I am analysing a set of data using your PsPM toolbox. As stimuli we had pictures with various degrees of aversiveness (from pretty much neutral to very aversive) presented for three seconds each. The inter-trial interval varried between 7 and 12 seconds. We want to extract a trial-by-trial estimate of SCR.
According to your papers and from what I read here in the forum I believe the most appropriate analyses method in our case is DCM, am I right?
From visual inspection fo the data I notice that, as expected, there is a huge variability in the amplitude and frequency of SCR and that these are barely visible for some participants. Thus, I would like to ask you advice on how to deal with potential non-responders, or those participants that could be classified as such by the traditional analyses based on trial-by-trial manual inspection. How does the model deal with such data?
Related to this, because output values will differ considerably between participants would you advise any kind of transformation (log, z) before entering the trial-by-trial values in a mixed general linear model?
Many thanks for the software and support.
Kind regards,
Ruben
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
please apologise the late reply; I just came back from holidays.
Using DCM for this problem appears a sensible choice. Regarding "non-responders", you could just use the amplitude estimates to define them - they should be very small and just slightly above zero (they will never be exactly zero, though). Whether you want to exclude them is another question that can be answered independent from the model-based approach. Depending on the research question, excluding "non-responders" may be favourable or could on the other hand bias your results. For example, it is sometimes implicilty assumed that "non-responders" fail to emit a physiological response to a psychological process. But these people may just as well differ in some psychological process, such as how much they attend to the stimuli.
Regarding transformations, there are of course different approaches to this, and this depends on how you set up the model. For an LME, it may be wise to transform the data. The general problem is that peripheral skin properties may introduce a non-linear, non-multiplicative transformation of the sweat gland activity, and it is not clear how to best deal with this. We have in the past used z-transformation of amplitude estimates (see Staib et al. 2015), and another way is to divide by the average response in a baseline/control condition. Both have their pros and cons and can be motivated in terms of a biophysical model of the skin.
Hope this helps
Dominik
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
following on the non-responders issue, how do I get the amplitude estimates you are referring to?
Shall I look at the reconstructed response amplitude per condition which comes out of the first-level GLM?
Many thanks an best,
Ambra
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Dear Dominik and Tobias,
I am analysing a set of data using your PsPM toolbox. As stimuli we had pictures with various degrees of aversiveness (from pretty much neutral to very aversive) presented for three seconds each. The inter-trial interval varried between 7 and 12 seconds. We want to extract a trial-by-trial estimate of SCR.
According to your papers and from what I read here in the forum I believe the most appropriate analyses method in our case is DCM, am I right?
From visual inspection fo the data I notice that, as expected, there is a huge variability in the amplitude and frequency of SCR and that these are barely visible for some participants. Thus, I would like to ask you advice on how to deal with potential non-responders, or those participants that could be classified as such by the traditional analyses based on trial-by-trial manual inspection. How does the model deal with such data?
Related to this, because output values will differ considerably between participants would you advise any kind of transformation (log, z) before entering the trial-by-trial values in a mixed general linear model?
Many thanks for the software and support.
Kind regards,
Ruben
Hi Ruben
please apologise the late reply; I just came back from holidays.
Using DCM for this problem appears a sensible choice. Regarding "non-responders", you could just use the amplitude estimates to define them - they should be very small and just slightly above zero (they will never be exactly zero, though). Whether you want to exclude them is another question that can be answered independent from the model-based approach. Depending on the research question, excluding "non-responders" may be favourable or could on the other hand bias your results. For example, it is sometimes implicilty assumed that "non-responders" fail to emit a physiological response to a psychological process. But these people may just as well differ in some psychological process, such as how much they attend to the stimuli.
Regarding transformations, there are of course different approaches to this, and this depends on how you set up the model. For an LME, it may be wise to transform the data. The general problem is that peripheral skin properties may introduce a non-linear, non-multiplicative transformation of the sweat gland activity, and it is not clear how to best deal with this. We have in the past used z-transformation of amplitude estimates (see Staib et al. 2015), and another way is to divide by the average response in a baseline/control condition. Both have their pros and cons and can be motivated in terms of a biophysical model of the skin.
Hope this helps
Dominik
Hello,
following on the non-responders issue, how do I get the amplitude estimates you are referring to?
Shall I look at the reconstructed response amplitude per condition which comes out of the first-level GLM?
Many thanks an best,
Ambra
Hi Ambra
for GLM, the amplitude estimate is the reconstructed amplitude per condition.
For DCM (to which the thread referrred originally), it is one of the several estimates found in the stats - they are recognisable by their names.
Dominik