# [FieldTrip] Pb with ft_freqstatistics when averaged over channels : NaNs

laetitia.lalla at inserm.fr laetitia.lalla at inserm.fr
Tue Mar 29 10:50:46 CEST 2016

```

Dear FieldTrip community,

I have a similar problem that may be of help here.

It's my first message on this mailing list so, first of all, thank you
very much for this free toolbox and the help you're providing. I'm a PhD
student in Marseille (France) and I'm doing LFP recordings with tetrodes
or silicon probes in freely moving rats. (I apologize in advance for
this very detailed - and long - email).

I'm trying to compare the power in a certain frequency range between 2
conditions (Trial and Baseline). I can see a clear rising during the
trial and I want to run the statistics now.

Like Zita, I'd like to run a cluster permutation test on data with
frequency and time information, and my signal is averaged over channels.

I followed this tutorial :
http://www.fieldtriptoolbox.org/tutorial/natmeg/statistics. [1]Based on
this, this is what I did :

1) I cut epochs of 6s (trials), filter [0.1 250Hz] and average over
channels to have an averaged signal per probe (32 channels = 4 shanks =
1 probe).

2) I do the same for the baseline.

3) I create a structure data_all which gather all the epochs, with the
appropriate design matrix (0 if baseline, 1 if trial).

4) I do a Time Frequency Transform averaged over time with
ft_freqanalysis, 'mtmfft', with keeptrials='yes'. It give me, for each
trial, a profile Power=f(Freq) averaged during the time of the epoch.
The difference in my frequency range is very clear when I plot the
profiles.

5) I use ft_freqstatistics like this :

cfg = [];

cfg.channel = 'Str';

cfg.design = data_all.designmatrix;

cfg.method = 'montecarlo';

cfg.numrandomization = 500;

cfg.statistic = 'indepsamplesT';

cfg.ivar = 1;

cfg.correctm = 'cluster';

cfg.neighbours = [];

cfg.alpha=0.05;

cfg.tail=0;

cfg.correcttail='alpha';

[stat]=ft_freqstatistics(cfg, FR_alltrials)

And this is not working because I have only NaNs...

stat =

stat: [NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN
NaN NaN NaN NaN NaN NaN NaN NaN NaN]

prob: [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1]

cirange: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]

mask: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]

ref: [NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN
NaN NaN NaN NaN NaN NaN NaN NaN NaN]

dimord: 'chan_freq'

freq: [1x25 double]

label: {'Str'}

cfg: [1x1 struct]

Would you have an idea to resolve my problem ? Maybe I forgot to precise
a parameter somewhere ?

Best regards,

Laetitia

On 24-03-2016 13:15, Zita Eva Patai wrote:

> Dear FT-ers
>
> I would like to run a cluster permutation test on data where i have frequency and time information, but have averaged over channels.
>
> I am setting my cfg for ft_freqstatistics as usual, but asking for all times and channels (no averaging). I am also not setting the cfg.neighbours option, because there aren't any. In this case it won't run and asks for that setting. If I then set cfg.neighbour = []; then it complains it cannot find the .powspctrm of my data.
>
> If I do not ask for cluster correction, it runs through fine and I get a stat structure out of it, but this is uncorrected.
>
> Is there an easy way to fix this?
>
> I am using the FT inside SPM12.
>
> Thanks very much,
> zita
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip