Commentary: Statistical significance vs. Clinical Importance: What is in a term?

- Prof. Jochanan Benbassat, MD

Unit of Sociology of Health, The Faculty of Health Sciences,
Ben Gurion University of the Negev, P.O.B. 653, 84105 Beer-Sheva, ISRAEL

Eubios Journal of Asian and International Bioethics 5 (1995), 94-5.
Ethical problems may arise not only from philosophic disagreement, but also from lack of data or misinterpretation of available information. In this issue Dr Goldsmith draws attention to two causes of such a misinterpretation: deliberate distortion of data and ambiguous terminology. The author posits that an ambiguous terminology may lead to conflicting interpretations of the same evidence over causal relationships (1).

Causal relationships, i.e., does cause "A" lead to consequence "B", are central in science, law and decision making in general. This is consistent with human tendency to organize events by schemes of cause-effect relations. Causal links dominate our thinking, and there is a general tendency to view causes as leading inevitably, rather than probabilistically, to their consequences (2).

More often than not however, causes lead probabilistically, rather than inevitably, to their consequences. Cause-effect relations are seldom "either or"; the relationship between causes and their consequences may range over the entire spectrum of certainty. The degree of confidence in the causal relationship that justifies a decision varies according to discipline and circumstances. What may be an adequate proof for causality in a liability law suit, may not suffice for a criminal conviction. What may be an ample reason for action in order to prevent a disaster, may not fulfil the scientific prerequisites for ruling out the null hypothesis. The same evidence that in a given context should be considered as sufficient for a decision, may still be viewed by scientists as statistically "non-significant", even though both judgements are correct. Only too often, a "non- significant" association between two variables, is erroneously perceived as a proof that they are not related. The consequences are frequent misunderstandings between investigators and decision makers, and even exasperation with conflicting conclusions of two experts from the same data.

To ensure an appropriate communication between investigators and decision makers we need a new vocabulary and a clear demarcation of their responsibility. While judgements are made in a deterministic "either yes or no" manner, inferences from available evidence are probabilistic, and subject to varying degrees of uncertainty. It is the task of the investigator to interpret evidence for causality in terms of probabilities. It is the duty of decision makers to rule whether these probability estimates justify a practical conclusion in the given context.

The need for such a redefinition has been alluded to in the past by authors who drew attention to the discrepancy between the statistically significant but clinically unimportant, and vice versa (3). Dr Goldsmith's main contribution is his plea for a more lucid presentation of the degree of certainty of inferred causality. Hopefully, this will reduce the bewilderment of decision makers with apparently conflicting expert conclusions, and will also prevent the deliberate misrepresentation and misapplication of scientific evidence.


1. Goldsmith JR., "Where the trail leads...Ethical problems arising when the trail of professional work lead to evidence of cover-up of serious risk and misrepresentation of scientific judgement concerning human exposures to radar", EJAIB 5 (1995), 87-91.
2. Tversky A, Kahneman D. "Causal schemes in judgement under uncertainty", pp. 117-28 in Kahneman D, Slovic P, Tversky A (eds): Judgement under uncertainty: Heuristics and biases. Cambridge University Press, London, 117-28, 1982.
3. DL Sackett, RB Haynes, P Tugwell eds., Clinical Epidemiology. A basic science for clinical medicine. Little, Brown Co, Boston 1985.

Go to Commentary - Frank Leavitt
Go back to EJAIB July 1995
Go back to EJAIB