Typo in FC? Confidence Interval

[MudPhud20XX]

---

Members don't see this ad.

Can anyone confirm that this is a typo from FC?

• If the CI range includes 0, the null hypothesis is not rejected → no statistically significant difference was found

It should be "If the CI range includes 1" right?

 

[Transposony]

Depends what you are calculating--Relative risk or Odds ratio.

For Relative risk CI range including 1 is not significant. In other word 1/1 risk is no risk.

For Odds ratio CI range including 0 is not significant. In other words the odds of something significant happening are Zero.

EDIT: It's same for Odds Ratio.

Here is a great explanation.

 

[SBR249]

Depends what you are calculating--Relative risk or Odds ratio.

For Relative risk CI range including 1 is not significant. In other word 1/1 risk is no risk.

For Odds ratio CI range including 0 is not significant. In other words the odds of something significant happening are Zero.

Here is a great explanation.

I would disagree with the above. Any time you are taking a ratio of 2 numbers be it risk (as in relative risk) or odds (as in odds ratio), a 1 is always NOT significant because mathematically it signifies that the 2 numbers are the same and you get a ratio = 1. The only exception is relative risk reduction which has 2 steps with the first being a subtraction (see below).

In the case of a CI that includes 0, the only time that is NOT significant is when your calculations involve a subtraction such as absolute risk or a risk reduction (absolute or relative). In that case, if the two numbers you are comparing is the same, then the subtraction will give you a 0 meaning it's NOT significant.

Edit: I think the document in the link above agrees with me on pg.3 when it says the measure of no effect for both RR and OR would be 1.

 

[thehundredthone]

The confidence interval can include 0 and have the null hypothesis not be rejected, if the null hypothesis is θ = 0. This is statistically correct. However, when we talk about RR and OR, the null hypothesis is that RR/OR = 1 because they are ratios. So Firecracker isn't technically wrong, but it is being counter-intuitive.

EDIT: One example would be the difference of means.

 

[MudPhud20XX]

The confidence interval can include 0 and have the null hypothesis not be rejected, if the null hypothesis is θ = 0. This is statistically correct. However, when we talk about RR and OR, the null hypothesis is that RR/OR = 1 because they are ratios. So Firecracker isn't technically wrong, but it is being counter-intuitive.

EDIT: One example would be the difference of means.

Thank you very much! That clarifies my confusion. Thank you all!!!

 

[dyspareunia]

If the CI between two means includes 0 that means there is some overlap between the two groups. If the CIs overlap at all (i.e. include 0) then you have to consider the fact that there may be no difference between the two groups.

 

[Transposony]

I would disagree with the above. Any time you are taking a ratio of 2 numbers be it risk (as in relative risk) or odds (as in odds ratio), a 1 is always NOT significant because mathematically it signifies that the 2 numbers are the same and you get a ratio = 1. The only exception is relative risk reduction which has 2 steps with the first being a subtraction (see below).

In the case of a CI that includes 0, the only time that is NOT significant is when your calculations involve a subtraction such as absolute risk or a risk reduction (absolute or relative). In that case, if the two numbers you are comparing is the same, then the subtraction will give you a 0 meaning it's NOT significant.

Edit: I think the document in the link above agrees with me on pg.3 when it says the measure of no effect for both RR and OR would be 1.

Thank You.
You are right.
I stand corrected.