FAQ

Why should I use PsPM?

There are two alternatives to PsPM for SCR analysis: standard peak scoring, and Ledalab. All these three approaches aim at making a statement about an unobservable psychological process (sympathetic arousal), given skin conductance data. We prefer to use any method that has the highest chance of recovering this unobservable process. But how can we know how well a method recovers a hidden process? One possibility is to induce known psychological states, for example showing aversive and neutral images which we know induce strong and weak sympathetic arousal. Ideally, these two conditions should be separable by any analysis method. PsPM separates such two known conditions much better (significantly better) than the other two approaches (Bach DR, 2014, Biological Psychology 103:63-68: A head-to-head comparison of SCRalyze and Ledalab, two model-based methods for skin conductance analysis).

I have previously used SCRalyze. What’s the difference to PsPM?

PsPM incorporates the previous software package SCRalyze and offers all features of SCRalyze plus many more. If you started working on a project with SCRalyze and want to continue, you can still find previous software versions, help, and recources on http://scralyze.sourceforge.net.

What’s the best way of recording SCR?

It’s extremely simple - there is no mystery to it. Only a few dos and don’ts: Do use cup electrodes, don’t use dry plate electrodes (to minimise artefacts). Do use 0.5% NaCl gel, don’t use EEG or ECG gel (to avoid distorting responses). Do record on palm (e.g., thenar/hypothenar, or two fingers) or plant (inner arch). We have shown that it does not really matter where precisely you record (see Bach et al., 2010, “Modelling event-related skin conductance responses”). Make sure the subject does not move the limb you’re recording from. Do make sure there is no electromagnetic noise. Do use any kind of voltage coupler (e.g. based on a Wheatstone bridge, or differential amplifier). If you are using non-standard equipment and are unsure whether your responses conform to the canonical response function: use 10-20 simple stimuli (e.g. white noise bursts), separated by 30-40 seconds of rest, average responses, and check out how they look.

Can I import text data?

Yes you can. Text files must contain one data channel per column and no headers or annotations. Note that text import can be inconvenient because it involves very large files and might push matlab to its memory limits. We have been able in the past to provide import interfaces for different proprietary data formats, and if you give us details of your format, we might be able to help you import your data in a quicker and more comfortable way.

How can I import a matlab variable into PsPM?

You cannot import variables from the matlab workspace; you need to save them to a .mat file - one file per data set (e. g. per subject). This .mat file will have to contain one variable called “data”. This will be either a data points x columns matrix, or a cell array where each cell in the array is a one-column vector pertaining to one data channel. SCRalyze will not recognize time stamps as such. Rather, it will ask you for a sample rate. If the samples are not spaced evenly, let me know; we could then think of how to import the data otherwise.

How do I construct a regressor file for GLM?

Regressor files for GLM are similar to the ones used in SPM. Imagine a simple 2 x 2 factorial design. Each of the four cells needs to be specified separately. A regressor file might be created like this:

names = {'Condition A', 'Condition B', 'Condition C', 'Condition D'};
onsets = {[5 34 55], [9 42 48], [18 30 38], [14 23 27]};
save('Regressors.mat', 'names', 'onsets');

where e.g. there are three trials for each cell of the design, specified in seconds from file start. After estimating the GLM parameters, you have one summary statistic per cell of the design, which you can use to test your main effects and interaction on the second level (i.e. across the group). Don’t forget to save your file with e.g. the last line above.

How do I specify my regressors?

In my experiment, there are choices (positive and negative) with different payout levels and type etc., which will comprise 60 conditions. Does each type of choice have to be in a separate column? Should each column correspond to a condition (60 columns)? Typically, psychological experiments use factorial designs (i. e. designs in which each condition corresponds to a combination of factor levels, and all factors are fully crossed). In this case, each regressor should correspond to a cell (that is, a condition, i. e. a particular combination of factor levels) of your design. You can then specify all sorts of contrasts (i. e. main effects, interactions, polynomial contrasts etc.) using the contrast feature, and test them on the second level, across the group. In this case, however, there are 60 conditions. Therefore, it is worth thinking about an alternative. In fact, typical factorial designs make no assumption about any ordering of the factor levels. The levels of the independent variable are treated as categorical. In the present case, some factors - for example, payoff value - are clearly numerical. Here, one can greatly simplify the number of regressors. Take the payoff value. A main effect of payoff value in a full factorial design tests the omnibus H1: “There is some effect whatsoever of payoff level”. This includes effects such as “The 23rd payoff level produces greater SCR than the 2nd level, in the absence of any other differences between payoff level.” This might be completely uninteresting in the case of a quasi-continuous, numerical factor. You might be more interested in hypotheses of the kind: “There is a linear effect of payoff level on SCR”, “There is an inverted U-relation of payoff level and SCR” etc. These are polynomial contrasts of lower order. You could surely use 60 regressors and define such polynomial contrasts later. But you can also use them in your design matrix: for example, specify ALL events in one regressor. Then, use “parametric modulators” to encode your conditions. A categorical condition (for example, positive vs. negative) will just consist of -1 and 1, and for a numerical factor, you could specify for example, linear and quadratic effects. Note that if your design is not fully crossed, then the parametric modulators will be correlated. They get decorrelated during processing, and this means that the ordering of pmods might have an influence on your results.

When I trim my EDA data for GLM analyses, can I still use onset files where the onsets are relative to the beginning of the measurement, or do I have to trim the onsets as well?

The onset file must be relative to the data file you feed into analysis. This means, you need to trim the onset file as well. Actually, there are two ideas behind trimming. One is to get rid of artefacts and long stretches of data before and after the experiment. The other idea is that if you record GSR inside the scanner, you can trim the SCRalyze data files in the same way that you trim your fMRI data. Then you can use the same onset files for SPM and SCRalyze.

How do I specify episodes for SF?

There are several ways of specifying episodes for SF. The simplest is if each data file contains exactly one episode. Simply specify “whole” in the time unit menu. If the data files contain more than just one episode, you can specify episodes manually, or, more conveniently, use episode files. A simple file might look like this:

names = {'onsets', 'offsets'};
onsets = {[10 50 100], [40 80 160]};
save('Regressors.mat', 'names', 'onsets');

This file contains three episodes, one from 10-40 seconds, another from 50-80 seconds, and a third one from 100-160 seconds. Don’t forget to save your file with e.g. the last line above. The same episodes can be specified with

epochs = [10 40; 50 80; 100 160];

Again, don’t forget to save.

I have calculated contrasts. What do the numbers mean?

The meaning of contrast estimates depends on the basis function you use. For the canonical SCRF provided in the software, contrast estimates correspond to the peak amplitude of an ideal SCR. For example, if you compare two conditions, and the contrast estimate is 1, then this means that the average difference between the SCR peak amplitudes in the two conditions is 1 unit. If you converted your data into µS during import (which we always recommend), then a contrast estimate of 1 corresponds to a peak amplitude difference of 1 µS.

In what cases should I use a DCM, and when should I not?

Dynamic causal modelling for SCR was developed to analyse experiments where the timepoint and duration of a response are unknown and have to be estimated. This is a pretty general description, and there are many cases where either DCM or GLM can be used. However, if you only analyse SCR evoked by short stimulus - don’t use DCM. In this case, a GLM is sufficient and will be much more robust.

How can I deconvolve spontaneous fluctuation of skin conductance (after filtered SCL) in order to get the onsets and durations so that I can use them in SPM for modeling?

In principle, this is simple enough you just need to do DCM for SF and manually extract the onsets and amplitudes of the spontaneous fluctuations - durations are assumed to be fixed. There are a few technical issues to be considered: (1) all DCMs for SCR are susceptible to artefacts, in particular spikes with downward deflection which change the apparent signal baseline. The most difficult artefacts I have seen so far were induced by small subject movements in the static field. (2) DCM inversion might take a while if your scanner runs are long - several hours per inversion is not unusual, but this of course depends on your computation power. (3) If you model the extracted onsets as stick functions, modulate these by amplitude of SF, and convolve with an HRF, the result will look pretty similar to the original (filtered) SCR data. This is because HRF and SCRF are somewhat similar. Of course, there all all sorts of other things one might do: for example, ignore SF amplitude (because SF amplitude does not depend on tonic arousal e. g. during public speaking anticipation, while the number of SF does), or only analyse SF in certain time windows, etc. This is how you do it: define episodes (e. g. scanner runs), and run DCM for SF (see SCRalyze manual). Load the resulting SF file and extract [sf.dcm.stat.t] (onset time from episode start in seconds) and [sf.dcm.stat.a] (amplitude in your units). Note that this type of DCM does not estimate the number of responses - it assumes a fixed number and estimates their amplitude. The amplitude of “unnecessary” SFs will be estimated close to zero. Therefore it is a good idea to exclude SF below a certain amplitude. A sensible treshold is 0.1 mcS (see Bach, Daunizeau et al., Psychophysiology 2011 for details).