Occam's razor pares the choice among statistical models

Our plain-English summary of Occam's razor pares the choice among statistical models, by K.J. Rothman (American Journal of Epidemiology, 1978).

The study

• Author: K.J. Rothman
• Journal: American Journal of Epidemiology
• Published: 1978;108(5):347-9
• Cited by: 24 articles, as counted by NIH iCite
• Read it: PubMed · Publisher
• In J-M's library: Johns-Manville's company medical library held a copy (J-M no. 002148).

Summary of the article

This commentary was the author's reply to a preceding paper by Walter and Holford, who had argued against his position that independent effects in epidemiology are additive, meaning that the joint effect of two or more factors equals the sum of their separate effects. The author argued that Walter and Holford confused statistical interaction, which depends on the choice of scale, with biologic interaction. He also argued that models leading to divergent conclusions about the same data could not be equally acceptable when the goal was scientific inference rather than description or prediction. He used the example of smoking and asbestos exposure in causing lung cancer. Whatever scale was used, smoking had a greater effect on lung cancer incidence among asbestos workers than among people not exposed to asbestos. He argued that a worker asking about this should not be told the answer was arbitrary. He stated that Walter and Holford's "one-hit" model corresponded to causation and their "no-hit" multiplicative model to prevention, and noted that he had derived the same multiplicative relationship in a 1976 paper. He said he did not regard his approach as the only appropriate method and pointed to a technique developed by Walker. He also corrected what he described as inaccuracies about the assessment of causality, the meaning of necessary and sufficient causes, and the meaning of Occam's razor.

PubMed has no abstract for this study, and the article doesn't print one. We wrote this summary from the full article with the help of AI. It is not the authors' wording.

In plain English

This 1978 article was a commentary, not a study, and it reported no data. The author replied to a paper by Walter and Holford in the same journal issue. That paper had disputed the author's earlier position that independent effects in epidemiology are additive, meaning the joint effect of two factors equals the sum of their separate effects. The author argued that statistical models leading to opposite conclusions about the same data cannot both be acceptable when the goal is scientific inference. As an example, the author used smoking and asbestos exposure in causing lung cancer. The author stated, without presenting data in this article, that smoking has a greater effect on lung cancer incidence among asbestos workers than among people not exposed to asbestos. The author held that the multiplicative model Walter and Holford preferred describes prevention, not causation. The author acknowledged it could reasonably be argued that defining biologic interaction by its usefulness for public health questions, though pragmatic, lacks a strong theoretical base. The author concluded that, insofar as Occam's razor bears on the dispute, it warns against accepting many unnecessary statistical models.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Commentary (a reply to another paper on statistical methods); no new data
• Who was studied: No study population; the article is a methodological argument
• Exposure: Smoking and asbestos exposure were discussed only as an example of two risk factors for lung cancer; no fiber types, doses or data were given
• Main results:
  • The author argued that statistical models leading to divergent conclusions about the same data cannot be equally acceptable when the goal is scientific inference
  • The author asserted, without presenting data in this article, that smoking has a greater effect on lung cancer incidence among asbestos workers than among those unexposed to asbestos, whatever scale transformation is used
  • The author disagreed with Walter and Holford that the answer to such a worker's question could depend on an arbitrary choice of criterion
  • The author held that the 'one-hit' and 'no-hit' models correspond to causation and prevention, so the choice between them is not arbitrary
  • The author stated that causality for a risk factor can be assessed without evaluating interaction between factors
  • The author said other methods, such as one by Walker that allocates disease burden to single and joint effects, merit consideration
• The authors' conclusion: The author concluded that, insofar as Occam's razor bears on the issues Walter and Holford addressed, it warns against accepting many unnecessary statistical models
• Limitations the authors state:
  • The author acknowledged it could reasonably be argued that defining biologic interaction by its usefulness for public health questions, though pragmatic, lacks a strong theoretical base

Topics

Medical terms in this study, explained in our Lexicon: Epidemiology, Asbestos Exposure, Lung Cancer, Causation.

Related studies

• Cervical endometriosis associated with malignant pleural mesothelioma mimicking cervical cancer--Occam's razor or the "third man" — J.B. Engel, S. Heuer, et al. (2011)
• Occam's Razor versus Hickam's Dictum — T.A. Newman, J.E. Takasugi, et al. (2017)
• Agreement between a brief food frequency questionnaire and diet records using two statistical methods — G.L. Ambrosini, N.H. de Klerk, et al. (2001)
• Ordinal regression models for epidemiologic data — B.G. Armstrong, M. Sloan (1989)
• Mesothelioma mortality in asbestos workers: implications for models of carcinogenesis and risk assessment — J. Peto, H. Seidman, et al. (1982)
• Environmental pathology: new directions and opportunities — B.T. Mossman (1999)

Browse more research in American Journal of Epidemiology, or return to medical research.