open-discussion > mc correction in StatNonParamTestPDM
Showing 1-3 of 3 posts
Aug 3, 2015 06:08 PM | Lara Foland-Ross - Stanford University
mc correction in StatNonParamTestPDM
Hello,
I'm using StatNonParamTestPDM to assess group differences in shape and have some basic questions regarding correction for multiple comparisons. I am running the following command:
StatNonParamTestPDM listfile.txt -out stat_results -surfList -numPerms 50000 -signLevel .05 -signSteps 1000
1. the text files for the FDR p values appear after running the above, but it contains only header information - no p values:
NUMBER_OF_POINTS = 0
DIMENSION = 1
TYPE = Scalar
...any tips on troubleshooting this error?
2. Can you please clarify the function of the signSteps flag? Reading up on this in your original paper, it seems that the maximal threshold is divided by the number of threshold steps (so here, is 0.05/1000 = 0.005). Does this imply therefore that correction for multiple comparisons is accomplished by using a more stringent significance threshold of 0.005?
While I understand that StatNonParamTestPDM is no longer supported, I'd still appreciate an answer to these two (hopefully) quick questions.
Many thanks in advance!
Lara
I'm using StatNonParamTestPDM to assess group differences in shape and have some basic questions regarding correction for multiple comparisons. I am running the following command:
StatNonParamTestPDM listfile.txt -out stat_results -surfList -numPerms 50000 -signLevel .05 -signSteps 1000
1. the text files for the FDR p values appear after running the above, but it contains only header information - no p values:
NUMBER_OF_POINTS = 0
DIMENSION = 1
TYPE = Scalar
...any tips on troubleshooting this error?
2. Can you please clarify the function of the signSteps flag? Reading up on this in your original paper, it seems that the maximal threshold is divided by the number of threshold steps (so here, is 0.05/1000 = 0.005). Does this imply therefore that correction for multiple comparisons is accomplished by using a more stringent significance threshold of 0.005?
While I understand that StatNonParamTestPDM is no longer supported, I'd still appreciate an answer to these two (hopefully) quick questions.
Many thanks in advance!
Lara
Aug 3, 2015 06:08 PM | Martin Styner
RE: mc correction in StatNonParamTestPDM
First, since StatNonParamTestPDM is really old, it is based on
itk-meta meshes (and not vtk meshes). There is a VTK2Meta converter
in the distribution that allows you to convert the fileformats. So,
convert your vtk meshes and change the information in the csv file,
hopefully then the files should be read correctly and you will get
a result.
The signSteps, is simply a way to get a step-wise description of your corrected threshold, as otherwise you simply get a binary pass/fail threshold decision (this is only the case for the permutation based correction, FDR does not need such a step information). Best don't change it.
Martin
The signSteps, is simply a way to get a step-wise description of your corrected threshold, as otherwise you simply get a binary pass/fail threshold decision (this is only the case for the permutation based correction, FDR does not need such a step information). Best don't change it.
Martin
Aug 3, 2015 06:08 PM | Lara Foland-Ross - Stanford University
RE: mc correction in StatNonParamTestPDM
Great. Thanks very much!
Lara
Lara