Back to the main page.
Bug 2049 - ft_statistics_montecarlo: implement confidence intervals for p-value
| Status | CLOSED FIXED |
| Reported | 2013-03-13 13:45:00 +0100 |
| Modified | 2014-03-12 12:21:39 +0100 |
| Product: | FieldTrip |
| Component: | core |
| Version: | unspecified |
| Hardware: | PC |
| Operating System: | Windows |
| Importance: | P3 normal |
| Assigned to: | Jörn M. Horschig |
| URL: | |
| Tags: | |
| Depends on: | |
| Blocks: | |
| See also: |
Jan-Mathijs Schoffelen - 2013-03-13 13:45:20 +0100
Implement it and store the result in two separate fields: lower bound and upper bound, e.g. stat.probupperbound = xxx stat.prbmclownbnd = yyy
Jan-Mathijs Schoffelen - 2013-03-13 13:48:30 +0100
stat.montecarlo stat.montecarhi
Jörn M. Horschig - 2013-03-20 11:44:09 +0100
hmm humm, ehm... if isfield(stat, 'posclusters') for i=1:length(stat.posclusters) stat.posclusters(i).var = Nrand*stat.posclusters(i).prob.*(1-stat.posclusters(i).prob); stat.posclusters(i).cirange = 1.96.*sqrt(stat.posclusters(i).var./Nrand); if stat.posclusters(i).prob<cfg.alpha && stat.posclusters(i).prob+stat.posclusters(i).cirange>=cfg.alpha warning('FieldTrip:posCluster_exceeds_alpha', sprintf('The p-value of positive cluster #%i is below %.3f but the confidence interval is above - consider increasing the number of permutations!', i, cfg.alpha)); end end end if isfield(stat, 'negclusters') for i=1:length(stat.negclusters) stat.negclusters(i).var = Nrand*stat.negclusters(i).prob.*(1-stat.negclusters(i).prob); stat.negclusters(i).cirange = 1.96.*sqrt(stat.negclusters(i).var./Nrand); if stat.negclusters(i).prob<cfg.alpha && stat.negclusters(i).prob+stat.negclusters(i).cirange>=cfg.alpha warning('FieldTrip:negCluster_exceeds_alpha', sprintf('The p-value of negative cluster #%i is below %.3f but the confidence interval is above - consider increasing the number of permutations!', i, cfg.alpha)); end end end hm?
Jörn M. Horschig - 2013-03-20 11:51:59 +0100
or rather a warning_once and be less specific about which cluster the niffy one is?
Jörn M. Horschig - 2013-03-20 11:53:39 +0100
tutorial data: Warning: The p-value of positive cluster #1 is below 0.025 but the confidence interval is above - consider increasing the number of permutations! > In ft_statistics_montecarlo at 396 In trunk\private\statistics_wrapper at 298 In ft_timelockstatistics at 111 In test_tutorial_clusterpermutationtimelock at 297
Jörn M. Horschig - 2013-03-20 11:55:24 +0100
last spammail: should the cirange field be there only for stat.posclusters and stat.negclusters or for stat.prob as well?
Jörn M. Horschig - 2013-03-20 14:03:25 +0100
Jm or Robert, before committing, is this correct? stddev = sqrt(stat.prob.*(1-stat.prob)); stat.cirange = 1.96*stddev/sqrt(Nrand);
Robert Oostenveld - 2013-03-22 11:29:32 +0100
(In reply to comment #6) I have no idea from the top of the head. I suggest to check with Eric (now CC)
Jörn M. Horschig - 2013-03-22 11:32:30 +0100
Eric, could you please check my last comment to verify that what I did is correct?