How to measure effect sizes for rational decision-making

(Philosophy of Science, Englisch)

n studies testing the effectiveness of treatments, the treatments' effect sizes are measured using so-called outcome measures. Such outcome measures provide information for subsequent choices between treatments. However, in the case of binary variables, two classes of outcome measures, absolute and relative ones, differ in how they describe a treatment's effect size. Which outcome measures can then inform a rational choice between treatments? In this talk, I argue that absolute measures are at least as good as, if not better than, relative ones for informing rational decision-making across choice scenarios.

To start, I model two choice scenarios using decision theory, one between a tested treatment and the control group treatment and another between treatments tested in distinct trials. Using these decision models, I analyze the conditions under which absolute or relative measures provide decision-relevant information. As argued by Sprenger & Stegenga (2017), absolute measures but not relative ones always do so for choices between a treatment and a control group treatment. I show that this argument does not hold for choices between treatments tested in distinct trials. Here, we need information about the difference in the probabilities of the outcome of interest given treatments to decide. To analyze when absolute or relative measures provide this information, I distinguish between three epistemic situations, differing in how much we know about the probabilities of the outcome given control group treatments. I argue that absolute measures are equally good or better than relative ones for informing choices across these epistemic situations. Overall, for informing rational decision-making, absolute measures dominate relative ones across choice scenarios.

The dominance of absolute measures challenges the current practice in biomedical research of only using relative measures, while also undermining the claim that only absolute measures are suited to inform choices (Sprenger and Stegenga 2017). Recognizing these aspects advances the debate on reporting outcome measures. I conclude my talk with three principles for reporting outcome measures, to be scrutinized in further work moving beyond the idealized perspective of decision models.

Chair: Maren Bräutigam

Zeit: 16:50-17:20, 07. September 2022 (Mittwoch)

Ort: SR 1.004

Ina Jäntgen

(University of Cambridge, United Kingdom)

© 2022 SOPhiA. Alle Rechte vorbehalten.
Letzte Aktualisierung: 2014-04-01.