position when closed in sleep.  But this upward roll is not an uncomplicated movement.  There takes place at the same time a relaxation of binocular convergence, which in sleep may be replaced by a slight divergence.  This tendency of the axes of vision to diverge as the eyes are raised is undoubtedly connected biologically with the distribution of distances in the higher and lower parts of the field of vision, of which mention has already been made.  Its persistence is taken advantage of in the artificial device of assisting the process of stereoscopic vision without instruments by holding the figures to be viewed slightly above the primary position, so that the eyes must be raised in order to look at them and their convergence thereby decreased.  It is by the concomitance of these two variables that the phenomena of both this and the preceding series of experiments are to be explained.  In the present case the elimination of a fixed point of regard is followed by a release of the mechanism of convergence, with a consequent approximation to parallelism in the axes of vision and its concomitant elevation of the line of sight.

The second fact to be noted is the reduction in amount of the mean variation.  The series of values under the three sets of experimental conditions hitherto described is as follows:  I. 7’.69; II. 31’.42; III. 17’.16.  This increase of regularity I take to be due, as in the case of the lighted room, to the presence of a factor of constancy which is not strictly an element in the judgment of horizontality.  This is a system of sensory data, which in the former case were transient—­the vision of familiar objects; and in the latter resident—­the recognition of specific experiences of strain in the mechanism of the eye.  The latter sensations exist under all three sets of conditions, but they are of secondary importance in those cases which include the presence of an objective point of regard, while in the case of judgments made in total darkness the observer depends solely upon resident experiences.  Attention is thus directed specifically toward these immediate sensational elements of judgment, and there arises a tendency to reproduce the preceding set of eye-strains, instead of determining the horizon plane afresh at each act of judgment upon more general data of body position.

If the act of judgment be based chiefly upon sensory data connected with the reinstatement of the preceding set of strains, progressions should appear in these series of judgments, provided a constant factor of error be incorporated in the process.  This deflection should be most marked under conditions of complete darkness, least in the midst of full illumination.  Such a progression would be shown at once by the distribution of positive and negative values of the individual judgments about the indifference point of constant error.  As instances of its occurrence all cases have been counted in which the first half of the series