I am trying to create a timing file for a non-linear scr model. We have 25 trials (5 of each condition/trial type) and each trial is 13 seconds followed by an ISI of 6-9 seconds (pessoa timings were used because it is a fMRI paradigm). I am defining the cell array "events" with 5 cells (one for each condition/trial type) and each cell includes a vector of onset times for each trial within that condition/trial type. But everytime I try to run the model, I get the following "Warning: Error in event definition, Either events are outside the file or trials overlap." I am defining the array as follows:
Because the non-linear model processes each trial individually there is no need to specify trial types before inverting the model. In other words, the DCM approach is agnostic about the trial type it's dealing with.
Thus, the events should be specified in one column, which would result in this.
If you specify the events like this, the non-linear model will run through and will create a file (as specified in model filename) containing the variable dcm.stats with information about the amplitude of each individual trial. You then have to perform your contrasts on this column, according to the trial types/category.
There still might be slight differences in the format of the timing file depending on the design of your experiment. In particular, defining the timings as a one-column vector would make PsPM assume that it's dealing with evoked, fixed-latency responses, while defining a window for each trial (with a two-column onset/offset vector) implies the search for an anticipatory response, which can also vary in peak latency. In order to help you with this, we would need to know some additional information about the study design.
Hope this helps,
Best regards
Tobias
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thank you so much for your quick reply! This is very useful information! When we deisgned the paradigm, we were assuming that it would be an evoked response experiement. Participants are shown an emotional (or neutral) picture and asked to perform an emotion regulation technique to change their emotional reaction. There is a possibility that the task is working as more of a cognitive load type task, rather than an emotional response task, because we are seeing a GSR spike at the very end of the ISI (which is about 6-9 seconds long), meaning the spike is occuring about 6 seconds after the picture is removed and the trial is over. Some participants, however, show a GSR spike 5-6 seconds after the picture initially appears. So we were thinking of running some initial analyses using both methods to see if we can make sense of anything. If you have any thoughts, we would certainly love the feedback.
Stephanie
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Your idea of comparing these two different hypotheses sounds interesting. If your evoked hypothesis does not work you could also try with an anticipatory framework. You then specify as a second events column a window wherein the response may occur. PsPM then tries to find in a data driven way the best fit per trial and participant. From this you will then obtain a matrix with three columns: amplitude (as before), latency and dispersion. Where each trial and participant has its own setting. Thereof, you can use the mean latency across trials to test whether you'll find a difference between the two groups.
Please, do not hesitate to contact me again for further question or comments.
Best regards
Tobias
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
We are playing around with the different analytic methods for the project I described in my message above. In addition to our event related data, we also have a 7 min "rest" period that we collect at the end of the MRI scan (participants just stare at a fixation corss while we collect functional resting state data). We applied the spontaneous fluctuation method to these data and I exported all 4 parameters (SCL, AUC, DCM and MP). I should mention that we broke the 7 minutes into 7 one minutes epochs. The SCL and AUC variables appear to provide some excellent data, but we are confused by the DCM and MP outputs. Most of the data points are zero. We also applied the SF technique to our event related data, just for fun, and are getting the same pattern of results. We aren't quite sure how to interpret these findings. What does it mean to have a value of zero for the DCM and MP parameters? Are we doing something wrong? Or is this normal? If it's normal, what does it mean for those participants that have non-zero values.
Thank you!!
Stephanie
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
could data scaling be an issue here? DCM & MP use thresholds to separate SF from noise (default 0.1 mcS), so if your data is in not in mcS but in arbitrary (or different) units, this will yield incorrect results. You can fix the units during import, or change thre threshold in the options of the SF model.
Hope this helps
Best
Dominik
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
We are pretty sure that our data is in mcS (as this is the labe for the values displayed during data collection along the y-axis of the waveform plot). Dr. Stevens has never used this in his research before, nor have any of the other investigators at our center. We are pretty sure that it is set up correctly, but we also aren't very confident in that because we have seen some participants whose GSR profiles are flat lines during data collection (i.e. no fluctuations or responses to IAPS emotionally charged pictures). What else could a zero value for a majority of DCM and MP data points indicate? There is a chance that I did something wrong when I set up the analyses, but we aren't familiar enough with measuring and analyzing EDA enough to know what it could possibly be.
Also, a quick basic question, what are the unit values for the amplitude parameter generated using the non-linear fixed and felxible events models? We were assuming they are mcS, but wanted to be sure because we have some very low values for this variable as well (0.0001 - 0.9).
Thank you in advance!
Stephanie
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
if you want I'd be happy to look into the discrepancy between DCM/MP and SCL/AUC - just send me a results file generated by PsPM.
Regarding the amplitudes, the units are "eSCR units". For eSCR, this is just mcS. For aSCR, this means that a neural input with the same amplitude, fed into the eSCR model, would have caused an eSCR with 1 mcS amplitude.
This sounds a bit complicated. The reason is that aSCR amplitude depends not only on the amplitude, but also on the duration of the neural input. The model separates the two. Hence, to specify the neural impulse amplitude in terms of ensuing SCR, one must assume a certain constant duration of the neural impulse - this is then just an eSCR.
Hope this makes sense.
Best
Dominik
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thank you for that explanation! I don't think we completely follow, but ultimately it sounds like the units for the amplitude values generated by the non-linear models are in mcS? After noticing a mistake with my script, I have rerun the analyses and some participants have response values as low as 0.00001 and some have much higher values (around 0.6 - 0.9). Assuming these are mcS values 1) does this pattern of results sound problematic? and 2) If they do sound valid, is it fair to use the same 0.1 mcS threshold that you use for peak detection in SF analyses as a benchmark for determining reliable response versus no response to the stimuli?
Ultimately we are just trying to make sure our data makes sense. We have never used any sort of measure of electrodermal activity before, and are concerned about the quality of the signal we are recording. Some participants look like they are not responding at all, while others are responding quite a bit to emotionally charged pictures. So we are just trying to make sure that our results are passing sanity checks (to be sure we are using your toolbox correctly and that are data is ultimately reliable).
And as for the DCM/MP and SCL/AUC, I will send you a file with some examples of our PsPM generated results over the weekend.
Thank you again!
Stephanie
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I think it is ok to say that aSCR units are in mcS (if your data are in mcS).
The large heterogeneity in responding is often observed. However, it would be good to make sure it is not a technical artefact. There is also a bug in some earlier SCRalyze versions if you don't do z-transforming. I can help you fix the values in case you used one of the affected versions https://sourceforge.net/p/scralyze/discussion/1136238/thread/3842e3e9/
What we often do is that we check that people have a reliable response to the US. However, I would use averaging to do that. The fact that the response amplitude is very small does not mean it is not there. For SF, we have to separate responses within one subject - this is why a threshold is necessary (and we haven't come up with a way of making it subject-specific).
I guess some heterogeneity is expected and relates to psychological processes rather than to SCR generation. I would tend to see this heterogeneity as part of the (psychological) picture.
Dominik
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I am trying to create a timing file for a non-linear scr model. We have 25 trials (5 of each condition/trial type) and each trial is 13 seconds followed by an ISI of 6-9 seconds (pessoa timings were used because it is a fMRI paradigm). I am defining the cell array "events" with 5 cells (one for each condition/trial type) and each cell includes a vector of onset times for each trial within that condition/trial type. But everytime I try to run the model, I get the following "Warning: Error in event definition, Either events are outside the file or trials overlap." I am defining the array as follows:
events = {[0; 182.766; 320.709; 362.073; 462.479] [22; 102.057; 160.766; 383.031; 442.153] [41.483; 82.311; 202.434; 281.514; 423.077] [60.562; 121.877; 221.633; 260.967; 403.424] [140.949; 241.371; 301.219; 340.847; 481.800]}
save ('timingfile', 'events')
I then load the generated 'timingfile.mat' file into the batch editor as my timings file.
I very well could be defining the cell array incorrectly, but I was wondering if you had any suggestions as to why I keep receiving the warning.
Thank you!
Stephanie
Hi Stephanie
Because the non-linear model processes each trial individually there is no need to specify trial types before inverting the model. In other words, the DCM approach is agnostic about the trial type it's dealing with.
Thus, the events should be specified in one column, which would result in this.
events = {[0; 22.0000; 41.4830; 60.5620; 82.3110; 102.0570; 121.8770; 140.9490; 160.7660; 182.7660; 202.4340; 221.6330; 241.3710; 260.9670; 281.5140; 301.2190; 320.7090; 340.8470; 362.0730; 383.0310; 403.4240; 423.0770; 442.1530; 462.4790; 481.8000]}
If you specify the events like this, the non-linear model will run through and will create a file (as specified in model filename) containing the variable dcm.stats with information about the amplitude of each individual trial. You then have to perform your contrasts on this column, according to the trial types/category.
There still might be slight differences in the format of the timing file depending on the design of your experiment. In particular, defining the timings as a one-column vector would make PsPM assume that it's dealing with evoked, fixed-latency responses, while defining a window for each trial (with a two-column onset/offset vector) implies the search for an anticipatory response, which can also vary in peak latency. In order to help you with this, we would need to know some additional information about the study design.
Hope this helps,
Best regards
Tobias
Hi Tobias,
Thank you so much for your quick reply! This is very useful information! When we deisgned the paradigm, we were assuming that it would be an evoked response experiement. Participants are shown an emotional (or neutral) picture and asked to perform an emotion regulation technique to change their emotional reaction. There is a possibility that the task is working as more of a cognitive load type task, rather than an emotional response task, because we are seeing a GSR spike at the very end of the ISI (which is about 6-9 seconds long), meaning the spike is occuring about 6 seconds after the picture is removed and the trial is over. Some participants, however, show a GSR spike 5-6 seconds after the picture initially appears. So we were thinking of running some initial analyses using both methods to see if we can make sense of anything. If you have any thoughts, we would certainly love the feedback.
Stephanie
Hi Stephanie,
Your idea of comparing these two different hypotheses sounds interesting. If your evoked hypothesis does not work you could also try with an anticipatory framework. You then specify as a second events column a window wherein the response may occur. PsPM then tries to find in a data driven way the best fit per trial and participant. From this you will then obtain a matrix with three columns: amplitude (as before), latency and dispersion. Where each trial and participant has its own setting. Thereof, you can use the mean latency across trials to test whether you'll find a difference between the two groups.
Please, do not hesitate to contact me again for further question or comments.
Best regards
Tobias
Hi Tobias,
We are playing around with the different analytic methods for the project I described in my message above. In addition to our event related data, we also have a 7 min "rest" period that we collect at the end of the MRI scan (participants just stare at a fixation corss while we collect functional resting state data). We applied the spontaneous fluctuation method to these data and I exported all 4 parameters (SCL, AUC, DCM and MP). I should mention that we broke the 7 minutes into 7 one minutes epochs. The SCL and AUC variables appear to provide some excellent data, but we are confused by the DCM and MP outputs. Most of the data points are zero. We also applied the SF technique to our event related data, just for fun, and are getting the same pattern of results. We aren't quite sure how to interpret these findings. What does it mean to have a value of zero for the DCM and MP parameters? Are we doing something wrong? Or is this normal? If it's normal, what does it mean for those participants that have non-zero values.
Thank you!!
Stephanie
Hi Stephanie
could data scaling be an issue here? DCM & MP use thresholds to separate SF from noise (default 0.1 mcS), so if your data is in not in mcS but in arbitrary (or different) units, this will yield incorrect results. You can fix the units during import, or change thre threshold in the options of the SF model.
Hope this helps
Best
Dominik
Hi Dominik,
We are pretty sure that our data is in mcS (as this is the labe for the values displayed during data collection along the y-axis of the waveform plot). Dr. Stevens has never used this in his research before, nor have any of the other investigators at our center. We are pretty sure that it is set up correctly, but we also aren't very confident in that because we have seen some participants whose GSR profiles are flat lines during data collection (i.e. no fluctuations or responses to IAPS emotionally charged pictures). What else could a zero value for a majority of DCM and MP data points indicate? There is a chance that I did something wrong when I set up the analyses, but we aren't familiar enough with measuring and analyzing EDA enough to know what it could possibly be.
Also, a quick basic question, what are the unit values for the amplitude parameter generated using the non-linear fixed and felxible events models? We were assuming they are mcS, but wanted to be sure because we have some very low values for this variable as well (0.0001 - 0.9).
Thank you in advance!
Stephanie
Hi Stephanie
if you want I'd be happy to look into the discrepancy between DCM/MP and SCL/AUC - just send me a results file generated by PsPM.
Regarding the amplitudes, the units are "eSCR units". For eSCR, this is just mcS. For aSCR, this means that a neural input with the same amplitude, fed into the eSCR model, would have caused an eSCR with 1 mcS amplitude.
This sounds a bit complicated. The reason is that aSCR amplitude depends not only on the amplitude, but also on the duration of the neural input. The model separates the two. Hence, to specify the neural impulse amplitude in terms of ensuing SCR, one must assume a certain constant duration of the neural impulse - this is then just an eSCR.
Hope this makes sense.
Best
Dominik
Thank you for that explanation! I don't think we completely follow, but ultimately it sounds like the units for the amplitude values generated by the non-linear models are in mcS? After noticing a mistake with my script, I have rerun the analyses and some participants have response values as low as 0.00001 and some have much higher values (around 0.6 - 0.9). Assuming these are mcS values 1) does this pattern of results sound problematic? and 2) If they do sound valid, is it fair to use the same 0.1 mcS threshold that you use for peak detection in SF analyses as a benchmark for determining reliable response versus no response to the stimuli?
Ultimately we are just trying to make sure our data makes sense. We have never used any sort of measure of electrodermal activity before, and are concerned about the quality of the signal we are recording. Some participants look like they are not responding at all, while others are responding quite a bit to emotionally charged pictures. So we are just trying to make sure that our results are passing sanity checks (to be sure we are using your toolbox correctly and that are data is ultimately reliable).
And as for the DCM/MP and SCL/AUC, I will send you a file with some examples of our PsPM generated results over the weekend.
Thank you again!
Stephanie
Hi Stephanie
I think it is ok to say that aSCR units are in mcS (if your data are in mcS).
The large heterogeneity in responding is often observed. However, it would be good to make sure it is not a technical artefact. There is also a bug in some earlier SCRalyze versions if you don't do z-transforming. I can help you fix the values in case you used one of the affected versions
https://sourceforge.net/p/scralyze/discussion/1136238/thread/3842e3e9/
What we often do is that we check that people have a reliable response to the US. However, I would use averaging to do that. The fact that the response amplitude is very small does not mean it is not there. For SF, we have to separate responses within one subject - this is why a threshold is necessary (and we haven't come up with a way of making it subject-specific).
I guess some heterogeneity is expected and relates to psychological processes rather than to SCR generation. I would tend to see this heterogeneity as part of the (psychological) picture.
Dominik