When doing a behavioural PLS, is it possible to have multiple behaviours in a single condition per subject?
In my experiment, subjects placed bids on each food item they saw (ie. 35 bids per run per subject). I'd like to include all of these bids in my behavioural PLS analysis, and tie them to the onset of each food item. Is that possible, and if so, how can I do it? Based on what I can take away from the tutorial, it looks like you enter you can enter only one bid per subject. I am not sure if my understanding of this is correct. The section in the PLS manual that specifies how to enter the behaviour is copied and pasted below.
Thanks!
-----------------------------------------------
For example, let's say you have group1 and group2 with condition1, and condition2. In group1, you have 2 subjects, and in group2 you have 3 subjects. Here is the order of the behavior file:
group1 condition1 subject1
group1 condition1 subject2
group1 condition2 subject1
group1 condition2 subject2
group2 condition1 subject1
group2 condition1 subject2
group2 condition1 subject3
group2 condition2 subject1
group2 condition2 subject2
group2 condition2 subject3
If I understood correctly, you would like to enter multiple behaviors for each condition per subject like the one listed below, is it correct?
=====================================================================
Label Behaviors
========================== ============================
group1 condition1 subject1 bid1 bid2 bid3 ... bid35
group1 condition1 subject2 bid1 bid2 bid3 ... bid35
group1 condition2 subject1 bid1 bid2 bid3 ... bid35
group1 condition2 subject2 bid1 bid2 bid3 ... bid35
group2 condition1 subject1 bid1 bid2 bid3 ... bid35
group2 condition1 subject2 bid1 bid2 bid3 ... bid35
group2 condition1 subject3 bid1 bid2 bid3 ... bid35
group2 condition2 subject1 bid1 bid2 bid3 ... bid35
group2 condition2 subject2 bid1 bid2 bid3 ... bid35
group2 condition2 subject3 bid1 bid2 bid3 ... bid35
Hi Jimmy,
Yes, that's correct. So if I upload one large file with all of the bids (for all subjects) in it, that's all I need to do then?
Thanks!
Deb
35 bids per run per subject). I'd like to include all of these bids in my behavioural PLS analysis, and tie them to the onset of each food item
Deb..
I think that the design Jimmy shows would allow you to determine which of the 35 bids in each condition is most strongly correlated with the AVERAGE response in the condition -
If you want to look at Condition1_Bid1_Food1 separately from Condition1_Bid35_Food35, you would need to turn off 'average across runs/trials' when you make the datamat - and be concerned that you have sufficient numbers of subjects to look at the data at this single trial correlation level..
Let us know if that is what you meant..
cheers
Nancy
35 bids per run per subject). I'd like to include all of these bids in my behavioural PLS analysis, and tie them to the onset of each food item
Deb..
I think that the design Jimmy shows would allow you to determine which of the 35 bids in each condition is most strongly correlated with the AVERAGE response in the condition -
If you want to look at Condition1_Bid1_Food1 separately from Condition1_Bid35_Food35, you would need to turn off 'average across runs/trials' when you make the datamat - and be concerned that you have sufficient numbers of subjects to look at the data at this single trial correlation level..
Let us know if that is what you meant..
cheers
Nancy
Hi Nancy,
The goal of my analysis is to look at how changing bid effects brain response, so it would be important to look at the differences between bid1_food1 and bid35_food35, as you suggested in your last message.
In terms of data, I have:
29 subjects * 35 bids per run per subject * 6 runs per subject = 6090 data points
Hopefully that will be enough data.
So in addition to uploading the file with all my datapoints, I'll need to turn off average across runs/trials. Is there anything else I need to do?
Thanks!
Deb
are all of the runs the same? - i.e., you have n=6 observations of condition1_bid1?
Hmm... I'm not sure what you mean by are all the runs the same?
It is an event related study and food pictures are presented in pseudo random order. The runs are the same in that subjects will always see 35 food items and make 35 bids, but they are also different in that the order of the stimuli they will see is different. Timing will also be different because they see the next food item only after they enter their bid for the previous stimuli.
Hope that clears things up!
I need to continue my question of "If I understood correctly, you would like to enter multiple behaviors for each condition per subject like the one listed below, is it correct?" here.
Since the answer is "yes", theoretically, the large file with all the bids under this "row order" can be loaded after you click "Load Behavior Data" button. However, in reality, you can't do this!
Assuming you only have 2 groups and 2 conditions (from the context above), you will end up with 140 rows. Depending on the brain size and temporal window size, the dataset could easily beyond the capability of your computer. Let's assuming that you have temporal window size of 8, and your volume size is [91 109 91], and your brain size could be 91*109*91/2 = 451314. Then, 140*8*451314 = 505471680. Assuming you are using single datatype (4 bytes), The total size of your dataset could be end up to 140*8*451314*4 = 2021886720, i.e. 2GB. Plus, the accessory data etc, I am sure that it will go beyond your computer (or even MATLAB) capability.
I think that the above design need to be reconsidered. One suggestion is that you enter 2~3 bids each time. i.e.:
============================================================
Label Behaviors
========================== ===================
group1 condition1 subject1 bid1 bid10 bid35
group1 condition1 subject2 bid1 bid10 bid35
......
Frankly speaking, I am an entrylevel application developer here, and have only very limited knowledge about the PLS design. If you encounter problem like "application crash", "how to use", etc. Please feel free to ask me. Other theoretical questions, or any question related to PLS design, I don't have much answer, and people with more PLS experience will provide you appropriate answer here.
I'm going to echo Jimmy on the design issue: if you have 35 bid/food pairings *6 runs, with no way to distinguish either bids or food (perhaps you do and weren't being clear), then you have a single vector of bid values and a matrix reflecting food onsets *#TRs, with no way to distinguish among them -- it's not clear at this point what your hypothesis/question is.
At a minimum - you should collapse the brain data across bid type (e.g., hi/med/low; ignoring the specific food presented). You could still use each subject's individual value for the three categories.
Or you could generate "mean bid value" for specific food types (e.g., fruits, veggies)
hope this helps
cheers
nancy
Thank you Jimmy and Nancy for your help! My apologies for the delay in responding -- I was waiting to hear back from my supervisor before writing more.
At any rate, colleagues of ours carried out the same analysis, so I'm sure it's possible, but perhaps I am not explaining it properly. For each subject, I want to input the following information:
subject1 bid1 bid2 bid3 ... bid35
subject2 bid1 bid2 bid3 ... bid35
...
Based on my previous assumption, if you have 6 runs and do "Merge Data Within Each Run" (i.e. turn off 'average across runs/trials'), the size of your dataset will exceed 6*2GB (12GB). Unless your colleagues who carried out the same analysis was using a very small raw data or they have huge amount of memory in their computer, running a dataset of 12GB is very challenging for most people. In order to enter all 35 bids in PLS without running out of memory, one possible solution could be resampling your raw data and making it very small. Tools "reslice_nii.m" & "clip_nii.m" can be used for this solution. The usage can be found under: http://research.baycrest.org/~jimmy/NIfTI
Baycrest is an academic health sciences centre fully affiliated with the University of Toronto
Privacy Statement - Disclaimer - © 1989-2024 BAYCREST HEALTH SCIENCE. ALL RIGHTS RESERVED