Confidence intervals for evoked potentials

Robert Oostenveld r.oostenveld at FCDONDERS.RU.NL
Tue Jul 11 15:50:09 CEST 2006


Hi Jasper,

On 11 Jul 2006, at 11:43, Jasper Poort wrote:
> Hi all, I have some questions regarding confidence intervals for
> timelocked data in FieldTrip. I noted that a field .var is returned
> by timelockanalysis
>
> line 510 timelockanalysis
> % compute the variance
> tmp1 = repmat(dofvec(:)', [nchan 1]);
> tmp2 = repmat(dofvec(:)', [nchan 1])-1;
> var = (ss - (s.^2)./tmp1) ./ tmp2;
>
> Assuming that the sampling distribution is normal I can presumably
> use this field for computation of confidence intervals of the .avg
> field?

Yes, you can use it to compute the channel-timepoint specific
standard-error of mean (SEM).

> Is nonparametric estimation of variance using a bootstrap or
> jackknife also implemented somewhere? In general I'm just curious
> to know how others compute confidence intervals for evoked
> potentials in FieldTrip, for example to get an idea at what point
> in time two conditions start to differ,

Bootstrap and jackknife estimates of variance are only different from
teh normal variance estimate if the measure involves a nonlinear or
biased estimate (i.e. coherence). The jackknife estimate of variance
of the ERP is exactly the same as the normal estimate of variance,
since the ERP is a plain (linear) average. The same holds for
bootstrap estimates.

So that means that for ERPs the jackknife and bootstrap only would be
a computationally expensive methods, resulting in the same variance
as usual. Therefore they are not implemented.

best regards,
Robert



More information about the fieldtrip mailing list