Could someone give me some advice on plotting curves that are generated by analyses using TENT vs TENTzero? I have an analysis in which I use either TENT or TENTzero to model some responses to videos, then use 3dANOVA3 to test the group differences. While the statistical maps are more or less the same, the plotted graphs tell different stories. Specifically, when using TENT the graph of an ROI shows condition A starting lower than condition B (i.e., having a lower % change value at time 0) but reaching the same peak value, while when using TENTzero I see then starting at zero (obviously) but then condition A reaching a higher peak value. In both cases, then, the trough-to-peak value for A is greater than that for B, so it makes sense that the stats are the same. My question is whether there are quidelines for when to use TENT vs TENTzero. I feel as though if a reviewer questioned me on why I chose TENTzero over TENT I wouldn’t know how to answer.
it’s expected to have some anticipation effect (the subject might respond earlier than the real stimulus onset), or
there might have some inaccuracies or mistakes in stimulus onset time specifications.
On other hand, if you don’t see much effect at the stimulus onset time with TENT or if you don’t believe either of the two scenarios happened, it’s definitely more preferable to go with TENTzero: a more parsimonious model (occam’s razor).
Gang – didn’t you get TENTzero and TENT exactly reversed in the previous post!?
I think this is really what I’m trying to figure out. I can imagine why there might be anticipation-related differences between the two conditions–one condition is blocks of the first halves of videos and the other condition is blocks of corresponding second halves. So in a sense in the one condition you know you’ll be seeing something entirely new and in the second you’ll be seeing something that is somewhat familiar. But of course this is not what we are interested in measuring, though I realize we can’t just ignore it.
So when using TENTzero all anticipatory effects are ignored/neutralized, and I guess it really comes down to whether it is justifiable to ignore them (assuming it ever is), while TENT leaves them in for better or for worse. I can’t shake the feeling that TENT is a more fair reflection of what is actually happening in the brain.
But on this logic TENTzero is only justified when a) there are no pre-stimulus differences; or b) when these exist but they are explicitly discussed and identified in a paper. In the case of a) there really isn’t going to be a difference between TENT and TENTzero anyway, while in b) the use of TENTzero really depends on how justified you are in ignoring these differences (and that you state that they existed in the paper).
In the specific case of my data, I’m really struggling with whether the choice between these two ways of modelling the response significantly effects the interpretation of the results.
Perhaps for my purposes the best thing to do is model at least one TR pre-stimulus using TENT, but I’d really like to hear what you think about what I’ve written, and any other thoughts you might have. In the end, I just want my paper to be a faithful reflection of the effects we observed.
Perhaps for my purposes the best thing to do is model at least one TR pre-stimulus using TENT, but I’d really like to hear what you
think about what I’ve written, and any other thoughts you might have. In the end, I just want my paper to be a faithful reflection of the
effects we observed.
With TENT, you’ve already assumed that the BOLD response started earlier than the stimulus onset times.
Here is one thought - How about this empirical approach? Try shifting the estimated HDR of the first (or second) condition rightward (or leftward) along the time axis a little bit until the two curves roughly match at the upstroke phase, and then compare the two curves in terms of their shape differences. That way you would be comparing the two curves after eliminating the anticipation effect.
You may also try CSPLIN/CSPLINzero instead of TENT/TENTzero to get smoother curve comparisons.
thanks a lot for your response. I have a few questions.
I’m still confused about whether it is justified to ignore any pre-stimulus anticipatory effects, which is what TENTzero seems to do. And yet, at least in my case the stats come out the same, it’s really just a question of how the plotted responses look.
Could you explain to me, more generally, why TENTzero exists? That is, what was the motivation for adding it to AFNI in the first place? If I understood that, perhaps I could answer my own questions.
I’m still confused about whether it is justified to ignore any pre-stimulus anticipatory effects, which
is what TENTzero seems to do. And yet, at least in my case the stats come out the same, it’s really
just a question of how the plotted responses look.
If you believe that anticipation effect may exist, go with TENT. If TENT(0, …) is not good enough to catch longer anticipation, use a negative number such for the first parameter in TENT.
Could you explain to me, more generally, why TENTzero exists? That is, what was the motivation for
adding it to AFNI in the first place? If I understood that, perhaps I could answer my own questions.
If you don’t expect any anticipation (nor stimulus timing errors) for your trials, use TENTzero. In other words, TENTzero is basically like the shape-fixed approach with GAM or BLOCK in terms of stimulus onset times with the assumption that BOLD response starts at the moment when each stimulus begins.
thanks a lot Gang, that makes sense. it’s interesting though–we had no expectation of an anticipation effect (and it’s not that big), but it does seem to be there. good to have both options I guess.
James
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.