What does this research mean for the field?

A novel two-step method combining clustering with robust outlier detection using the median and median absolute deviation matches or outperforms traditional one-step methods for identifying reporting irregularities in administrative data. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To develop and evaluate a robust two-step outlier detection framework that combines clustering with median absolute deviation to identify reporting irregularities in high-volume administrative data.

September 4, 2026Statistical Journal of the IAOS

Assessing administrative data quality: Detecting outliers in monthly reporting through historical analysis

View Full Paper

Ask AI

Bookmark

Authors

JBJD BunkerRTI International DLDan LiaoRTI International MBMarcus BerzofskyRTI International

Discussion

Member takes

Overview

Simulation study reveals a two-step clustering and robust deviation method accurately detects irregular administrative reporting, indicating improved data quality for downstream estimation.

Key Points

To develop and evaluate a robust two-step outlier detection framework that combines clustering with median absolute deviation to identify reporting irregularities in high-volume administrative data.
Designed a two-step detection framework pairing cluster analysis with robust outlier metrics (median and median absolute deviation).
Evaluated performance against mean ± SD, boxplot, and ratio-to-median methods across 10,000 simulations spanning 10 distinct data scenarios.
Applied the framework to empirical monthly counts from 136 National Incident-Based Reporting System (NIBRS) agencies and conducted sensitivity analyses on the primary tuning parameter k.
The proposed two-step framework matched or outperformed traditional one-step outlier detection approaches in 8 out of 10 simulation scenarios, ranking second best in the remaining 2.
Empirical evaluation of 136 NIBRS agencies demonstrated that the method effectively isolated irregular reporting frequencies and extreme counts relative to standard ratio-to-median benchmarks.

Cite This Study

Bunker et al. (2026) studied this question.

synapsesocial.com/papers/6a9a85e75d9e33f25c6314ac https://doi.org/10.1177/18747655261477979

View Full Paper

Ask AI

Bookmark

September 4, 2026Statistical Journal of the IAOS

Assessing administrative data quality: Detecting outliers in monthly reporting through historical analysis

View Full Paper

Ask AI

Bookmark

Authors

JBJD BunkerRTI International DLDan LiaoRTI International MBMarcus BerzofskyRTI International

Discussion

Member takes

Overview

Simulation study reveals a two-step clustering and robust deviation method accurately detects irregular administrative reporting, indicating improved data quality for downstream estimation.

Key Points

To develop and evaluate a robust two-step outlier detection framework that combines clustering with median absolute deviation to identify reporting irregularities in high-volume administrative data.
Designed a two-step detection framework pairing cluster analysis with robust outlier metrics (median and median absolute deviation).
Evaluated performance against mean ± SD, boxplot, and ratio-to-median methods across 10,000 simulations spanning 10 distinct data scenarios.
Applied the framework to empirical monthly counts from 136 National Incident-Based Reporting System (NIBRS) agencies and conducted sensitivity analyses on the primary tuning parameter k.
The proposed two-step framework matched or outperformed traditional one-step outlier detection approaches in 8 out of 10 simulation scenarios, ranking second best in the remaining 2.
Empirical evaluation of 136 NIBRS agencies demonstrated that the method effectively isolated irregular reporting frequencies and extreme counts relative to standard ratio-to-median benchmarks.

Cite This Study

Bunker et al. (2026) studied this question.

synapsesocial.com/papers/6a9a85e75d9e33f25c6314ac https://doi.org/10.1177/18747655261477979

View Full Paper

Ask AI

Bookmark