0todd0000
commented
2 years ago Thank you for these questions!
- The endpoints for the first cluster (from left-to-right) can be retrieved in Matlab as follows:
spmi = spm1d.stats.ttest(y).inference(0.05);
endpoints = spmi.clusters{1}.endpoints;
- Since the result is reported as
d = 1.03 - 1.05, I think this means that the d-value ranges between 1.03 and 1.05 within this region.
- Yes. Just like means and SD values can be calculated at an arbitrary point along the 1D domain, so too can functions of the mean and SD, like the t-value and like Cohen's d. Cohen's d for an independent samples test (i.e., a two-sample test) can be calculated using the code in the cited issue: #76. A MATLAB translation is:
Q = 101; %number of continuum nodes
JA = 8; %sample size, Group A
JB = 8; %sample size, Group B
yA = randn( JA, Q ); %random sample, GroupA
yB = randn( JB, Q ); %random sample, GroupB
mA = mean( yA, 1 ); %mean, Group A
mB = mean( yB, 1 ); %mean, Group B
sA = std( yA, [], 1 ); %st.dev., Group A
sB = std( yB, [], 1 ); %st.dev., Group B
s = sqrt( ( (JA-1)*sA.^2 + (JB-1)*sB.^2 ) / (JA+JB) ); %pooled st.dev.
d = ( mA - mB ) ./ s; %Cohen's d
- I think it's fine to report d-values, provided you do not use standard interpretations. Consider a threshold d-value like
d=1.20, which is often interpreted as indicative of a "very large" effect. The problem is that --- when there is no true effect --- a random 1D process is much more likely to produce a d-value of 1.20 than a random 0D process. So the d-value itself is fine. It's only the interpretation that is potentially problematic. So I suggest reporting the d-values if requested, and then citing the literature (for example: this paper) to show that standard, 0D interpretations of various statistical quantities like t-values and d-values cannot be applied directly to 1D processes.
fangtongzhang
Hi, I have 4 questions regarding the independent samples t-test in SPM1d statistical analysis：
After I conducted spm statistical analysis in matlab, the result shows that there is a significant difference in a data cluster, but there is no start frame and end frame as in the literature [Ankle kinematics, center of pressure progression, and lower extremity muscle activity during a side-cutting task in participants with and without chronic ankle instability], so my question is: if I want to have the results of the calculation of the start frame and end frame of the region with significant difference, how can I implement it in matlab?
The above literature reports the d-value of the spm analysis results, but it is a numerical value, not in the form of a curve, as you described in the forum #54 #76, so did he report the d-value improperly?
After browsing the #76 discussion, would this curve of cohen’s d be a result of calculating the cohen's d value for each corresponding point, such as the 34th point on the cohen's d curve, from the mean, standard deviation and pooled std of the 34th data for all subjects? Moreover, how to calculate the cohen's d for independent sample t-test using matlab?
I understood what you mentioned in discussion #54, suggesting not to report d-values. Because the d-values in the spm analysis are different from the d-values we used to get from statistical analysis of discrete data, and the d-values obtained from the spm analysis may be larger than the actual ones, so reporting the d-values is not that meaningful. But if the reviewers insist that we should report d-values, then my concern is that it seems a bit strange to report d-values in the form of a curve plot? Since I don't have a deep enough understanding of SPM analysis approach, I might be a bit wordy on these questions and would appreciate your answers. Please let me know if I have not made it clear enough, thanks!