian marginal effects
Description
Tests the differences between all marginal effects in the "bayesmeanscale_marg" object that is passed to it. This is particularly useful for testing non-linear interaction effects.
Usage
bayesMargCompareF(marg_list,
ci = 0.95,
hdi_interval = TRUE,
centrality = 'mean',
digits = 4)
Arguments
marg_list |
An object of class "bayesmeanscale_marg." |
ci |
The level of the credible interval. |
hdi_interval |
If TRUE, the default, computes the highest density credible interval. If FALSE, computes the equal-tailed interval. |
centrality |
Centrality measure for the posterior distribution. Options are "mean" or "median". |
digits |
The number of digits to report in the summary table. |
Details
Differences of marginal effects contained in the "bayesmeanscale_marg" object are computed.
Value
A list of class "bayesmeanscale_margcompare" with the following components:
diffTable |
summary table of the differences of the marginal effects |
diffDraws |
posterior draws of the differences of the marginal effects |
Author(s)
David Dalenberg
References
Mize, Trenton D. 2019. "Best Practices for Estimating, Interpreting, and Presenting Non-linear Interaction Effects." Sociological Science 6: 81-117.
Examples
## Logit model ##
if(require(rstanarm)){
m1 <- stan_glm(switch ~ dist + educ + arsenic + assoc,
data = rstanarm::wells,
family = binomial,
refresh = 0,
iter = 500)
m1Marg <- bayesMargEffF(m1,
marginal_effect = 'arsenic',
start_value = 2.2,
end_value = .82,
at = list(educ=c(0, 5)),
n_draws = 500)
bayesMargCompareF(m1Marg)
}
Bayesian marginal effects on the mean scale.
Description
Computes Bayesian marginal effects on the mean scale for models fit using the package "rstanarm". Marginal effects can be averaged over the values of the covariates in the data (average marginal effects), or the covariates can be held at their means (marginal effects at the means). Also, "at" values can be specified to fix covariates at particular values. The marginal effects can be specified in terms of discrete changes or instantaneous rates of change.
Usage
bayesMargEffF(model,
n_draws = 2000,
marginal_effect,
start_value,
end_value,
ci = .95,
hdi_interval = TRUE,
centrality = 'mean',
digits = 4,
at = NULL,
at_means = FALSE,
h = .0001)
Arguments
model |
A model object of class "stanreg." |
n_draws |
The number of draws to take from the joint posterior distribution. |
marginal_effect |
A character vector of terms that you want to estimate the marginal effect for. |
start_value |
The start value(s) for the marginal effect(s). Must be a list when specifying multiple values. Specify "instantaneous" when you want to calculate an approximate instantaneous rate of change. |
end_value |
The end value(s) for the marginal effect(s). Must be a list when specifying multiple values. Specify "instantaneous" when you want to calculate an approximate instantaneous rate of change. |
ci |
The level for the credible intervals. |
hdi_interval |
If TRUE, the default, computes the highest density credible interval. If FALSE, computes the equal-tailed interval. |
centrality |
Centrality measure for the posterior distribution. Options are "mean" or "median". |
digits |
The number of digits to report in the summary table. |
at |
Optional list of covariate values to estimate the marginal effect at. |
at_means |
If FALSE, the default, the marginal effects are averaged across the rows of the data. If TRUE, the covariate values that are not specified in the "at" argument are held at their means. |
h |
Step size for calculating the approximate instantaneous rate of change. |
Details
The following families of fixed-effect models fit using "rstanarm" are supported: 'beta', 'binomial', 'Gamma', 'gaussian', 'neg_binomial_2', and 'poisson.'
Value
A list of class "bayesmeanscale_marg" with the following components:
diffTable |
summary table of the marginal effects |
diffDraws |
posterior draws of the marginal effects |
Author(s)
David Dalenberg
References
Agresti, Alan. 2013. Categorical Data Analysis. Third Edition. New York: Wiley
Long, J. Scott and Sarah A. Mustillo. 2018. "Using Predictions and Marginal Effects to Compare Groups in Regression Models for Binary Outcomes." Sociological Methods & Research 50(3): 1284-1320.
Mize, Trenton D. 2019. "Best Practices for Estimating, Interpreting, and Presenting Non-linear Interaction Effects." Sociological Science 6: 81-117.
Examples
## Logit model ##
if(require(rstanarm)){
m1 <- stan_glm(switch ~ dist + educ + arsenic + assoc,
data = rstanarm::wells,
family = binomial,
refresh = 0,
iter = 500)
# marginal effect averaging over sample values of covariates #
bayesMargEffF(m1,
marginal_effect = 'arsenic',
start_value = 2.2,
end_value = .82,
n_draws = 500)
}
Bayesian marginal effects on the mean scale.
Description
Computes Bayesian marginal effects on the probability of outcomes for cumulative link models with a proportional odds structure fit using the package "rstanarm". Marginal effects can be averaged over the values of the covariates in the data (average marginal effects), or the covariates can be held at their means (marginal effects at the means). Also, "at" values can be specified to fix covariates at particular values. The marginal effects can be specified in terms of discrete changes or instantaneous rates of change.
Usage
bayesOrdinalMargEffF(model,
n_draws = 2000,
marginal_effect,
start_value,
end_value,
ci = .95,
hdi_interval = TRUE,
centrality = 'mean',
digits = 4,
at = NULL,
at_means = FALSE,
h = .0001)
Arguments
model |
A model object of class "stanreg" and "polr." |
n_draws |
The number of draws to take from the joint posterior distribution. |
marginal_effect |
A character vector of terms that you want to estimate the marginal effect for. |
start_value |
The start value(s) for the marginal effect(s). Must be a list when specifying multiple values. Specify "instantaneous" when you want to calculate an approximate instantaneous rate of change. |
end_value |
The end value(s) for the marginal effect(s). Must be a list when specifying multiple values. Specify "instantaneous" when you want to calculate an approximate instantaneous rate of change. |
ci |
The level for the credible intervals. |
hdi_interval |
If TRUE, the default, computes the highest density credible interval. If FALSE, computes the equal-tailed interval. |
centrality |
Centrality measure for the posterior distribution. Options are "mean" or "median". |
digits |
The number of digits to report in the summary table. |
at |
Optional list of covariate values to estimate the marginal effect at. |
at_means |
If FALSE, the default, the marginal effects are averaged across the rows of the data. If TRUE, the covariate values that are not specified in the "at" argument are held at their means. |
h |
Step size for calculating the approximate instantaneous rate of change. |
Details
The following links for fixed-effect ordinal models fit using "rstanarm" are supported: logit, probit, and clogclog.
Value
A list of class "bayesmeanscale_marg" with the following components:
diffTable |
summary table of the marginal effects |
diffDraws |
posterior draws of the marginal effects |
Author(s)
David Dalenberg
References
Agresti, Alan. 2013. Categorical Data Analysis. Third Edition. New York: Wiley
Long, J. Scott and Sarah A. Mustillo. 2018. "Using Predictions and Marginal Effects to Compare Groups in Regression Models for Binary Outcomes." Sociological Methods & Research 50(3): 1284-1320.
Mize, Trenton D. 2019. "Best Practices for Estimating, Interpreting, and Presenting Non-linear Interaction Effects." Sociological Science 6: 81-117.
Examples
## Proportional odds logistic regression ##
if(require(rstanarm) & require(MASS)){
m1 <- stan_polr(Sat ~ Infl + Type,
data = MASS::housing,
prior = rstanarm::R2(0.2, 'mean'),
refresh = 0,
iter = 500)
# marginal effects holding covariates at means #
bayesOrdinalMargEffF(m1,
marginal_effect = "Infl",
start_value = "Low",
end_value = "High",
at = list(Type = c("Tower", "Apartment")),
at_means = TRUE,
n_draws = 500)
}
Bayesian predictions for ordinal models on the mean scale.
Description
Computes Bayesian outcome predictions for cumulative link models with a proportional odds structure fit using the package "rstanarm". Predictions can be averaged over the values of the covariates in the data (average marginal predictions), or the covariates can be held at their means (marginal predictions at the means). Also, "at" values must be specified to fix at least one covariate at particular values.
Usage
bayesOrdinalPredsF(model,
at,
n_draws = 2000,
ci = .95,
hdi_interval = TRUE,
centrality = 'mean',
digits = 4,
at_means = FALSE,
data_slice = 'full')
Arguments
model |
A model object of class "stanreg" and "polr." |
at |
List of covariate values to estimate the predictions at. |
n_draws |
The number of draws to take from the joint posterior distribution. |
ci |
The level for the credible intervals. |
hdi_interval |
If TRUE, the default, computes the highest density credible interval. If FALSE, computes the equal-tailed interval. |
centrality |
Centrality measure for the posterior distribution. Options are "mean" or "median". |
digits |
The number of digits to report in the summary table. |
at_means |
If FALSE, the default, the predictions are averaged across the rows of the model data for each unique combination of "at" values. If TRUE, the covariate values that are not specified in the "at" argument are held at their means. |
data_slice |
The number of rows of data to average over for the predictions. Defaults to all rows. This can be useful for very large data sets. |
Details
The following links for fixed-effect ordinal models fit using "rstanarm" are supported: logit, probit, and clogclog.
Value
A list of class "bayesmeanscale_pred" with the following components:
predTable |
summary table of the predictions |
predDraws |
posterior draws of the predictions |
Author(s)
David Dalenberg
References
Agresti, Alan. 2013. Categorical Data Analysis. Third Edition. New York: Wiley
Long, J. Scott and Sarah A. Mustillo. 2018. "Using Predictions and Marginal Effects to Compare Groups in Regression Models for Binary Outcomes." Sociological Methods & Research 50(3): 1284-1320.
Mize, Trenton D. 2019. "Best Practices for Estimating, Interpreting, and Presenting Non-linear Interaction Effects." Sociological Science 6: 81-117.
Examples
## Proportional odds logistic regression ##
if(require(rstanarm) & require(MASS)){
m1 <- stan_polr(Sat ~ Infl + Type,
data = MASS::housing,
prior = rstanarm::R2(0.2, 'mean'),
refresh = 0,
iter = 500)
# marginal predictions holding covariates at means #
bayesOrdinalPredsF(m1,
at = list(Type = c("Tower", "Apartment")),
at_means = TRUE,
n_draws = 500)
}
Bayesian predictions on the mean scale.
Description
Computes Bayesian predictions on the mean scale for models fit using the package "rstanarm". Predictions can be averaged over the values of the covariates in the data (average marginal predictions), or the covariates can be held at their means (marginal predictions at the means). Also, "at" values must be specified to fix at least one covariate at particular values.
Usage
bayesPredsF(model,
at,
n_draws = 2000,
ci = .95,
hdi_interval = TRUE,
centrality = 'mean',
digits = 4,
at_means = FALSE,
data_slice = 'full')
Arguments
model |
A model object of class "stanreg." |
at |
List of covariate values to estimate the predictions at. |
n_draws |
The number of draws to take from the joint posterior distribution. |
ci |
The level for the credible intervals. |
hdi_interval |
If TRUE, the default, computes the highest density credible interval. If FALSE, computes the equal-tailed interval. |
centrality |
Centrality measure for the posterior distribution. Options are "mean" or "median". |
digits |
The number of digits to report in the summary table. |
at_means |
If FALSE, the default, the predictions are averaged across the rows of the model data for each unique combination of "at" values. If TRUE, the covariate values that are not specified in the "at" argument are held at their means. |
data_slice |
The number of rows of data to average over for the predictions. Defaults to all rows. This can be useful for very large data sets. |
Details
The following families of fixed-effect models fit using "rstanarm" are supported: 'beta', 'binomial', 'Gamma', 'gaussian', 'neg_binomial_2', and 'poisson.'
Value
A list of class "bayesmeanscale_pred" with the following components:
predTable |
summary table of the predictions |
predDraws |
posterior draws of the predictions |
Author(s)
David Dalenberg
References
Agresti, Alan. 2013. Categorical Data Analysis. Third Edition. New York: Wiley
Long, J. Scott and Sarah A. Mustillo. 2018. "Using Predictions and Marginal Effects to Compare Groups in Regression Models for Binary Outcomes." Sociological Methods & Research 50(3): 1284-1320.
Mize, Trenton D. 2019. "Best Practices for Estimating, Interpreting, and Presenting Non-linear Interaction Effects." Sociological Science 6: 81-117.
Examples
## Logit model ##
if(require(rstanarm)){
m1 <- stan_glm(switch ~ dist + educ + arsenic + assoc,
data = rstanarm::wells,
family = binomial,
refresh = 0,
iter = 500)
# marginal predictions holding covariates at means #
bayesPredsF(m1,
at = list(arsenic = c(.82, 1.3)),
at_means = TRUE,
n_draws = 500)
}