LDP Correction Estimators

Recover true statistics from LDP-perturbed data

How This Demo Works

1. Original data is fetched from the database. 2. LDP perturbation is applied with your selected ε=2. 3. Correction estimators recover the true statistics using the same ε.

Important: The correction formula must use the same epsilon that was used for perturbation. Changing ε re-processes the data with that privacy level, then corrects it accordingly.

Frequency Correction

F̂ = (observed - n×p) / (q-p)

Truth Probability (q)

q = e^ε / (e^ε + d - 1)

Lie Probability (p)

p = 1 / (e^ε + d - 1)

Scenario

Field

Mechanism

Epsilon (ε): 2.0

More PrivacyMore Accuracy

Truth probability: 64.9%

Confidence Level

No data available. Please check your database connection.

SQL Examples (GRRM)

-- Frequency correction

SELECT anon.ldp_frequency_estimate(observed_count, total_n, 2, 5);

-- Variance estimation

SELECT SQRT(anon.ldp_frequency_variance(observed_count, total_n, 2, 5)) as std_error;

-- Mean correction

SELECT anon.ldp_mean_from_frequencies(ARRAY[c1, c2, ...], 2, 5);

-- Confidence interval

SELECT anon.ldp_ci_lower(count, n, 2, 5, 0.95),
anon.ldp_ci_upper(count, n, 2, 5, 0.95);