R/use_weights.R
summarize.adjustr_weighted.Rd
Uses weights computed in adjust_weights
to compute posterior
summary statistics. These statistics can be compared against their reference
values to quantify the sensitivity of the model to aspects of its
specification.
An adjustr_weighted
object.
Name-value pairs of expressions. The name of each argument will be
the name of a new variable, and the value will be computed for the
posterior distribution of eight alternative specification. For example,
a value of mean(theta)
will compute the posterior mean of
theta
for each alternative specification.
Also supported is the custom function wasserstein
, which computes
the Wasserstein-p distance between the posterior distribution of the
provided expression under the new model and under the original model, with
p=1
the default. Lower the spacing
parameter from the
default of 0.005 to compute a finer (but slower) approximation.
The arguments in ...
are automatically quoted and evaluated in the
context of .data
. They support unquoting and splicing.
Whether to compute summary statistics by first resampling
the data according to the weights. Defaults to FALSE
, but will be
used for any summary statistic that is not mean
, var
or
sd
.
Stan model data, if not provided in the earlier call to
adjust_weights
.
An adjustr_weighted
object, with the new columns specified in
...
added.
if (FALSE) {
model_data = list(
J = 8,
y = c(28, 8, -3, 7, -1, 1, 18, 12),
sigma = c(15, 10, 16, 11, 9, 11, 10, 18)
)
spec = make_spec(eta ~ student_t(df, 0, 1), df=1:10)
adjusted = adjust_weights(spec, eightschools_m)
summarize(adjusted, mean(mu), var(mu))
summarize(adjusted, wasserstein(mu, p=2))
summarize(adjusted, diff_1 = mean(y[1] - theta[1]), .model_data=model_data)
summarize(adjusted, quantile(tau, probs=c(0.05, 0.5, 0.95)))
}