@Anking-Maintainers
The zero and null value are overlapping and can be confusing
Don’t think this change is really necessary, find it to be pretty clear as is (personally)
the phrase “zero or the null value” makes it sound like zero and the null value are two different things, but in some cases (like mean differences), zero is the null value
the problem with the original is that it can actually mislead people because for ratios (like odds ratios or relative risk), the null value is one, not zero. so if someone reads this and thinks zero is always the null value, they might end up misjudging statistical significance when it comes to ratios
also with @beejumm image there’s now a slight contradiction between the text and the image, which adds to the confusion
Source?
I see there has been some disagreement on this. Going off the AMBOSS definition, would the dissenters (@Jwill @herstein.jacob) accept something like below, which I think makes thing more clear and avoids confusion, or still not sold on this change?
“…must not contain the null values {{c1::zero}} (for ratios) or {{c1::one}} (for differences)”
I see how this may be confusing, but the real issue IMO is that this card is focused specifically on ratios. If anything, would make more sense to say just clarify which type of “study”.
(Not trying to get too focused on stats outside of what is needed for Step, but it is possible to have a null value that is not 0 or 1, so I think that this card is referring to that possibility)
How’s this:
To be statistically significant at the 95% confidence level, the confidence interval must not contain the null value. For differences (e.g., mean differences), the null value is {{c1::zero}}. For ratios (e.g., odds ratios, relative risk), the null value is {{c1::one}}.
I just still feel like if the answer isn’t in reference to “zero”, the Q stem is going to tell you what the null value is and therefore remove any potential confusion.
Is there a QID that proves the current version is insufficient for our needs?
UW 20214: you need to know that one is a null value for RR to answer the question
UW 20103: you need to know one is a null value for the odds ratio to exclude other options
UW 3909: you need to know that one is a null value for RR to infer the answer
UW 4019: you need to know that one is a null value for RR to infer the answer
I haven’t done all the QID yet, but in general, I do believe that 1 as a null value is more common than zero, as questions tend to be about RR, OR, HR.
I am not sure I understand what you mean “the Q stem is going to tell you what the null value is”. The questions will assume we know that zero is a null value for differences and one is a null value for ratios.
I’m okay with not changing the card, but I don’t see why we shouldn’t or why changing it doesn’t make sense.
If anything, this card can be misleading because having zero in the confidence interval can still make the study significant if we’re talking about ratios (e.g., RR = 0.4, CI: 0–0.8). However, this card suggests [or can be understood] that my example would not be a significant result, which isn’t accurate
Even if there are other null values I don’t see this as a tested area or sth that we should care about at this level.
I agree with Mohannad (and big ups for those killer UW citations)! I also still think the card would benefit from additional clarification along the lines of my or @Eli.Tanenbaum’s suggestions to address the concerns of some.
Thanks for your detailed response. I’d support:
I would like to bring this back up,
tagged QID 20464 requires you to know:
The CI for differences CANNOT contain the NULL VALUE “zero”
In the Answer explanation, it states the above and then goes on to point out that it is important to also know for other circumstances:
The CI for relative risks or odds ratios CANNOT contain the NULL VALUE “one”
I think that this, in addition to what @mohannadkh10 stated above there is “significant” 
data to support an update to improve the accuracy of the card.
I have two thoughts:
-
For the results of a study to be statistically significant , the 95% confidence interval must not contain the null values: {{c1::zero}} (for differences) or {{c1::one}} (for ratios and risks) -
Or we just change the format of the card entirely, which I don’t usually prefer, but might just be the better call here (± the c2):
For the results of a study to be statistically significant , the 95% confidence interval must not contain the {{c2::null value}}
For differences = {{c1::zero}}
For odds ratios and relative risks = {{c1::one}}
Agree and fine with either one
@AnKing-Maintainers, please consider voting
For formatting, what about this:
For the results of a study to be statistically significant, the 95% confidence interval must not contain the null values:
For differences (e.g., ARR, AR): {{c1::Zero}}
For ratios (e.g, RR, OR): {{c1::One}}
Personally I would keep card as original, but I know for a fact that when I was doing this card, I knew something was off (because of the 1 and 0 thing) but I just went along with it because BNB and my school lectures explained this well enough that it didn’t cause confusion.
So for clarity sake, yes, this change is good for the majority
We were so close to coming to a conclusion on this never-ending discussion 
. I think @Ahmed7’s propsed wording is good and it has (statistically significant?) support. To avoid complicating this thread any further, I’ve created a new suggestion with Ahmed’s wording. Unless there are any objections or last minute changes, I’ll push this change in 72 hours, if that’s ok.
For the results of a study to be statistically significant, the 95% confidence interval must not contain the null value:
For differences (e.g., ARR, AR): {{c1::zero}}
For ratios (e.g., RR, OR): {{c1::one}}
Edit: PUSHED 