New Exact MLE Formula Proposed for Randomized Response Data

Global: Estimating the True Distribution of Data Collected with Randomized Response

Researchers from multiple institutions have introduced a simple closed‑form solution for the maximum‑likelihood estimate (MLE) of data gathered through randomized response, a protocol that provides local differential privacy, in a study submitted to arXiv on January 13, 2026. The work aims to improve the accuracy of histogram reconstruction while avoiding the negative‑value issue that can arise with traditional debiasing methods.

Background of Randomized Response

Randomized response (RR) enables the collection of categorical information while preserving individual privacy by having each participant report a randomly altered version of their true value. Major technology firms have incorporated RR into analytics pipelines to obtain aggregate insights without exposing raw user data.

Challenges with Standard Debiasing

The conventional debiasing rule applied to RR data can produce estimated histograms that contain negative entries, which are not interpretable as probabilities. Because there is no widely accepted remedy, practitioners face uncertainty when selecting an appropriate estimation technique.

Iterative Bayesian Update and Its Limitations

The Iterative Bayesian Update (IBU) algorithm offers an elegant approach by iteratively refining estimates until they converge to the MLE. However, the iterative nature of IBU can be computationally intensive, especially for large‑scale datasets, making it less practical for real‑time applications.

Exact MLE Formula Introduced

The authors present a direct formula that yields the exact MLE for RR‑collected data without requiring iterative computation. By deriving the solution analytically, the method eliminates the convergence overhead associated with IBU while guaranteeing the same statistical optimality.

Experimental Comparison

Empirical tests on synthetic and real‑world datasets compare the new formula against IBU, standard debiasing, and other recent estimators. Results show that the exact MLE matches IBU’s accuracy but executes orders of magnitude faster, and it consistently avoids negative histogram entries.

Implications for Practitioners

According to the study, developers of privacy‑preserving analytics tools can adopt the closed‑form MLE to achieve reliable histogram estimates with reduced computational cost. The authors suggest that future research may explore extensions to multi‑dimensional data and integration with existing privacy frameworks.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.