I am new to this forum, and I need some helps in bootstrapping.

I am using Matlab.

So, here's the study design:

I have three groups (Group1, Group2, Group3), 30 subjects/group each. They all have some brain imaging scans (FDG-PET).

Within each group, I calculated some parameter (called global efficiency using graph theory analysis).

So, each group has only one value (i.e., global efficiency), so I cannot do statistical comparison for the group differences.

Now, using bootstrap resampling, I could do some statistical comparisons, correct?

Within each group, I did 1,000 bootstrap resampling (using bootstrp in Matlab), and I calculated the "global efficiency" from each boostrap samples, so I have 1,000 global efficiency values for each group.

Now, I have seen other published works that some people just do one-way ANOVA and post-hoc tests on this 3,000 bootstrap samples, which I believe seriously overestimate the significance.

My initial alternative is to calculate 95% confidence interval empirically from each groups' 1,000 bootstrap samples, e.g., sort the "bootstrap global efficiency" and take the 25th and 975th values and compare across the groups.

For example, if the group1's 25th value is bigger than group2's 975th value, group1's global efficiency is greater than group2 (estimated p<0.05). Is this correct way to do this?

Nevertheless, with this kind of approach, I cannot actually do one-way ANOVA-like test...is there any way?

Please let me know.

Best,

Ji-Hyun