Hello everyone!
Would you consider an Odds ratio of 1.18 (95% CI 1.01 - 1.35) statistically significant?
What I mean is that the lower bound of the 95% CI is pretty close to 1.00. The 1.01 value would represent a really small risk association of 1%. Would you still consider these results statistically significan but clinically irrelevant? or just not statistically significant?
I would like to hear your thoughts.
Thanks in advance.
It is statistically significant at the 5% level (alpha = .05) since 1 is not included in the 95% CI. Whether this effect is clinically or otherwise "practically" significant is a question that needs to be addressed/judged by those who are active in this field/area of research. Small effect sizes can sometimes be quite meaningful statistically, especially when they concern things like the risk of dying, becoming seriously ill, etc. Applied to an entire population the increased probability of something good or bad happening may be quite relevant.
Hello Andrés!
This is a topic that I've given some thought on.
For starters, the odds ratio is definitely significant at the p
Good point Anastasios Stylianos Tsarouchas : the size of the OR of a continuous variable depends on the unit used to give the continuous variable. So in order to get an impression of the practical relevance or an observed OR it should be based on a practically meaningful unit of the variable.
However, I am sceptical about using an SD as such a practically meaningful unit. It sometimes may be ok, but it may also be rather nonsensical in other cases. And even if it would be ok, it can be a considerable problem to estimate the SD with sufficient precision (particularily with relatively small samples). A further complication comes in when the model if multivariable, where the SD of the marginal distribution may not be relevant at all and some kind of "adjusted SD" may be difficult to define or to interpret. Having a unit that is practically meaningful on theoretical grounds would be clearly preferrable over anything that is derived just from an observed sample.
Yes, it's statistically significant even lower bound of CI is close to one.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude