When Joseph Di Lorenzo and I 
examined this question, 
therefore, we used a large 
frame, one subtending a visual 
angle of 54 degrees. We found 
that the frame’s tilt was then 
underestimated. That being the 
case, it is understandable why 
a strong illusion of rod 
orientation occurred. If, for 
example, a frame tilted by 30 
degrees appears to be tilted by 
only 20 degrees, then a vertical 
rod within it that is displaced 
by an angle of 30 degrees from 
the frame must appear to be 
tilted by 10 degrees in the 
direction opposite to that of the 
frame. To appear upright, 
therefore, the rod would have 
to be tilted by 10 degrees toward 
the tilted frame. In fact there 
was a high correlation between 
the underestimation of the 
frame’s tilt and the illusory tilt 
of the rod. Other evidence we 
obtained consistently 
supported the conclusion that 
whatever factor caused the 
frame to appear less tilted than 
it was also led to a 
correspondingly large effect on 
the appearance of the rod.

    This conclusion about the 
rod-and-frame effect is quite 
consistent with the strong 
effect on perceived body and 
object orientation that occurs 
when the observer is inside a 
tilted room. The underlying 
principle seems to be this: A 
large structure with 
rectangular coordinates tends 
to become the surrogate of the 
vertical-horizontal coordinate 
system of space. But to do so, it 
must surround the observer or, 
equivalently, it must occupy a 
large angle of the visual field. 
However, when the observer is 
inside the structure, visual 
capture seems to be more or less 
complete, with the result that 
gravity-based information is no 
longer a conflicting factor. 
When the observer is outside 
the structure, some degree of 
visual capture undoubtedly 
occurs as well. Experiments 
have shown that observers 
then err in their perception of 
how their own bodies are 
oriented. But, because the 
phenomenon of visual capture 
is not complete, there is a 
conflict between gravity 
information and visual 
structure. The outcome is a 
compromise. The frame still 
looks tilted, but less tilted than 
in fact it is.