Scenario 1: Let's assume you acquired a dataset referenced to A1 and A1 is part of your active electrode matrix. Next you want to rereference to linked ears. A1, since present, will be flat and A2 will contain whatever activity was present relative to A1. If you specify both A1 and A2 for building of the new reference, you will create an A1/A2 averge which actually contains half of A2's activity as A1 is flat. However, the activity at A1 is determined by the mean of the activity of all electrodes in the matrix, since all were measured referenced to A1. So in order to get the real A1 back, Analyzer calculates the true reference voltage (with implicit checked) and injects this voltage instead of A1. But beware, DO NOT specify A1 when building the new reference, because the implicit term ALREADY IS A1! If you do specify A1, A2 AND implicit you will get (A1+A2)/3.

Scenario 2: You acquired in the same fashion referenced to A1, but did not include A1 in your active electrodes matrix. The procedure is the same as above, but your customers will ask HOW to rereference to linked ears. The answer, again, is to specify A2 for the new reference AND to include the implicit, which in fact, is A1.

Scenario 3: You recorded referenced to A1, have A1 in your matrix and want to go to AvgRef. You then specify all electrodes except bipolar ones and except A1 for the new reference, say include implicit in order to get A1 into the reference and apply this reference to all electrodes including A1 but without the bipolar ones. This will get you an AvgRef, just like what you have from the start with ActiveTwo.

So here goes ActiveTwo:

You already have an Average reference, so all you have to do is to specify your new target electrode(s) as the new reference AND include them in the list of electrodes that the new reference applies to. Here, you DO NOT check implicit, because you do not want for the new reference to be the average of your average reference plus your target, but instead you want for the new reference to be only the electrode(s) you specified.

Message: Whenever you want to explicitly use "that which you can not see" as a virtual electrode into the building of your reference, check the box "Include...". Do not check the box AND include the same electrode that served as the reference for data acquisition into the new reference if you have the "Include..." option checked.

If you have data that was recorded based on a bipolar pair, but that resulted in just one trace, the settings are different.  You actually have a bipolar trace, but only ONE such trace and not two individual traces that together would make up the bipolar signal as above.  In this case you would have to choose the option "Common reference", because to Analyzer there is no dedicated reference channel available for your VEOG and/or HEOG trace which means that to Analyzer this is the same as using the "Common reference" of all other (unipolar) channels.

Now, you will have seen that Analyzer tries to be intelligent (~) about naming history nodes with conflicting names. Creating a second average with different or the same parameters based on the exact same data would lead to two nodes called "Average", so Analyzer goes and names the second one "Average 2", etc.

However, if the averages are computed in different branches of your history file, Analyzer sees no problem in these Averages all having the name "Average" and does not change any names.

This in turn can easily lead to a situation where you are expecting Analyzer to find a certain node called "Average" for instance in the third branch, but there also is one such node in the first branch. This node in the first branch is the one that will be found by Analyzer, because it resides higher up in the branching-hierarchy.

In other words, the first thing you should try to do would be to rename any other occurrences of your target node names in one of the files that fail. If it ceases to do so, you will know you are on the right track.

In other words, if you have an S1, S2, S3 and S4 condition in your data, you would do common preprocessing steps (Ocular Correction, Baseline, Filtering, Detrending...) for all and then you would segment your data into separate data nodes for your S1, S2, S3 and S4 conditions, followed by individual averaging.

b) There are two parts to this. First you would select the S1 trigger as the one to segment by (move it from left to right listbox). This is important, because whatever you write in the ABE field is tested against the primary condition (S1 is present!) you just set. Then you enter the following in the ABE field: "NOT S2(-MaxTime,0)" with "MaxTime" being the time limit for the presence of S2 stimuli before S1 stimuli (pls note the minus sign).

The other, more complex condition would translate to "(S1(0,0) and S7(0,MaxTime)) or (S2(0,0) and S8(0,MaxTime))". For this condition, you would have to select both S1 and S2 in the upper listbox windows of the Segmentation transformation, else none of the intended matches would be found.

You may wonder why you have to select S1 and S2 in the top windows AND have to reference them in the ABE. This is because your condition is NOT "give me any S7 or S8 following any S1 or S2", but instead the S7 is directly linked to the presence of an S1 and the S8 ist equally linked to the presence of your S2. This is why you have to include them in the ABE also.

The reason for this obviously is that different n's will definitely lead to different signal-to-noise ratios between the channels in your setup and thus will automatically lead to different standard deviations due to the increased amount of residual spontaneous EEG activity left in the resulting average for lower-n channels.

The reason this option was included in BrainVision Analyzer is because we have a lot of customers working with clinical populations such as schizophrenics, that are very difficult to work with, and tend to produce enormous amounts of potential artifacts. For this type of population, our customers are glad for any trial/channel combination they can use for the average, regardless of whether or not this means making compromises regarding the validity of subsequent statistical comparisons. 

Thus, this option is definitely not recommended for "regular" science with "normal" subjects...