Participants performed five, 30-second blocks of a task. I plan to analyze these data using a GLM approach with a Gaussian basis function with parameters 1.0 1.0 and a tRange of -2.0 30.

My understanding is that I would receive 32 beta values per channel per participant (as well as group averaged beta values for each channel).

First, I am just looking to confirm that what is visualized in Homer after running a GLM are the model-predicted values.

Second, I have seen people take the model predicted values and, for example, collapse these across the task block for each channel to get a single number/channel that is interpreted to represent underlying cortical activity. I have also seen people take a similar approach but using the beta values as opposed to (what I assume are) the model-predicted values.

I know that these represent slightly different things, though I suspect that the final values would correlate quite well.

Are there advantages to choosing one approach over the other?

Hi Cory, I am also interested in this question - did you happen to figure any of that out since this post?

First, I am just looking to confirm that what is visualized in Homer after running a GLM are the model-predicted values.

Yes, when you visualize the HRF time course in Homer you are seeing the beta values times their associated temporal basis functions

Second, I have seen people take the model predicted values and, for example, collapse these across the task block for each channel to get a single number/channel that is interpreted to represent underlying cortical activity. I have also seen people take a similar approach but using the beta values as opposed to (what I assume are) the model-predicted values.

I don't know what you mean by beta values versus model-predicted values. In any case, when using the consecutive Gaussian model for the HRF, we generally average the HRF (i.e. the beta values convolved with their associated temporal basis functions) over the time range of interest to get a single value that we then use in statistical testing.