Since I’m taking things from BoingBoing now, check these out.
To quote: “The ‘blue’ tiles on the top face of the left cube are the same color as the “yellow” tiles in the top of the right cube.”
No statistical content here at all. But maybe you could draw an analogy to hypothesis testing, the idea that two studies could give identical results, but one could be statistically significant and the other non-significant, if the two studies were embedded in different experimental designs. Or in meta-analysis, two different studies could be interpreted differently if surrounded by different sets of other studies in a hierarchical model. In that case, though, the perceptions of difference would be real and the fact that the two studies were, in isolation, “the same color,” would miss the point.
P.S. Regarding Bill’s comment below: the illusion is not new, it’s just a cool presentation of a result that is well-known among vision researchers.
Well, this is weird. I see four blue tiles on the top of each cube, and they are the same ones.
Oh, sorry, I misunderstood what the point is. The actual "colors" of the tiles may be the same, but they "appear" different. This is a well-known property of color vision. Edwin Land showed in the '50s that you can get a very good approximation of color vision with only two actual colors. What color a particular part of the field appears to be depends strongly on what is nearby.
> maybe you could draw an analogy to hypothesis
> testing, the idea that two studies could give
> identical results, but one could be statistically
> significant and the other non-significant,
> if the two studies were embedded in different
> experimental designs.
Usually the correct conclusion to draw from that is that the two studies ARE "the same colour" and the hypothesis test is misleading. Exactly when that's the right conclusion to draw is the same question as when the likelihood principle is applicable. If only someone was studying that. Oh wait, I'm studying that. Oh good :-)