Methods for objects of class rctglm — rctglm

Methods for extracting information about the estimand and about the underlying GLM fit from an rctglm object. See details for information on each method.

Usage

estimand(object)

# S3 method for class 'rctglm'
estimand(object)

est(object)

# S3 method for class 'rctglm'
est(object)

# S3 method for class 'rctglm'
coef(object, ...)

# S3 method for class 'rctglm'
predict(object, ...)

# S3 method for class 'rctglm'
print(x, digits = max(3L, getOption("digits") - 3L), ...)

# S3 method for class 'rctglm'
summary(object, level = 0.95, ...)

# S3 method for class 'summary.rctglm'
print(x, digits = max(3L, getOption("digits") - 3L), ...)

# S3 method for class 'rctglm'
plot(x, level = 0.95, ...)

# S3 method for class 'rctglm'
formula(x, ...)

# S3 method for class 'rctglm'
family(object, ...)

# S3 method for class 'rctglm'
fitted(object, ...)

# S3 method for class 'rctglm'
model.frame(formula, ...)

# S3 method for class 'rctglm'
model.matrix(object, ...)

# S3 method for class 'rctglm'
nobs(object, ...)

# S3 method for class 'rctglm'
residuals(object, ...)

# S3 method for class 'rctglm'
logLik(object, ...)

# S3 method for class 'rctglm'
deviance(object, ...)

# S3 method for class 'rctglm'
AIC(object, ..., k = 2)

# S3 method for class 'rctglm'
terms(x, ...)

# S3 method for class 'rctglm'
weights(object, ...)

# S3 method for class 'rctglm'
vcov(object, type = c("coefficients", "estimand"), ...)

# S3 method for class 'rctglm'
confint(object, parm, level = 0.95, ...)

# S3 method for class 'rctglm'
update(object, formula., ..., evaluate = TRUE)

Arguments

object: an object of class rctglm
...: additional arguments passed to methods
x: an object of class rctglm
digits: a numeric with the number of digits to display when printing
level: a numeric confidence level in $(0, 1)$, used for the Wald interval on the estimand in summary, plot, and confint. Defaults to 0.95.
type: for vcov, one of "coefficients" (default) or "estimand". "coefficients" returns vcov(object$glm) (the variance-covariance matrix of the GLM coefficient vector); "estimand" returns the estimated variance of the marginal-effect estimand as a 1x1 matrix.
parm: for confint, either omitted (returns the Wald CI for the estimand) or the character string "coefficients" (delegates to stats::confint on the underlying GLM).
formula.: for update, a formula, updated via stats::update.formula. See stats::update().
evaluate: for update, if TRUE (default) the updated call is evaluated; if FALSE the unevaluated call is returned.
formula: for model.frame, an object of class rctglm.
k: for AIC, the penalty per parameter. See stats::AIC().

Value

estimand/est return a data.frame with columns Estimate and Std. Error with the estimate and standard error of the estimand.

summary returns a summary.rctglm object. plot returns a ggplot object. confint returns a 1x2 matrix (or the GLM confint output). vcov returns a variance-covariance matrix (default) or a 1x1 matrix (for the estimand). update returns a new rctglm object (or an unevaluated call when evaluate = FALSE).

See coef() and predict.glm() for details on what is returned by the corresponding methods.

Details

The generics estimand (and its alias est) extract a one-row data.frame with the estimate and standard error of the estimand.

summary.rctglm returns a summary.rctglm object printed with print.summary.rctglm. The object holds a coefficient-style table for the estimand with columns Estimate, Std. Error, z value, Pr(>|z|), and the lower/upper Wald CI bounds at the requested level. The Wald test uses the null value $r(\psi, \psi)$ (which equals 0 for the ATE and 1 for rate/odds ratios). The underlying GLM summary is stored in the glm_summary element for full access to the GLM diagnostics.

plot.rctglm returns a ggplot2 object with the estimand estimate and its Wald CI drawn as a point and horizontal error bar, plus a dashed vertical reference line at the null value.

Thin wrappers to the corresponding methods on the underlying glm are provided for coef, predict, formula, family, fitted, model.frame, model.matrix, nobs, residuals, logLik, deviance, AIC, terms, weights. vcov.rctglm returns the GLM coefficient variance-covariance matrix by default and the estimand variance when type = "estimand". confint.rctglm returns the Wald CI for the estimand by default (or delegates to confint(object$glm) when parm = "coefficients"). update.rctglm re-runs rctglm() with an updated call.

Examples

# Generate some data to showcase example
n <- 100
exposure_prob <- .5
dat_binom <- glm_data(
  Y ~ 1+1.5*X1+2*A,
  X1 = rnorm(n),
  A = rbinom(n, 1, exposure_prob),
  family = binomial()
)

# Fit the model
ate <- rctglm(formula = Y ~ .,
              exposure_indicator = A,
              exposure_prob = exposure_prob,
              data = dat_binom,
              family = binomial,
              estimand_fun = "ate")
#> 
#> ── Symbolic differentiation of estimand function ──
#> 
#> ℹ Symbolically deriving partial derivative of the function 'psi1 - psi0' with respect to 'psi0' as: '-1'.
#> • Alternatively, specify the derivative through the argument
#> `estimand_fun_deriv0`
#> ℹ Symbolically deriving partial derivative of the function 'psi1 - psi0' with respect to 'psi1' as: '1'.
#> • Alternatively, specify the derivative through the argument
#> `estimand_fun_deriv1`

print(ate)
#> 
#> Object of class rctglm 
#> 
#> Call:  rctglm(formula = Y ~ ., exposure_indicator = A, exposure_prob = exposure_prob, 
#>     data = dat_binom, family = binomial, estimand_fun = "ate")
#> 
#> Counterfactual control mean (psi_0=E[Y|X, A=0]) estimate: 0.569
#> Counterfactual active mean (psi_1=E[Y|X, A=1]) estimate: 0.815
#> Estimand function r: psi1 - psi0
#> Estimand (r(psi_1, psi_0)) estimate (SE): 0.246 (0.07336)
summary(ate)
#> 
#> Summary of rctglm fit
#> 
#> Call:  rctglm(formula = Y ~ ., exposure_indicator = A, exposure_prob = exposure_prob, 
#>     data = dat_binom, family = binomial, estimand_fun = "ate")
#> 
#> Counterfactual control mean (psi_0=E[Y|X, A=0]): 0.569
#> Counterfactual active  mean (psi_1=E[Y|X, A=1]): 0.815
#> 
#> Estimand function r: psi1 - psi0
#> Null value (r(psi, psi)): 0
#> 
#>          Estimate Std. Error z value  Pr(>|z|)  2.5 % 97.5 %
#> Estimand  0.24602    0.07336   3.354 0.0007973 0.1022   0.39
#> 
#> (Wald-type 95% CI; standard error computed from the full-model influence-function variance.)
#> Access the underlying GLM summary via `<summary_object>$glm_summary`.
estimand(ate)
#>    Estimate Std. Error
#> 1 0.2460163 0.07335603
est(ate)
#>    Estimate Std. Error
#> 1 0.2460163 0.07335603
coef(ate)
#> (Intercept)          X1           A 
#>    0.460642    1.871062    1.857131 
confint(ate)
#>              2.5 %    97.5 %
#> Estimand 0.1022411 0.3897915
if (FALSE) { # \dontrun{
plot(ate)
} # }

Methods for objects of class `rctglm`

Usage

Arguments

Value

Details

See also

Examples