I am trying to use bootstrapping to enable the comparison of gait curves (specifically knee angle during a gait cycle) between two groups of subjects. Essentially, I was to see at what points in the gait cycle (from 0-100%) the two groups differ. Because this is an 'unpaired' analysis, I believe I need to use a stratified bootstrapping as described in the answer to this question. I have also read Michael Chernick's book, but am having trouble translating the stratified example into practical analysis.

My problem is that when I run the stratified bootstrapping I end up with very tight confidence intervals that are significantly tighter than the confidence intervals of either of my initial datasets. My understanding is that I can consider any position along the x-axis (time) in which the difference between the means of the two datasets (including the confidence intervals) does not include 0 as significant. However, with such tight confidence intervals, this seems to grossly overestimate significance. What am I doing wrong that leads to such tight confidence intervals?

To clarify, here is my experimental setup: Data set 1: Gait curve from population A. m subjects by n values over time; m=30 Data set 2: Gait curve from population B. m subjects by n values over time; m=30

Because the data are from different populations, they are "unpaired." Thus, I am trying to run a stratified bootstrapping algorithm. First I run bootstrapping on Data set 1 with 1000 samples. Then I run bootstrapping on Data set 2 with 1000 samples. Then I subtract the mean curve for Data set 1 and for Data set 2. I save this mean difference curve. The whole process is repeated 1000 times, and then I run the bootstrapping on these 1000 samples x n positions over time. Finally, I am looking at the mean and confidence intervals from this bootstrapping. Here is where I have incredibly tight confidence intervals, much more so than either of my initial data sets, or the confidence intervals from the initial two bootstraps. Any suggestions would be welcome!



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.