help

help >

**RE: Permutation Testing for One Group/Time Point**Sep 9, 2019 06:09 PM | Andrew Zalesky

RE: Permutation Testing for One Group/Time Point

Hi Cassie,

For a one-sample t-test, permutation work by randomly flipping the signs of the values. Flipping signs should not change the sample mean on average of a random variable under the null hypothesis of zero mean.

However, note that testing a correlation between change in symptom scores in a single patient group is not a one-sample test. This is still a t-test (not one-sample t-test). The permutation in this case is still done by randomizing the correspondence between patient and symptom scores. For example, patient 7 might be assigned the symptom score for patient 4.

In summary, don't use a one-sample test to test an association between a symptom score in single patient group. Use a t-test.

Andrew

For a one-sample t-test, permutation work by randomly flipping the signs of the values. Flipping signs should not change the sample mean on average of a random variable under the null hypothesis of zero mean.

However, note that testing a correlation between change in symptom scores in a single patient group is not a one-sample test. This is still a t-test (not one-sample t-test). The permutation in this case is still done by randomizing the correspondence between patient and symptom scores. For example, patient 7 might be assigned the symptom score for patient 4.

In summary, don't use a one-sample test to test an association between a symptom score in single patient group. Use a t-test.

Andrew

*Originally posted by Cassie C:*Apologies if this question has already been
answered in the forum.

I'm aware that permutation testing with two groups involves randomly exchanging data points from sample A with data points from data B but I am wondering how this works for a one-sample t-test.

For example, I have a single patient group and am examining how a measure of change in symptom scores correlates with connectivity at time point 1, using a a contrast of [0 -1] for negative correlation and a contrast of [0 1] for positive correlation. Wondering how the permutation testing actually works for this single group.

I'm aware that permutation testing with two groups involves randomly exchanging data points from sample A with data points from data B but I am wondering how this works for a one-sample t-test.

For example, I have a single patient group and am examining how a measure of change in symptom scores correlates with connectivity at time point 1, using a a contrast of [0 -1] for negative correlation and a contrast of [0 1] for positive correlation. Wondering how the permutation testing actually works for this single group.

## Threaded View

Title | Author | Date |
---|---|---|

Cassie C |
Sep 8, 2019 | |

Andrew Zalesky |
Sep 9, 2019 | |