Comment by havan_agrawal

Shouldn't there also be a few "control" challenges sprinkled in where all three are the same color and there's no "right" answer? If the test is implemented well and/or there is no human bias (either from the previous question or from the positioning of the circles), then you'd expect to see a uniform distribution of answers on the control. If there is bias (e.g. some innate preference for the top circle (say)), that should get adjusted for in the final analysis.