Based on the residuals, I do cluster correction with 3dClustSim, that reports a minimum cluster size for significancy.
What would be the correct words for reporting the results of the Chi-Square tests and contrasts of 3dlmer?
I am interested in the grouptime interaction, so I mainly looked at the result of [-gltCode Int-Ctlr_over_Time 'Group : 1Intervention -1*Control Time :']
when I get a significant result there, can I call it a significant ‘interaction effect’? What is the difference to the result of Time:Group Chi-sq, which is slightly different to the other one (in regards to the cluster size)?
I mainly looked at the result of [-gltCode Int-Ctlr_over_Time 'Group : 1Intervention -1Control Time :']
when I get a significant result there, can I call it a significant 'interaction effect'?
Yes, the chi-square value provides the assessment for the statistical evidence of the interaction between Group Time.
What is the difference to the result of Time:Group Chi-sq, which is slightly different to the other one (just
below the cluster threshold suggested by 3dClustSim)?
Do you mean "-gltCode Int-Ctlr_over_Time 'Group : 1Intervention -1Control Time :' "? If so, the two essentially intend to achieve the same goal. The slight difference is due to two different ways to assess the statistical evidence.
Do you mean "-gltCode Int-Ctlr_over_Time 'Group :
1Intervention -1Control Time :' "? If so, the
two essentially intend to achieve the same goal.
The slight difference is due to two different ways
to assess the statistical evidence.
Thanks for your reply. Yes, I meant the difference between the chi-sqare values (from group*time, generated automatically) and the z values from that contrast. So if they're essentially the same, can I report either one?
Also, a question about reporting the main effects: My goal was to find out if there is a group*time effect. Do I have to report the main effects Time and Group, too?
I meant the difference between the chi-sqare values (from group*time, generated automatically) and the z values
from that contrast. So if they're essentially the same, can I report either one?
Yes, either one is fine since they are conceptually equivalent.
My goal was to find out if there is a group*time effect. Do I have to report the main effects Time and Group, too?
No, you don't have to if the main effects are not your focus.
I do have one more question regarding these results, I hope you don’t mind:
Does the mathematical test in 3dLMEr, which produced the z values for my above mentioned group*time interaction effect, have a name?
Since you said chi-square results for group:time and z values for [-gltCode Int-Ctlr_over_Time ‘Group : 1Intervention -1Control Time :’] are conceptually equivalent and it’s possible to use either one, I decided to use the z values, because those also show the direction of the group*time effect.
I just want to make sure that readers of the papers know which of those values were chosen. So if I had taken the the chi-square values, I would write "Chi-square tests show that … ". - Now that I’ve chose the z values instead: What is the correct phrase to use for the z values?
Does the mathematical test in 3dLMEr, which produced the z values for my above mentioned group*time interaction effect, have a name?
group:time is the interaction between group and time. It assesses the difference of the slope (or marginal) effect for time between the two groups.
Since you said chi-square results for group:time and z values for [-gltCode Int-Ctlr_over_Time
'Group : 1Intervention -1Control Time :'] are conceptually equivalent and it's possible to use
either one, I decided to use the z values, because those also show the direction of the group*time effect.
Yes, it makes sense to use the Z-values because it contains the directionality information.
What is the correct phrase to use for the z values?
It shows the statistical evidence/strength for the group difference for the slope effect of time. It would be better to show the magnitude of the group difference (the sub-brick right before the Z-values) for the slope effect of time (as UnderLay on the AFNI GUI).
It shows the statistical evidence/strength for the group difference for the slope effect of time.
It would be better to show the magnitude of the group difference (the sub-brick right before the Z-values) for the slope effect of time
(as UnderLay on the AFNI GUI).
Thanks! In fact, I am actually using the z values sub-brick to threshold, and then the sub-brick before as overlay. Is that what you mean?
However, we found that values of -0.07 (for example) are very hard to grasp on their own. That is why we are focussing on the presence of interaction effects (along with their direction) first. (And then later on present figures regarding the magnitude.)
>> Does the mathematical test in 3dLMEr, which produced the z values for my above mentioned group*time interaction effect, have a name?
group:time is the interaction between group and time. It assesses the difference of the slope (or marginal) effect for time between the two groups.
Okay, so these are basically show the slope difference between both groups. Someone was asking me if there was a specific word to explain the background of these values. Similar to that you could write e.g. "Results of chi-sqare test/ F-test indicate that there is a significant group x time interaction at (peak x,y,z)" if you took values from the chi-square sub-bricks instead of the above mentioned z sub-brick...
Also, do you know any reference that mentions the differences between the chi-square sub-brick and the one that includes directions and explains why it is possible to take either either one as a source for group x time interactions? Someone was wondering if [-gltCode Int-Ctlr_over_Time 'Group : 1Intervention -1Control Time :'] is not already a pos-hoc test and you have to do some kind of unidirectional omnibus-/F-/Chi-Square-/... Test first before checking direction of the interaction effect (=checking the first sub-bricks with chi-square values). Unfortunately, I'm lacking background knowledge to answer that question well. (As far as I understand, however, an example for a post-hoc test would be 'Group : 1*Intervention Time : ')
I am actually using the z values sub-brick to threshold, and then the sub-brick before as overlay. Is that what you mean?
Yes, this is usually preferable.
we found that values of -0.07 (for example) are very hard to grasp on their own.
Depending on the context, that value may have some physical meaning and is interpretable, unlike statistical value (which is dimensionless).
do you know any reference that mentions the differences between the chi-square sub-brick and the one that includes
directions and explains why it is possible to take either either one as a source for group x time interactions?
These two statistical values are computed from two different perspectives. The first one with chi-square is considered an omnibus test: it's usually computed as F-statistics under the conventional AOVA; in 3dLMEr it's assessed with chi-square. An omnibus statistics is usually formulated across two or more effects. The second one (Z-statistic) is assessed as a contrast between two groups: this is a special case for the omnibus test between two effects. These two are equivalent simply because there are only two groups in your case. If you had three groups, you would not have this equivalence situation.
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.