Talk:Tweedie distribution

	This article is rated B-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects:
Statistics Mid‑importance This article is within the scope of WikiProject Statistics, a collaborative effort to improve the coverage of statistics on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks. Mid This article has been rated as Mid-importance on the importance scale.

On 31 July 2025, it was proposed that this article be moved from Tweedie distribution to Tweedie-Bar-Lev-Enis distribution. The result of the discussion was not moved.

Power law?

What is the independent variable in the variance-to-mean power law? Without knowing what is being varied, it's hard to make sense of it. Is it a mean parameter of the distribution? Is there always just one obvious parameter to vary? Dicklyon (talk) 17:09, 3 May 2013 (UTC)[reply]

Requested move 31 July 2025

The following is a closed discussion of a requested move. Please do not modify it. Subsequent comments should be made in a new section on the talk page. Editors desiring to contest the closing decision should consider a move review after discussing it on the closer's talk page. No further edits should be made to this discussion.

The result of the move request was: not moved. (non-admin closure) Tenshi! (Talk page) 00:46, 15 August 2025 (UTC)[reply]

---

Tweedie distribution → Tweedie–Bar-Lev–Enis distribution – Several reliable sources describe this family as independently and rigorously characterized by M. C. K. Tweedie (1984) and by Shaul K. Bar-Lev & Peter Enis (1986), and recent literature adopts the triple eponym. The current single-eponym title reflects later secondary usage and obscures the dual/independent characterization. Per WP:COMMONNAME and WP:CONSISTENCY (distribution pages use the singular), the title should be Tweedie–Bar-Lev–Enis distribution. Sources: Tweedie 1984; Bar-Lev & Enis 1983, 1986; Brown 1986 (cites the 1983 TR); Jørgensen 1987; Bar-Lev 2019; Cohen & Huillet 2022; Kokonendji et al. 2020; Truquet, Cohen & Doukhan 2024. Stochastics101 (talk) 21:18, 31 July 2025 (UTC) ~~~~ — Relisting. Ivey (talk - contribs) 23:31, 7 August 2025 (UTC)[reply]

— Stochastics101 (talk • contribs) has made few or no other edits outside this topic. —⁠ ⁠BarrelProof (talk) 23:38, 31 July 2025 (UTC)[reply]

• Leaning opposed: The nominator also made a closely related major edit of the article a few days ago to. That and this renaming proposal are the only edits ever made by the nominating account. I strongly suspect some agenda is being pushed here. The "misnamed" remark inserted into the article seems particularly heavy-handed. It is not "misnaming" to refer to something using the name of the first person who published a substantial discussion of the subject, and Tweedie appears to have been first, if I understand correctly. Even using a name with a murkier justification than that is not necessarily "misnaming". It seems obvious that most of the sources being cited by the nominator do not refer to the "Tweedie–Bar-Lev–Enis distribution", and I suspect some cherry-picking as well. It seems clear that none of the pre-2019 sources described by the nominator use the term "Tweedie–Bar-Lev–Enis distribution". The 2019 source appears to be primarily an advocacy article complaining about the naming that was written by one of the people proposed to be added to the list of credits, and thus it is not an independent reliable source. The Kokonendji 2020 source is paywalled, so I don't know what it says, and it has only two independent citations in Google Scholar, so it does not appear especially influential (although low citations might be expected for such a recent article). One of those two citations is from one of the people proposed to be added to the list of credits in the name. No clear citation was provided by the nominator for "Truquet, Cohen & Doukhan 2024", and that one is not cited in the Wikipedia article. I found an ArXiv paper that fits that description. It does not contain the term "Tweedie–Bar-Lev–Enis distribution". Not that this is necessarily conclusive on the subject, but the Tweedie 1984 paper is cited about 7× as much in Google Scholar as the most-cited of the Bar-Lev–Enis articles. The Jørgensen 1987 article and the Tweedie 1984 article are relatively highly cited, with about 1,000 citations each in Google Scholar. As stated in the Wikipedia article, it appears that the Jørgensen 1987 article calls this the "Tweedie distribution". I do not find a record of a "Bar-Lev & Enis 1983" publication. It seems unlikely that the 1985 Bar-Lev & Enis Metrika paper contains much that is especially influential, as that paper has only one citation in Google Scholar that wasn't written by Bar-Lev (and that referencing publication is incompletely described). —⁠ ⁠BarrelProof (talk) 23:38, 31 July 2025 (UTC)[reply]

  Key Facts:

  1. Bar-Lev and Enis (1986, Annals of Statistics) investigated the reproducibility property of natural exponential families (NEFs) on the real line. Originally written as a 1983 SUNY Buffalo technical report, it was submitted to Annals of Statistics in November 1983 and published in 1986 (submission and acceptance dates are indicated on the published paper's front page). They proved that if a steep NEF is reproducible, it must have a power variance function of the form V(m)=Am^{p}, A>0, p∈R, m∈M (the mean parameter space). They identified all possible values of p for which the NEF is genuine and derived various statistical and probabilistic properties of such NEFs for each permissible value of p. The technical report had previously been distributed to multiple researchers, including Lawrence D. Brown, who cited it in his 1986 monograph "Fundamentals of Statistical Exponential Families" (Institute of Mathematical Statistics Lecture Notes-Monograph Series, Vol. 9). Brown did not cite Tweedie's paper. The technical report was also sent to Bent Jørgensen, who cited the published Bar-Lev and Enis paper as appearing in 1987.
  2. Tweedie (1984): Tweedie's 1984 paper entitled "An Index Which Distinguishes between Some Important Exponential Families" was published in the proceedings of the Indian Statistical Institute Golden Jubilee International Conference, edited by J.K. Ghosh & J. Roy in Calcutta (pp. 579–604). The first citation of Tweedie's proceedings paper appeared in Bent Jørgensen's fundamental 1987 paper "Exponential Dispersion Models" (with discussion) in the Journal of the Royal Statistical Society B, where both Tweedie and Bar-Lev served as discussants. Two main points should be emphasized:

  • Tweedie's analysis contained a significant error: he incorrectly claimed that NEFs with power variance functions do not exist for p<0. This claim was mathematically incorrect, as NEFs with power variance functions do exist for p<0. These NEFs are generated by stable distributions with stable index α∈(1,2), and the corresponding NEFs are supported on R (the whole real line). At the same time, their mean parameter space is M=R⁺ (positive real numbers).

  • Availability issue: To this date, as far as I could search, a full text or PDF of Tweedie's 1984 proceedings article does not appear to be available online.

  3. The term "TWEEDIE DISTRIBUTION" or "TWEEDIE SCALE" was first coined in Bent Jørgensen's monograph (1997), "The Theory of Dispersion Models" (Chapman and Hall), without acknowledging the significant contribution of Bar-Lev and Enis. Since then, Tweedie's paper has been cited hundreds of times.

  4. The term "Tweedie-Bar-Lev-Enis" has been used since Bar-Lev (2019) and appears 12 times in the literature, including in Bar-Lev papers and by other authors such as Cohen and Huillet (2022) in Journal of Statistical Physics, Springer.

  5. Jørgensen's influential designation: Jørgensen's 1987 paper refers to Tweedie (1984) as having provided the "most complete study" of NEF–PVFs, which subsequently led to naming this class after Tweedie—despite the fact that Bar-Lev & Enis submitted their work earlier.

  6. The Tweedie distribution in Wikipedia: It should be noted that all GLM procedures, whether additive or multiplicative, appearing under this entry belong to Jørgensen's findings as presented in his 1997 monograph. Neither Tweedie nor Bar-Lev contributed to these findings. Accordingly, Jørgensen's contributions should be acknowledged.

  7. Historical precedent - The Rao-Blackwell example: Bar-Lev (2019) argues that proper attribution should follow the precedent established with the Rao-Blackwell theorem: "Both Bar-Lev and Enis should have received appropriate credit by renaming the class of NEF-PVFs to include the names of Tweedie, Bar-Lev, and Enis. This would resemble the dignified and elegant manner in which Lehmann and Scheffé acted regarding the Rao-Blackwell Theorem." The parallel is compelling: the Rao-Blackwell theorem provides an established precedent for how the mathematical community should handle independent discoveries. Rao (1945) published his result in the Bulletin of the Calcutta Mathematical Society, while Blackwell (1947) independently published the same result two years later in The Annals of Mathematical Statistics. These were not joint publications—each mathematician discovered the result independently. When Lehmann and Scheffé (1950) later referenced this work, they recognized both contributors by naming it the "Rao-Blackwell theorem," despite the two-year gap between publications. As Dr. Calyampudi Radhakrishna Rao notes in Wikipedia: "The result on the one parameter case was published by Rao (1945) in the Bulletin of the Calcutta Mathematical Society and by Blackwell (1947) in The Annals of Mathematical Statistics. Lehmann and Scheffé (1950) called the result the Rao-Blackwell theorem. In computational terminology, it is called the Rao-Blackwellized Filter." This precedent demonstrates that Lehmann and Scheffé acted in a dignified and logical manner by acknowledging both independent discoverers. Similarly, the mathematical community should recognize all hree contributors—Tweedie, Bar-Lev, and Enis—who independently provided the rigorous mathematical foundation for this important family of distributions, rather than attributing the work to Tweedie alone.

  8. Citation patterns: The higher number of citations for Tweedie compared to Bar-Lev and Enis is a natural consequence of Jørgensen coining the name "Tweedie distribution." Once this terminology was established, subsequent citations naturally followed this naming convention.

  9. Call for justice: To achieve historical justice and prevent academic injustice, the Tweedie distribution should be called the Tweedie-Bar-Lev-Enis distribution.

  10. Final note: I trust that historical justice will be served. Stochastics101 (talk) 23:36, 6 August 2025 (UTC)[reply]

• Oppose for now – Let's discuss what sources say and converge on what the article should say about the history of this class of distributions before doing a title change. I have no objection to historical justice, but it has be supported by sources without WP:OR or much WP:SYNTH. Dicklyon (talk) 02:50, 7 August 2025 (UTC)[reply]

Relisting comment: Stimulating some additional participation just in case Ivey (talk - contribs) 23:31, 7 August 2025 (UTC)[reply]

Note: WikiProject Statistics has been notified of this discussion. Ivey (talk - contribs) 23:31, 7 August 2025 (UTC)[reply]

• Looking in books, I don't find this name for the distribution. And books that mention Bar-Lev and Enis (1986) mostly also mention Hougaard (1986) in parallel. Including all these in the name of the distribution would be unwieldy, so I don't expect anyone much does it, and neither should we. In scholarly papers, the version that omits Hougaard was introduced by Bar-Lev as you note above, which doesn't sound likely to be seen as a fair alternative. Let's wait, and if it's widely adopted of course we can use it. Dicklyon (talk) 02:12, 9 August 2025 (UTC)[reply]

The discussion above is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

Request to revisit move: Tweedie → Tweedie–Bar-Lev–Enis distribution

The earlier move request (closed on 15 August 2025) was decided as “not moved.” Unfortunately, I was unable to respond further at the time due to health issues. Now that I am able to contribute again, I would like to reopen the discussion because there is new evidence and additional sources that were not fully considered in the original request.

Recent scholarship shows increasing use of the term Tweedie–Bar-Lev–Enis. Examples include Cohen & Huillet (2022) in Journal of Statistical Physics, Kokonendji et al. (2020) in Sankhya A, and Truquet, Cohen & Doukhan (2024). Touré’s 2021 PhD thesis also adopts the terminology, and Mildenhall (2025) has published a detailed essay specifically entitled The Class of Bar-Lev–Enis–Tweedie Distributions. This indicates that the triple eponym is not confined to Bar-Lev’s own writing but is now appearing across independent venues.

There is also clearer evidence that the exclusive “Tweedie” naming originated from misattribution in Jørgensen (1987). His paper not only described Tweedie’s study as “the most complete” but also misdated Bar-Lev & Enis’s earlier work and treated narrower contributions (Morris, Hougaard) as if they were parallel. This error had the practical effect of erasing Bar-Lev & Enis from subsequent naming practice, despite their independent and rigorous characterization.

Finally, there is a strong precedent in mathematical nomenclature for recognizing multiple independent discoverers, most notably the Rao–Blackwell theorem. The Tweedie–Bar-Lev–Enis case is directly analogous, with contemporaneous and independent results that shaped the same family of distributions. Given these developments, I believe it is appropriate to revisit the move and adopt the more accurate and historically just name Tweedie–Bar-Lev–Enis distribution. Stochastics101 (talk) 22:34, 9 September 2025 (UTC)[reply]