help > Any effect vs. Simple main effect
Showing 1-5 of 5 posts
Feb 22, 2017 06:02 PM | Jenna Traynor - McMaster University
Any effect vs. Simple main effect
Hi Alfonso,
I am running a within-group bivariate correlation analysis looking at the association between functional connectivity and 6 different behavioural scores, using Conn V.17. I have a question regarding my findings:
When I set up the findings to look for any effect among my behavioural variables 1-6, I select my patient group and the six variables and set up the contrast as so: [0 1 0 0 0 0 0; 0 0 1 0 0 0 0; 0 0 0 1 0 0 0; 0 0 0 0 1 0 0; 0 0 0 0 0 1 0; 0 0 0 0 0 0 1]. I then select all of my seeds and enter into results explorer, which shows that there is no significant effect at all.
However, if while in the second level results window, I select the simple main effect of each of my behavioural scores separately i.e. [0 1], and all of my seeds, and then enter into results explorer, I get an abundance of significant values for each behavioural score.
My experimental question is: is there an association between FC and each of my behavioural scores - does this mean that I can look at the simple main effects even if the overall F test was not significant? If so, how can I explain this statistically? I know there is some practice of looking at simple main effects despite overall ANOVA being insignificant and I am wondering if that applies here?
Thank you,
Jenna
I am running a within-group bivariate correlation analysis looking at the association between functional connectivity and 6 different behavioural scores, using Conn V.17. I have a question regarding my findings:
When I set up the findings to look for any effect among my behavioural variables 1-6, I select my patient group and the six variables and set up the contrast as so: [0 1 0 0 0 0 0; 0 0 1 0 0 0 0; 0 0 0 1 0 0 0; 0 0 0 0 1 0 0; 0 0 0 0 0 1 0; 0 0 0 0 0 0 1]. I then select all of my seeds and enter into results explorer, which shows that there is no significant effect at all.
However, if while in the second level results window, I select the simple main effect of each of my behavioural scores separately i.e. [0 1], and all of my seeds, and then enter into results explorer, I get an abundance of significant values for each behavioural score.
My experimental question is: is there an association between FC and each of my behavioural scores - does this mean that I can look at the simple main effects even if the overall F test was not significant? If so, how can I explain this statistically? I know there is some practice of looking at simple main effects despite overall ANOVA being insignificant and I am wondering if that applies here?
Thank you,
Jenna
Mar 1, 2017 02:03 AM | Alfonso Nieto-Castanon - Boston University
RE: Any effect vs. Simple main effect
Hi Jenna,
Yes, the combination of entering all of your 6 covariates simultaneously as well as entering all of your seeds simultaneously is perhaps making the resulting analysis severely underpowered (this depends on the number of subjects, number of seeds&covariates, and expected effect sizes of the covariate-functional associations). If you have a priori hypotheses about individual covariates and/or individual seeds you should test those separately (in these cases it is often useful to use something like a study preregistration to help you have a clear plan of confirmatory analyses, vs. any other exploratory analyses that may be guided by what you actually find in your data). If you do not have specific a priori hypotheses about individual covariates one possible way to reduce the dimensionality of your predictors (and increase the power of your analyses) is to perform a PCA decomposition first of your 6 covariates and enter just the first few component scores into your second-level analyses. Similarly, if you do not have specific hypotheses about the individual seeds (and there are many seed areas that you would like to simultaneously test), you could use for example masked-ICA to find first a reduced set of potential subnetworks within those areas, and then use the resulting (reduced set of) networks as seeds instead of your (larger set of) original seed areas.
Hope this helps
Alfonso
Originally posted by Jenna Traynor:
Yes, the combination of entering all of your 6 covariates simultaneously as well as entering all of your seeds simultaneously is perhaps making the resulting analysis severely underpowered (this depends on the number of subjects, number of seeds&covariates, and expected effect sizes of the covariate-functional associations). If you have a priori hypotheses about individual covariates and/or individual seeds you should test those separately (in these cases it is often useful to use something like a study preregistration to help you have a clear plan of confirmatory analyses, vs. any other exploratory analyses that may be guided by what you actually find in your data). If you do not have specific a priori hypotheses about individual covariates one possible way to reduce the dimensionality of your predictors (and increase the power of your analyses) is to perform a PCA decomposition first of your 6 covariates and enter just the first few component scores into your second-level analyses. Similarly, if you do not have specific hypotheses about the individual seeds (and there are many seed areas that you would like to simultaneously test), you could use for example masked-ICA to find first a reduced set of potential subnetworks within those areas, and then use the resulting (reduced set of) networks as seeds instead of your (larger set of) original seed areas.
Hope this helps
Alfonso
Originally posted by Jenna Traynor:
Hi Alfonso,
I am running a within-group bivariate correlation analysis looking at the association between functional connectivity and 6 different behavioural scores, using Conn V.17. I have a question regarding my findings:
When I set up the findings to look for any effect among my behavioural variables 1-6, I select my patient group and the six variables and set up the contrast as so: [0 1 0 0 0 0 0; 0 0 1 0 0 0 0; 0 0 0 1 0 0 0; 0 0 0 0 1 0 0; 0 0 0 0 0 1 0; 0 0 0 0 0 0 1]. I then select all of my seeds and enter into results explorer, which shows that there is no significant effect at all.
However, if while in the second level results window, I select the simple main effect of each of my behavioural scores separately i.e. [0 1], and all of my seeds, and then enter into results explorer, I get an abundance of significant values for each behavioural score.
My experimental question is: is there an association between FC and each of my behavioural scores - does this mean that I can look at the simple main effects even if the overall F test was not significant? If so, how can I explain this statistically? I know there is some practice of looking at simple main effects despite overall ANOVA being insignificant and I am wondering if that applies here?
Thank you,
Jenna
I am running a within-group bivariate correlation analysis looking at the association between functional connectivity and 6 different behavioural scores, using Conn V.17. I have a question regarding my findings:
When I set up the findings to look for any effect among my behavioural variables 1-6, I select my patient group and the six variables and set up the contrast as so: [0 1 0 0 0 0 0; 0 0 1 0 0 0 0; 0 0 0 1 0 0 0; 0 0 0 0 1 0 0; 0 0 0 0 0 1 0; 0 0 0 0 0 0 1]. I then select all of my seeds and enter into results explorer, which shows that there is no significant effect at all.
However, if while in the second level results window, I select the simple main effect of each of my behavioural scores separately i.e. [0 1], and all of my seeds, and then enter into results explorer, I get an abundance of significant values for each behavioural score.
My experimental question is: is there an association between FC and each of my behavioural scores - does this mean that I can look at the simple main effects even if the overall F test was not significant? If so, how can I explain this statistically? I know there is some practice of looking at simple main effects despite overall ANOVA being insignificant and I am wondering if that applies here?
Thank you,
Jenna
Mar 5, 2017 07:03 PM | Jenna Traynor - McMaster University
RE: Any effect vs. Simple main effect
Hi Alfonso,
Thank you so much for your help. I have chosen to follow your advice and use a masked-ICA to find a potential set of subnetworks to enter into the analysis. However, it has been a long time since I first ran the ROI step in SetUp, and since then I updated to Conn v.17. Therefore, in order to examine ICA networks I need to go back to set up and enable voxel-to-voxel analysis. I tried to do this (skipped preprocessing, as my images are already preprocessed in SPM) and when I get to step 5/7 ROIs I get this error:
ERROR DESCRIPTION:
Error using conn_process (line 763)
duplicated ROI name BA.9 (R). Dorsolateral Prefrontal Cortex
Error in conn_process (line 15)
case 'setup', disp(['CONN: RUNNING SETUP STEP']); conn_process([0:4,4.5,5]);
Error in conn (line 3195)
else conn_process('setup');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM8 + Beamforming DEM FieldMap MEEGtools TOM8
Matlab v.2016a
storage: 49.3Gb available
I thought maybe it was because CONN was using my old ROI files that I loaded from Conn V.14, so I reloaded all of my ROIs from Conn V.17 utils folder, but I am getting the same error. Any thoughts on how to fix this?
Thank you,
Jenna
Thank you so much for your help. I have chosen to follow your advice and use a masked-ICA to find a potential set of subnetworks to enter into the analysis. However, it has been a long time since I first ran the ROI step in SetUp, and since then I updated to Conn v.17. Therefore, in order to examine ICA networks I need to go back to set up and enable voxel-to-voxel analysis. I tried to do this (skipped preprocessing, as my images are already preprocessed in SPM) and when I get to step 5/7 ROIs I get this error:
ERROR DESCRIPTION:
Error using conn_process (line 763)
duplicated ROI name BA.9 (R). Dorsolateral Prefrontal Cortex
Error in conn_process (line 15)
case 'setup', disp(['CONN: RUNNING SETUP STEP']); conn_process([0:4,4.5,5]);
Error in conn (line 3195)
else conn_process('setup');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM8 + Beamforming DEM FieldMap MEEGtools TOM8
Matlab v.2016a
storage: 49.3Gb available
I thought maybe it was because CONN was using my old ROI files that I loaded from Conn V.14, so I reloaded all of my ROIs from Conn V.17 utils folder, but I am getting the same error. Any thoughts on how to fix this?
Thank you,
Jenna
Mar 7, 2017 01:03 AM | Alfonso Nieto-Castanon - Boston University
RE: Any effect vs. Simple main effect
Hi Jenna,
I believe there might be a patch for this issue described here (http://www.nitrc.org/forum/message.php?m... ), or alternatively simply install 17b which already contains this same patch. Let me know if everything seems to be working fine now
Best
Alfonso
Originally posted by Jenna Traynor:
I believe there might be a patch for this issue described here (http://www.nitrc.org/forum/message.php?m... ), or alternatively simply install 17b which already contains this same patch. Let me know if everything seems to be working fine now
Best
Alfonso
Originally posted by Jenna Traynor:
Hi Alfonso,
Thank you so much for your help. I have chosen to follow your advice and use a masked-ICA to find a potential set of subnetworks to enter into the analysis. However, it has been a long time since I first ran the ROI step in SetUp, and since then I updated to Conn v.17. Therefore, in order to examine ICA networks I need to go back to set up and enable voxel-to-voxel analysis. I tried to do this (skipped preprocessing, as my images are already preprocessed in SPM) and when I get to step 5/7 ROIs I get this error:
ERROR DESCRIPTION:
Error using conn_process (line 763)
duplicated ROI name BA.9 (R). Dorsolateral Prefrontal Cortex
Error in conn_process (line 15)
case 'setup', disp(['CONN: RUNNING SETUP STEP']); conn_process([0:4,4.5,5]);
Error in conn (line 3195)
else conn_process('setup');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM8 + Beamforming DEM FieldMap MEEGtools TOM8
Matlab v.2016a
storage: 49.3Gb available
I thought maybe it was because CONN was using my old ROI files that I loaded from Conn V.14, so I reloaded all of my ROIs from Conn V.17 utils folder, but I am getting the same error. Any thoughts on how to fix this?
Thank you,
Jenna
Thank you so much for your help. I have chosen to follow your advice and use a masked-ICA to find a potential set of subnetworks to enter into the analysis. However, it has been a long time since I first ran the ROI step in SetUp, and since then I updated to Conn v.17. Therefore, in order to examine ICA networks I need to go back to set up and enable voxel-to-voxel analysis. I tried to do this (skipped preprocessing, as my images are already preprocessed in SPM) and when I get to step 5/7 ROIs I get this error:
ERROR DESCRIPTION:
Error using conn_process (line 763)
duplicated ROI name BA.9 (R). Dorsolateral Prefrontal Cortex
Error in conn_process (line 15)
case 'setup', disp(['CONN: RUNNING SETUP STEP']); conn_process([0:4,4.5,5]);
Error in conn (line 3195)
else conn_process('setup');
Error in conn_menumanager (line 119)
feval(CONN_MM.MENU{n0}.callback{n1}{1},CONN_MM.MENU{n0}.callback{n1}{2:end});
CONN v.17.a
SPM8 + Beamforming DEM FieldMap MEEGtools TOM8
Matlab v.2016a
storage: 49.3Gb available
I thought maybe it was because CONN was using my old ROI files that I loaded from Conn V.14, so I reloaded all of my ROIs from Conn V.17 utils folder, but I am getting the same error. Any thoughts on how to fix this?
Thank you,
Jenna
Mar 13, 2017 01:03 PM | Jenna Traynor - McMaster University
RE: Any effect vs. Simple main effect
Hi Alfonso,
Thank you very much, I downloaded the latest version of CONN and it worked perfectly. I did a group-PCA, however, I am wondering how to do the PCA with just my group of patients. I also have a group of control scans that are loaded into the set up. Part of the analysis was comparing connectivity between the groups. The second part of the analysis is looking at within group connectivity in my patient group only and the association between 6 behavioural scores/covariates and connectivity within this patient group (by doing a PCA first).
How can I performed a PCA within the patient group only? I am assuming this is what I would want to do, since I found many differences is rs-fc between the two groups. Now, I am only interested in the principle components within this patient group and the correlation between these principle ROIs/networks and scores.
Thank you,
Jenna
Thank you very much, I downloaded the latest version of CONN and it worked perfectly. I did a group-PCA, however, I am wondering how to do the PCA with just my group of patients. I also have a group of control scans that are loaded into the set up. Part of the analysis was comparing connectivity between the groups. The second part of the analysis is looking at within group connectivity in my patient group only and the association between 6 behavioural scores/covariates and connectivity within this patient group (by doing a PCA first).
How can I performed a PCA within the patient group only? I am assuming this is what I would want to do, since I found many differences is rs-fc between the two groups. Now, I am only interested in the principle components within this patient group and the correlation between these principle ROIs/networks and scores.
Thank you,
Jenna