Check whether a private release is worth its budget, before spending it

Reports f = d / (epsilon * N), the fraction of each prior's width that survives as noise, and the resulting fold-error. This consumes no privacy budget and reads no data: it depends only on the configuration.

Usage

pmx_preflight(priors, epsilon, n_subjects, covariates = NULL)

Arguments

priors

A pmx_priors() object.

epsilon

The privacy budget under consideration.

n_subjects

Number of subjects in the fit.

covariates

Optional pmx_covariates(), so the reported d matches a fit that also releases covariate summaries.

Value

A pmx_preflight report. The arithmetic behind it is worked through at https://iamstein.github.io/synpmx/articles/privacy-background.html.

Details

The fold-error is exp(f * span), capped at the prior's half-width because clipping prevents a release from landing outside the prior. The uncapped form is accurate for f below roughly 0.25 and increasingly pessimistic above it; see the feasibility article at https://iamstein.github.io/synpmx/articles/feasibility.html.