Do you know papers that discuss this subject?
We are looking for objective methods to decide about inclusion/exclusion of data sets.
For example: Article Bootstrap Signal-to-Noise Confidence Intervals: An Objective...
We are doing a study that looks at the detection rate of an effect (N400) on a single subject basis.
We have some very noisy data sets and are wondering where to draw the line for inclusion.
What criteria/methods do you suggest?
- Sufficient number of trials left per condition after preprocessing
- Sufficient SNR
- ...?
We are looking forward to your replies!
Thank you.
Dear Anna
It actually strongly depends on what you want to do with your data.
If you just doing a simple amplitude/latency assessment you don't need that high signal quality.
If you want to perform source analysis/time-frequency you need much cleaner data (high SNR). Note that the SNR is defined as a baseline to signal of interest ratio (not the overall quality)
Finally in the case of single single-trial analysis you definitely need clean data
On the other hand, what problem do you have with your data?
Is it high noise? - maybe it can be filtered out
is it an artifact (EKG, blinks)? - that can be easily reduced
Dear Anna
It actually strongly depends on what you want to do with your data.
If you just doing a simple amplitude/latency assessment you don't need that high signal quality.
If you want to perform source analysis/time-frequency you need much cleaner data (high SNR). Note that the SNR is defined as a baseline to signal of interest ratio (not the overall quality)
Finally in the case of single single-trial analysis you definitely need clean data
On the other hand, what problem do you have with your data?
Is it high noise? - maybe it can be filtered out
is it an artifact (EKG, blinks)? - that can be easily reduced
Hi Anna,
You can first try with some pre-processing steps, like applying appropriate filters, removal of line noise, eye blinks, ECG, dc removal, etc., and then if it still seems to be noisy, it might not be good enough to draw correct interpretations from it.
Also, in the case of ERP analysis, it is important that you have more number of trials, that would help you to increase SNR as well.
Hope it helps!
- Badges
- Science topic