I’m interested in developing a block design experiment for a passive viewing task. 3 groups will undergo the task, and each subject will view the images under 3 different instructions (randomized over the course of the run): InstructionA, InstructionB, InstructionC. In each Instruction, there will be 3 cue types presented: CueA, CueB, CueC. Note that all mentioned "C"s serve as the control for the “A” and "B"s (i.e., GroupC is a control group, InstructionC is the control instruction, and CueC is the control cue). This is a pretty ambitious design, so I’m hoping you can provide some insight to maximize power and efficiency.
Each Cue block will comprise 4 images for 4s each (e.g., [CueA] below is a 16s block with 4 images presented 4s each). Blocks are separated by 10s null events:
InstructionA: [CueA] - null - [CueB] - null - [CueC] - null - [CueA] - null - [CueB] - null - [CueC] - null
InstructionB: [CueA] - null - [CueB] - null - [CueC] - null - [CueA] - null - [CueB] - null - [CueC] - null
InstructionC: [CueA] - null - [CueB] - null - [CueC] - null - [CueA] - null - [CueB] - null - [CueC] - null
Instructions and Cue orders would be randomized. This puts the run at ~8 minutes.
The main contrasts of interest would be:
InstructionA_CueA > InstructionB_CueA
InstructionA_CueB > InstructionB_CueB,
InstructionC_CueA > InstructionC_CueC,
InstructionC_CueB > InstructionC_CueC
The issue here is that I would only have 2 blocks sampled per Cue in each Instruction. Whereas if this was an event related design, I could model each image within a block as a single event, and the 2 CueA blocks above would be 8 events. I’m unsure if this is an actual issue, or if I’m panicking at the thought of seeing a timing file with just 2 rows!
I could increase the number of blocks presented, but then the scan run length may be too long for subjects to endure. The other issue is that if I was to present 4 runs, I’d be nearing 32 minutes of scan time, and I need to keep this in the 24-26 minute range altogether.
How would you advise trimming the fat from this design without compromising the structure? Can I get away with shorter ITIs? Ditch this altogether and adopt an event-related design?
OR…If CueC is not integral for Instructions A and B, could I redesign this in a manner where InstructionA and InstructionB consist solely of CueA and CueB blocks, and only InstructionC includes all 3 Cue blocks? I could then present 3 CueA blocks and 3 CueB blocks each within InstructionA and InstructionB. Statistically, would this cause problems when, for instance, I make an InstructionA_CueA > InstructionC_CueA type of contrast becuase the former image was derived from 3 CueA blocks and the latter from 2 CueA blocks? This is assuming that the absence of CueC is not a psychological confound, of course.