# Should one derive risk difference from the odds ratio?

**URL:** <https://discourse.datamethods.org/t/should-one-derive-risk-difference-from-the-odds-ratio/4403>\
**Category:** bayes\
**Tags:** absolute-risk\
**Created:** [May 13, 2021, 11:49pm UTC](https://discourse.datamethods.org/t/should-one-derive-risk-difference-from-the-odds-ratio/4403 "2021-05-13T23:49:41Z")\
**Posts on this page:** 3\
**Page:** 32

<div class="post-metadata">

**Author:** ![s\_doi](https://discourse.datamethods.org/letter_avatar_proxy/v4/letter/s/ea666f/32.png) [@s\_doi](https://discourse.datamethods.org/u/s_doi)\
**Post date:** [March 10, 2025, 5:31pm UTC](https://discourse.datamethods.org/t/should-one-derive-risk-difference-from-the-odds-ratio/4403/631 "2025-03-10T17:31:11Z")

</div>

If SR is given by:  
 ![SR](https://discourse.datamethods.org/uploads/default/original/2X/2/275086fa123630ca653c3770242fb2caed148df4.png)  
Then this defaults to 1-_a_ where _a_ is the probability of death when the harmful intervention is applied and therefore has to be a constant.

---

<div class="post-metadata">

**Author:** ![AndersHuitfeldt](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/andershuitfeldt/32/113_2.png) [@AndersHuitfeldt](https://discourse.datamethods.org/u/AndersHuitfeldt)\
**Post date:** [March 10, 2025, 7:07pm UTC](https://discourse.datamethods.org/t/should-one-derive-risk-difference-from-the-odds-ratio/4403/632 "2025-03-10T19:07:10Z")

</div>

Yes. This is why the survival ratio is constant across different baseline risks (given a specific, simple data generating mechanism that may occur when exposure increases risk of death). Importantly, no similar result exists for other measures of effect.

Do you think this is somehow an argument against the survival ratio? If so, you need to flesh out the argument and give me something to respond to

---

<div class="post-metadata">

**Author:** ![AndersHuitfeldt](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/andershuitfeldt/32/113_2.png) [@AndersHuitfeldt](https://discourse.datamethods.org/u/AndersHuitfeldt)\
**Post date:** [March 17, 2026, 2:43pm UTC](https://discourse.datamethods.org/t/should-one-derive-risk-difference-from-the-odds-ratio/4403/633 "2026-03-17T14:43:25Z")

</div>

Those of you who followed this thread may be interested in an upcoming discussion meeting at the Royal Statistical Society, which will be held on March 24th. Sign up at [RSS - Discussion paper meetings](https://rss.org.uk/training-events/events/discussion-papers/) , either for the physical meeting at Imperial College or for the livestreamed event.

This work was originally motivated by our interest in models based on the switch relative risk, but took on a life of its own and turned into a new, general framework for regression modelling that includes most existing models as special cases.

RBC can be understood as a principled language for modular model building, such that the statistical properties of each parameter can be understood as being determined by the algebraic properties of the [flow](https://en.wikipedia.org/wiki/Flow_(mathematics))used to construct a model based on that parameter.

[Previous page](https://discourse.datamethods.org/t/should-one-derive-risk-difference-from-the-odds-ratio/4403.md?page=31)