judgment by revolving the roller _C_. The dividing line between the
open and filled spaces was continuously recorded on the paper from
below by a pin not shown in the illustration.

The rollers, of which I had three, were easily removed or turned
about, so that the open space was presented first. In one of the
distances on each roller both spaces were unfilled. This was used at
frequent intervals in each series and served somewhat the same purpose
as reversing the order in which the open and filled spaces were
presented. With some subjects this was the only safe way of securing
accurate results. The absolute distances measured off were not always
a sure criterion as to whether the filled space was under-or
overestimated. For example, one rather erratic subject, who was,
however, very constant in his erratic judgments, as an average of
fifty judgments declared a filled space of 4 cm. to be equal to an
open space of 3.7 cm. This would seem, on the surface, to mean that
the filled space had been underestimated. But with these fifty
judgments there were alternated judgments on two open spaces, in which
the first open space was judged equal to the second open space of 3.2
cm. From this it is obvious that the effect of the filling was to
cause an overestimation--not underestimation as seemed at first sight
to be the case.

In another instance, this same subject judged a filled space of 12.0
cm. to be equal to an open space of 12.9 cm., which would seem to
indicate an overestimation of the filled space. But an average of the
judgments on two open spaces that were given in alternation shows that
an equivalence was set up between the two at 13.7 cm. for the second
open space. This would show that the filling of a space really
produced an underestimation.