The following research questions were posed at the top of Chapter 4:
The research described in this dissertation has made significant progress towards answering these questions. A measure based on visual-inertial nulling in the horizontal plane was introduced; in Experiment AIE2, it was used to find a main effect in agreement with prediction and with a trend in reported presence; a reasonable test-retest correlation (.83) was found; and data was collected on the correlation between the nulling measure and the Embedded Figures Test.
That said, one must point to the limitations of this research. In Table 4.7 (page ) there is no difference between conditions on the nulling measure for 13 matched pairs out of 24. This implies that the measure was having difficulty distinguishing the conditions. And this for a meaningful/random comparison, which one might expect to produce large differences between conditions.
This lack of sensitivity was probably due in part to equipment limitations. The resolution was set quite low (240x320 pixels) to maintain a 60 frames-per-second update rate, and the FOV, at 48, was somewhat below what is usually considered the threshold for high presence. Both of these may have introduced floor effects.
However, the lack of interactivity inherent to the procedure used in Chapter 4 may also have played a role. An environment which does not support interactivity may not tend to ``draw people in'' with an intensity needed to clearly demonstrate differences between conditions. In addition to introducing a possible floor effect, the lack of interactivity limits what the procedure of Chapter 4 can be applied to. While it is suited to investigating general display factors such as the relative importance of FOV and resolution, or measuring the foreground occlusion effect, interactive virtual environments fall outside of its scope. A possible visual-inertial nulling measure suited to interactive environments, based on the ``induced motion'' effect found in Pilot Study AIIIP2 (see Appendix D), will be discussed in Chapter 8. The underlying perceptual phenomenon will be discussed below in Section 7.4.
A possible (if untested) benefit of nulling presence measures may be a reduced anchor effect (see Section 2.3.3). A cost of the anchor effect is that one can not readily make comparisons between scores on conditions which were not directly compared in the same experiment. This makes it difficult to incrementally build knowledge by a series of disjoint experiments.
As an example of the anchor effect, consider the reported presence ratings in Experiment AIE1 (Table 4.3, page ) and Experiment AIE2 (Table 4.7, page ). The condition labeled ``48'' in Experiment AIE1 was identical to the condition labeled ``Meaningful'' (MRP) in Experiment AIE2. Yet the reported presence for this condition from the two experiments is quite different. Averaged across first and second reports, the value for the ``48'' condition in Experiment AIE1 was 4.8, whereas the value for the ``Meaningful'' condition in Experiment AIE2 was 2.75. This difference is highly significant on a two-sample t-test (p < .001). The difference presumably arises because in Experiment AIE2 the comparison was against a random scene, which tended to lower over-all presence ratings. In Experiment AIE1, the comparison was against two similar conditions of slightly narrower FOV.
Anchor effects for reported presence are not surprising. Humans were not evolved to assign numbers to mental states and have no robust means for doing so. While there are currently no data on this, it may be that a nulling presence measure, being more deeply-rooted psychologically than self-reported presence values, may exhibit much more consistency across experiments.
Both Experiment AIE2 (Table 4.7, page ) and Pilot Study AIP4 (Table B.2, page ) found roughly a factor of 10 between-participant difference in where the cross-over amplitude occurred. Similar differences were found in Pilot Studies AIP1 and AIP2, using different equipment. A factor of 10 is a conservative estimate: participants exceeded the range of measurement in both directions. A possibility is that the between-participant variation in the cross-over amplitude reflects a more sensitive measure of field dependency than has previously been available. Like the EFT and the rod-and-frame test, the visual-inertial nulling measure requires participants to extract a signal from a conflicting or distracting visual pattern. But the visual-inertial nulling procedure uses a more compelling visual stimulus than either of the other two tests; and, in addition, the visual-inertial nulling procedure avoids the strong gravitational cue which is present in the rod-and-frame test. The visual-inertial nulling procedure might also be used clinically as a means to diagnose vestibular damage.
About a third of the participants in Experiment AIE1 did not follow the general pattern that the 48 condition produced higher reported presence than the other two conditions. This suggests that for a minority of the participants, the effect of a foreground occlusion increasing presence may have been stronger than the effect of wider FOV increasing presence, at least for relatively narrow FOV's.
The protocol for Experiment AIE1 and Experiment AIE2 called for reported presence data to be gathered with the visual and inertial motions congruent, but with different sinusoidal amplitudes (30/sec and 20/sec, respectively). An informal observation is that participants did not seem to be aware of what the relationship was between the visual and inertial amplitudes, even to the extent of knowing whether or not they were equal. While visual-inertial phase differences are very apparent, amplitude differences are not, at least at the conscious level.
SSQ data were gathered before and after every session. These data were not formally analyzed, as they were gathered primarily to make a rough check that participants were not experiencing serious malaise. Simulator sickness did not appear to be a serious problem. The lack of strong symptoms in an experiment which involved a clear visual-inertial sensory conflict may seem surprising. It is probably due to the short duration of exposures to the conflicting stimuli. Exposures tended to be about a minute long, separated by a rest with eyes closed of one or two minutes while conditions were changed. It appeared that participants exposed to conflicting stimuli for longer periods of time did develop symptoms of simulator sickness.