Grand-averaging regressed ERPs against different predictors

Support & Discussions
,
1
  • MNE-Python version: 0.21.2
  • operating system: macOS 11.1

I am working on regressed ERPs. As far as I can understand, the functions linear_regression_raw (for raw data) and linear_regression (for epoched data) calculate regressed ERPs against a given set of regressors and store coefficients (betas, t’s and p’s) for each predictor in a complex, dictionary-like object. For example, the dictionary lm has the predictor s_freq as key and a series of Evoked objects as values:

lm = {'s_freq': lm(beta=<Evoked | '' (average, N=1), [-0.1, 0.8] sec, 32 ch, ~144 kB>, stderr=<Evoked | '' (average, N=1), [-0.1, 0.8] sec, 32 ch, ~144 kB>, t_val=<Evoked | '' (average, N=1), [-0.1, 0.8] sec, 32 ch, ~144 kB>, p_val=<Evoked | '' (average, N=1), [-0.1, 0.8] sec, 32 ch, ~144 kB>, mlog10_p_val=<Evoked | '' (average, N=1), [-0.1, 0.8] sec, 32 ch, ~144 kB>)}

I am trying to understand what is the best way to store the data for each regressor, coefficient participant, channel, and time, so to calculate grand averages for each regressor and coefficient. What I have trouble with is the inner structure of the lm object above (i.e., the structure within a given key of the dictionary). Would anyone be able to help?

Thanks!

2

As a follow-up – I think what I am really looking for is how to extract the values for each coefficient of each predictor. In the lm object above, for example, the key s_freq maps to a complex value that is a mne.stats.regression.lm object.

type(lm['s_freq'])
Out[112]: mne.stats.regression.lm

This object can be further broken down into different parts (which are the coefficients). The beta coefficients may be for example extracted as follows, which is an EvokedArray object.

lm['s_freq'].beta
type(lm['s_freq'].beta)
Out[113]: mne.evoked.EvokedArray

Am I understanding this well? If so, how could I extra with EvokedArray object of each coefficient for each predictor so that I can stored them in separate dictionaries?

Thanks!

3

did you have a look at the examples in

https://mne.tools/dev/generated/mne.stats.linear_regression.html#mne.stats.linear_regression

?

you can click on variables to know what their type is and on the source button to read the source code if it’s not clear

HTH
Alex

1 Like
4

Yes, I think I got it now. Thanks!