Hi Mike, the large difference in trial count is a genuine problem I am having right now so I really hope you see this 😅 Your videos are the only resource I can find that addresses this problem. Are there any solutions to overcome this? One I could think of is bootstrapping the unit vectors - would this make sense? Or are there actual solutions/papers/resources that I can read? Your videos have always been a tremendous help. Thank you 🙂
Hi April. I discuss this in more detail in my ANTS book. You could sub-select trials (e.g., based on reaction time or some other feature, or randomly selecting) to match trial count, do permutation testing to determine a significance threshold, or empirical bootstrapping to get 95% confidence intervals. But if the smaller of the two conditions still has a "reasonable" number of trials, then the imbalance shouldn't be a problem (what does "reasonable" mean? yeah, hard to say... maybe 30ish?).
@@mikexcohen1 thank you very much for the guidance! And if I may be a bit pedantic, would the permutation be between channels i.e. [ch1_trialN, ch2_trialM], or permuting random time lags between 2 channels? or maybe even other ways
You'd randomize the labeling of trials. So condition "A" would always have, say, 20 trials and condition "B" would always have 80 trials, but at each iteration in permutation testing you would randomly draw trials to be labeled as "A" or "B". This video is about ITPC (within channel). If you're computing synchronization across channels with trial imbalances, you might also consider using this method: www.sciencedirect.com/science/article/abs/pii/S1053811911000917
Hi Mike, the large difference in trial count is a genuine problem I am having right now so I really hope you see this 😅 Your videos are the only resource I can find that addresses this problem. Are there any solutions to overcome this? One I could think of is bootstrapping the unit vectors - would this make sense? Or are there actual solutions/papers/resources that I can read? Your videos have always been a tremendous help. Thank you 🙂
Hi April. I discuss this in more detail in my ANTS book. You could sub-select trials (e.g., based on reaction time or some other feature, or randomly selecting) to match trial count, do permutation testing to determine a significance threshold, or empirical bootstrapping to get 95% confidence intervals. But if the smaller of the two conditions still has a "reasonable" number of trials, then the imbalance shouldn't be a problem (what does "reasonable" mean? yeah, hard to say... maybe 30ish?).
@@mikexcohen1 thank you very much for the guidance! And if I may be a bit pedantic, would the permutation be between channels i.e. [ch1_trialN, ch2_trialM], or permuting random time lags between 2 channels? or maybe even other ways
You'd randomize the labeling of trials. So condition "A" would always have, say, 20 trials and condition "B" would always have 80 trials, but at each iteration in permutation testing you would randomly draw trials to be labeled as "A" or "B".
This video is about ITPC (within channel). If you're computing synchronization across channels with trial imbalances, you might also consider using this method: www.sciencedirect.com/science/article/abs/pii/S1053811911000917
@@mikexcohen1 wow that actually chears up a lot, in terms of properly defining the unit of observation, once again thank you!