Official Statistics Workflow with DPrivStats

The hardware and bandwidth for this mirror is donated by dogado GmbH, the Webhosting and Full Service-Cloud Provider. Check out our Wordpress Tutorial.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]dogado.de.

library(DPrivStats)
set.seed(11)

This vignette sketches an official-statistics workflow: combining multiple DP tabulations under a single budget, applying post-processing constraints, and comparing composition rules.

Combining releases under one budget

budget <- new_privacy_budget(epsilon = 4.0, delta = 1e-6, composition = "rdp")

if (can_spend(budget, 1.5)) {
  budget <- spend(budget, 1.5, "mean income by region table")
}

if (can_spend(budget, 1.5)) {
  budget <- spend(budget, 1.5, "education histogram")
}
budget
#> 
#> Privacy Budget (composition: rdp )
#>   total epsilon:  4.000
#>   spent epsilon:  0.000
#>   accumulated rho: 0.0000
#>   remaining:      4.000

Post-processing constraints

DP histograms can contain negative noisy counts; truncating at zero and normalizing are pure post-processing steps that preserve DP:

data(example_microdata)
h <- dp_histogram(example_microdata$age, epsilon = 1.0,
                  breaks = seq(10, 90, by = 10), normalize = TRUE)
h$estimate # already non-negative by construction
#>    [10,20]    (20,30]    (30,40]    (40,50]    (50,60]    (60,70]    (70,80] 
#> 0.03190307 0.12534167 0.28326799 0.33126897 0.16600733 0.05101880 0.01119218 
#>    (80,90] 
#> 0.00000000

Composition comparison

For a fixed workflow of small releases, RDP is typically much tighter than basic composition:

eps_seq <- c(1.5, 1.5, 1.0)
compare_composition(eps_seq, delta = 1e-6)
#>   composition  epsilon
#> 1       basic  4.00000
#> 2    advanced 29.32444
#> 3         rdp 11.46696

Disclosure risk intuition

The privacy loss random variable of a Laplace release is exponential; its tail probabilities quantify the chance of large losses:

laplace_plr_tail(c(0, 3, 6), epsilon = 1.0)
#> [1] 1.000000000 0.049787068 0.002478752

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.
Health stats visible at Monitor.