baseline for analyzing pre- and post-stimulus activity

[Raquel London]

Hi Mike,

I hope all is well over there. I have a question about choosing a baseline when looking at pre-stimulus activity.

I have EEG data recorded during a classic Navon task. We presented subjects with large letters made up of smaller letters and they had to respond as fast as possible to either the large letters (global attention condition) or small letters (local attention condition). The local and global letters could be the same (congruent condition) or different (incongruent condition).

To get a more robust effect with fewer trials, global attention vs. local attention was a blocked variable. Congruent vs incongruent was randomized within blocks. I am interested in pre-stimulus activity as well as post-stimulus activity. Because the global vs local condition was blocked, we don't have a clear baseline time period for each trial. There is no cue telling them if they have to respond to the global or local letter. I was thinking of calculating a condition-average baseline based on the entire epoch, but I want to be able to compare pre-stimulus alpha power between local and global conditions which I'm not sure if I can do in that case.

Is there a strategy you could recommend for choosing a baseline in this situation?

Thanks so much!

Raquel

[Mike X Cohen]

Hi Raquel. This is a good example of how the out-of-the-box baseline normalization is not always the best, or the easiest to interpret. I see two options here. 

One option is to use a single baseline that comprises all trials from all conditions (and all blocks). Any block-wise condition differences in the pre-stim period would show up as a condition-specific red or blue pre-stimulus blob, indicating relatively more or less power compared to the average of all conditions.

Another option is to do two sets of analyses -- one focusing on the post-stim dynamics and here you use a baseline normalization, and one in which you apply no baseline normalization and focus on pre-stimulus alpha. The difficult parts of interpreting non-normalized data are (1) comparing across frequencies (because of 1/f power scaling), and (2) separating tonic ongoing effects from phasic trial-related effects. In this case, you actually want to quantify the tonic (block-wise) effects. 

Hope that helps,

Mike

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.
To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.
Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
mikexcohen.com

[Raquel London]

Hi Mike,

Thank you very much for your answer. I have a few more, please let me know if I'm taking up too much of your time.

In the first option, would it be a good idea to choose a baseline that is as long as the pre-stimulus period I am interested in, up to -100?

And in the second option, for the post-stimulus analysis, would I then take a condition specific baseline?

In a previous analysis, instead of doing a baseline normalization, I got rid of the 1/f scaling like this:

Log-log transform the data, fit a linear function, create a vector from this function, subtract the vector from the loglog transformed data, inverse loglog transform the data. (I guess I could have used detrend, but I didn't know about this function then..)

Does this make sense to you or is it a bad idea? I thought in this way I would at least solve the first of the two problems you mentioned with non normalized data.

Thanks again,

Raquel

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimeseriesdata+unsub...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

[Mike X Cohen]

Yes, the baseline could be the entire pre-stimulus period, although you wouldn't want to include dynamics from the previous trial. Condition-average or condition-specific baseline would both work, but would tell you slightly different things (condition-average would focus more on block-wise tonic differences whereas condition-specific would focus more on phasic effects).

I'm generally not a big fan of log-transforming power and taking the residual of a linear fit. The problem is that the 1/f is not a neat logarithm, so the residual will likely look more awkward than the raw power. I would just use the raw power spectrum. Make sure to max-value-normalize the wavelet.

Mike

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
mikexcohen.com

--

You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

[Raquel London]

Hi Mike,

Thank you for your answer, I can see what you mean.

Regarding the max value wavelet normalization; we were just discussing this yesterday because in the code (from the book) we've seen up to now, sometimes this normalization is used and sometimes its not. We couldn't figure out when to use it and when not to.How do we decide?

Cheers,

Raquel

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimeseriesdata+unsub...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
mikexcohen.com

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimeseriesdata+unsub...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

[Mike X Cohen]

You can always normalize. There are a few threads on this list about it. Briefly:

cmw = <create your wavelet>

cmwX = fft(cmwX,nConv); % FFT of wavelet

cmwX = cmwX./max(cmwX); % max-value normalize in the frequency domain

% continue with convolution...

It's that final line that does the normalization. That will make sure the results are in microvolts (or whatever is the original scale of the data).

Mike

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
mikexcohen.com

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
mikexcohen.com

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

[Raquel London]

Hi Mike,

Thank you very much for your answer. Ok great, always normalize then :). 

cheers,

Raquel

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimeseriesdata+unsub...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
mikexcohen.com

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimeseriesdata+unsub...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
mikexcohen.com

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.

To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimeseriesdata+unsub...@googlegroups.com.

Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.