The answer is: kind of!
This new study by Pierre Lebreton and colleagues looked at ways of conducting a type of ‘eye tracking’ study on MTurk using some interesting metrics including gaze estimation. This article is a ‘how to’ of how to conduct this kind of study and is quite interesting.
The authors describe how they recruited 30 participants and ten participants had to be booted for a variety of reasons:
Out of the 30 workers, the input provided by 20 of them was meaningful for our experiments. The excluded videos included cases where the gaze tracking algorithm failed to provide reliable results due to very large head movements (2 cases), movement of the head out of the image (2 cases), bad illumination conditions (3 cases) and strong re ections from glasses which completely occluded the eyes (3 cases).
This does make me wonder what sample size would be necessary to get valid results on these types of studies.