The Evolution and Interpretation of "Statistical Purposes"
Source: arXiv:2607.11778 · Published 2026-07-13 · By Michael B. Hawes, John L. Eltinge, Paul S. Marck, Danielle C. Neiman, Sallie Ann Keller
TL;DR
This paper investigates the historical evolution, legal definitions, and broader ethical context of the phrase "for statistical purposes only" as used by National Statistical Organizations (NSOs), particularly in the United States. While ubiquitous in data collection communications and policy, the term lacks a clear and consistent operational definition. The authors trace its origin from early 20th century census practices through successive U.S. laws, culminating in the 2002 CIPSEA statute and the 2019 Evidence Act, which formalize statistical purposes as producing aggregate-level data intended for public benefit while protecting respondent confidentiality and prohibiting identifiable data use for enforcement or regulatory actions. The study reveals significant ambiguities and gaps in common interpretations, especially around trust, objectivity, and broader public good considerations. It compares U.S. usage to international NSO definitions, showing convergence but also differences in emphasis on public benefit. The authors argue for a broader, ethically grounded definition that explicitly incorporates scientific integrity, respondent trust, representativeness, and public benefit to provide clearer guardrails for NSOs' operational and ethical decisions. The paper closes by outlining areas—like trust rebuilding and transparency—that merit further research given the declining public trust in government statistics and growing privacy concerns.
Key findings
- Two predominant statutory criteria define 'statistical purposes': producing statistical info about large population aggregates for public benefit, and protecting confidentiality to prevent identifiable data use for legal or regulatory harm (CIPSEA §502(5),(7)).
- Historical usage has ambiguities: early 20th century census promised confidentiality but census data were used in draft evasion and immigration enforcement cases.
- World War II Second War Powers Act allowed individual census info disclosure to law enforcement, breaking prior confidentiality promises.
- Declining public trust in federal government (from 77% in 1964 to 23% in 2024) coincides with rising survey nonresponse, raising challenges for NSOs relying on 'statistical purposes' for trust.
- Public cognitive testing shows respondents often misunderstand 'statistical purposes,' not recognizing confidentiality protections and fearing data sharing with other agencies.
- 2019 Evidence Act codifies four federal statistical agency responsibilities including objectivity, relevance, confidentiality, and trust protection, extending the traditional definition of statistical purposes.
- International NSOs (EUROSTAT, Statistics Canada, UK ONS) employ similar but varied and sometimes more inclusive definitions emphasizing public good and research use.
- There is a gap between concise statutory definitions and the broader ethical, professional, and scientific frameworks NSOs must navigate to ethically justify data uses.
Threat model
The adversary is anyone who might misuse identifiable statistical data about individuals or organizations to cause harm, including legal, administrative, regulatory, or enforcement actions against data subjects. Statistical agencies must assume potential insider or external threats capable of unauthorized disclosure or re-identification. The agencies cannot allow the use of data collected under the premise of "statistical purposes" to be employed for nonstatistical, potentially harmful purposes or allow access by unauthorized parties. Assumes the adversary cannot circumvent well-established functional and legal separations protecting identifiable data.
Methodology — deep read
The paper employs a comprehensive historical-legal-policy review supplemented by literature on cognitive testing and trust research to analyze how the term 'statistical purposes' has been used and interpreted.
Threat model & assumptions: The adversary concept is implicit—statistical agencies must protect data subjects from misuse of identifiable respondent data that could cause legal, regulatory, or personal harm. The authors explore how laws and policies limit data use and access to prevent such harms.
Data: This is conceptual analysis rather than empirical experimental data. The authors draw on archival and public legal documents (e.g., U.S. Census laws 1929, 1954 Title 13, CIPSEA 2002, Evidence Act 2019), government policy documents, prior surveys on public trust and understanding, and international NSO definitions.
Architecture/algorithm: Not applicable as this is legal/ethical research.
Training regime: Not applicable.
Evaluation protocol: The authors systematically trace statutory language evolution and cross-reference with ethical principles (Belmont Report, FIPPs, ASA ethical guidelines) and professional standards for statistical practice. They incorporate results from cognitive studies (e.g., Landreth et al. 2008) demonstrating respondent confusion. They compare U.S. definitions with international counterparts, highlighting convergence and differences.
Reproducibility: As a legal and historical analysis, no code or datasets are provided, but all cited laws and references are publicly accessible. Empirical survey references are from prior independent studies.
Concrete example: The 1910 Census, publicly proclaimed to be solely for statistical purposes with confidentiality promises, was nonetheless used for legal enforcement activities (draft evasion, deportation). This illustrates tension between stated purpose and actual use, motivating statutory clarifications starting in 1929, with cyclical erosion in WWII, before stronger statutory protections in later decades.
Technical innovations
- First comprehensive historical-legal synthesis of the ambiguities and evolution of 'statistical purposes' in U.S. federal statistics over 100+ years.
- Integration of legal statutes with scientific integrity and professional ethics frameworks to propose a broader, multi-dimensional definition.
- Comparison of U.S. definitions with international NSO approaches to highlight commonalities and potential improvements.
- Identification of the conceptual gap between statutory language and public misunderstanding/trust challenges informed by cognitive testing.
Datasets
- Public understanding and trust surveys referenced (Pew Research Center 1964-2024; Landreth et al. 2008 cognitive testing)
- Archival legal documents: U.S. Census Acts (1929, 1954), Privacy Act (1974), CIPSEA (2002), Evidence Act (2019), War Powers Act (1942)
- International NSO definitions: EUROSTAT (2023), Statistics Canada (2017), UK ONS (2023), UN Statistical Division (2019)
Limitations
- No original empirical survey data or experimental validation provided—relies on prior secondary sources for public understanding and trust data.
- Focus is primarily on U.S. federal statistical agencies with some international comparison—may not capture all global NSO nuances.
- Does not directly address technical methods for disclosure avoidance or privacy-preserving data releases, only legal and ethical frameworks.
- The broader proposed definition is conceptual without operational metrics or concrete procedures for implementation.
- Limited discussion on how advancing technologies (e.g., big data, AI) challenge traditional definitions of statistical purposes.
- Ambiguities remain in balancing competing priorities like confidentiality, public benefit, relevance, timeliness, and cost.
Open questions / follow-ons
- How can NSOs systematically integrate notions of public trust and scientific integrity into operational definitions and workflows related to statistical purposes?
- What are effective strategies to improve public understanding of 'statistical purposes' and build trust in official statistics amid growing privacy concerns?
- How should emerging data collection modalities and administrative data linkages be governed under evolving conceptions of statistical purposes?
- What metrics or formal quality standards can codify the broader ethical and scientific elements proposed to augment statutory definitions?
Why it matters for bot defense
For bot-defense and CAPTCHA practitioners, this paper highlights the critical role of clear, ethically grounded definitions and communications about data use to maintain public trust. Although focused on large-scale official statistics, the challenges around user understanding of data collection purposes and confidentiality protections are broadly transferable to online data collection and bot mitigation contexts. Ambiguities in terms like 'statistical purposes' can undermine trust and cooperation, just as unclear data use policies can degrade user compliance with CAPTCHA or anti-bot measures. Furthermore, the emphasis on functional separation and limiting data use strictly for well-defined, beneficial purposes is a principle that can inform ethical design of user verification systems that collect sensitive behavioral or biometric data. This work encourages practitioners to consider deeper transparency, clear purpose specification, and aligning data use policies with broader public benefit and integrity principles—beyond legal compliance—to build more trustworthy and effective defenses against automated abuse.
Cite
@article{arxiv2607_11778,
title={ The Evolution and Interpretation of "Statistical Purposes" },
author={ Michael B. Hawes and John L. Eltinge and Paul S. Marck and Danielle C. Neiman and Sallie Ann Keller },
journal={arXiv preprint arXiv:2607.11778},
year={ 2026 },
url={https://arxiv.org/abs/2607.11778}
}