exp1: probe detection

[eivinasbutkus]

Settling on experimental design.

1. We start off with 10 scenes (either generated using pure Brownian motion, or ISR).
2. We come up with 40 trials (4 different probe placements per scene):
   • 20 TD trials with high and low attention
   • 20 DC trials with high and low attention
3. Each condition has these 40 trials. The probe placement is the same in every condition, but the queried dot ("is this a target?") is different.
4. To pre-select the queried dot, we effectively choose uniformly from the first two targets (dots 1 and 2) and the first two distractors (dots 5 and 6). This should be fine since we do not care about the exact accuracy comparison between model and humans per trial, but rather to just make sure that humans are still doing MOT and not only probe detection, and that they have a reasonable overall accuracy.

@belledon, is this what you were thinking?

[belledon]

its close. a few comments

• I think we should have TD and DC have separate experiments. having the same scene show up 4 times is a bit much. I was beginning to recognize repeated scenes and I'm assuming that the average subject might do so with 4 reps.

• note that 50% of the trials need have no probe. right now this is done by picking the top 20 peak attention trials and picking a random subset of 10 to have probes. that being said the total number of trials per condition will still be 40 (10x2 + 10x2).

@iyildirim , the think we are wrestling with is whether or not to include 8 conditions. In some sense we not need to know the accuracy per target since all we techincally need is average accuracy, assuming that subjects with similar accuracy have similar attention dynamics. However, it could also be the case that some targets (perhaps the one with the probes) are harder to track and thus we might want to include the accuracy of tracking the probed tracker into our analysis (although I see a potential case of controlling a mediator if we do).

[iyildirim]

Thanks for these thoughts. Let's place the probes always on targets. We will then have half of the trials no probe.

Assuming we do the first probe detection experiment using Exp0 stimuli and noting the fact that we have average performance for all trials in that stimuli thanks to our Exp0:

What if we have 120 trials (each a unique scene from Exp0) as before, and two conditions. In each condition, we will have 30 probes from TD, 30 probes from DC and 60 no probe trials. 15 of TD trials and 15 of DC trials be high attention probes and the remaining 15 in each kind be low attention probes.

Between conditions we swap whether scenes 1 to 30 are based on TD or scenes 31 to 60 are based on TD.

Putting both TD and DC in the same experiment will allow us to do within subject comparisons, strongest kind one would like. But I can see the argument against this to consider two different experiments. We can choose a design and start with that.