Randomization Fixed Seed Question

[Dahlia Kamel]

We are hoping to begin using MST in our study and are looking for some clarification. We have a longitudinal design with 3 separate timepoints. From what I have understood, we can randomize participants to either Set C or Set D and use randomization fixed seeds 1-2-3 and ensure no overlap in stimulus items. Is this correct or are there certain fixed randomization seeds that do contain overlap? We are planning to use the 40 item per condition setting. Thank you!

[Craig E.L. Stark]

Sets C and D use different stimuli, so using them, you could have 2 independent runs.  If you shift over to the Sets 1-6 (what we use now), you get 6 totally separate sets of stimuli.  Note, though, that the images in C&D will appear in 1-6 (along with images from E&F and others).  So, the stimulus set determines which actual images have a chance of being shown.

Where the seed comes in is in two aspects: 1) Which stimuli get assigned to which condition, and 2) What order the images are shown in.  So, 001a/b.jpg may be a rubber duck pair in Set 1 (or C or ...).  If you use seed 1, it might be that 001a.jpg is shown as the 5th study item and 001b shown as the 10th test item, here as a lure.  Change the seed and 001a.jpg might be shown as the 12th study item and 001a shown as the 8th test item, here as a repeat.  But, go back to that seed=1 and it'd replicate the first run perfectly.

*What we in the lab do*: Everybody gets a unique *numeric only ID*.  So, I might be 12000.  My first run through, I might get Set 3 and my second I might get Set 1.  But, that ID number is always used.  We also set the randomization to be based on the ID so that, for example, if I want to stick a day between study and test, all I need to do us enter in that ID number and have the randomization seed set to use the ID.  On the first day, run Phase 1 and on the second day run Phase 2.  You'll know that the same stimulus randomization happened so that if the study phase assumed 001a jpg was to be shown at study and repeated identically at test, that's what'll happen in the test phase.

[Background if that didn't make sense.  Computer random number generators aren't really random.  If you start with the same "seed", you'll take the same apparent random walk through numbers.  So, if I set seed=1, and asked for 3 random numbers between 1-100 I might get: 42, 19, 87.  Seed=2 might give: 93, 12, 19.  Set seed=1 and ask again and you'd get: 42, 19, 87.  *That's* why, if we can control the seed, we can replicate exactly how randomization happened.  FWIW, if you want things to appear to be nicely random and not always do the same thing the first time you run your code, you just do something like set the seed to be the number of microseconds the computer has been on.  You won't replicate that any time soon.]

Craig

Craig Stark, Ph.D.

James L. McGaugh Chair in Neurobiology of Learning and Memory
Professor, Department of Neurobiology and Behavior

Director, Facility for Imaging and Brain Research (FIBRE) & Campus Center for Neuroimaging (CCNI)

School of Biological Sciences, University of California, Irvine