We also tested if dominance switches and changes in awareness had any temporal relationship to one another. In all subjects, there was indeed a correlation between the timing of the onsets and offsets of the awareness intervals and the dominance switches (BE tests,
p < 0.01;
Materials and methods). This is illustrated in the cross-correlograms of
Figure 4A in which the density of awareness onsets (
Figure 4A1) and offsets (
Figure 4A2) is plotted as a function of time for one of the coherence conditions (90%). Note that there was a clear peak in the occurrence of awareness onsets immediately
after the dominance switches (
Figure 4A1, lag time about 150–400 ms) while the occurrence of awareness offsets clearly peaked just
before these switches (
Figure 4A2; lead time about 200–500 ms). Both effects were contrast dependent with the weakest and strongest coincidences occurring in the low- (light gray) and high- (black) contrast conditions, respectively. Similar results were obtained for the other coherence conditions (not shown). These findings suggest that dominance changes tend to trigger awareness while loss of awareness or attention, in turn, tends to trigger a dominance change. An alternative interpretation, however, is that these events are caused by a common mechanism. Adaptation reciprocal-inhibition models (Blake,
1989; Freeman,
2005; Lehky,
1988; Noest et al.,
2007; Wilson,
2003) suggest, for example, that switches arise from adaptation of the currently dominant population and recovery of the suppressed population. It is conceivable, therefore, that awareness is more easily lost at the end of a dominance epoch due to progressive adaptation of the dominant population and that, at the moment of the switch, awareness is more easily (re)gained because the now dominant population has recovered from its previously adapted state. If so, one would expect that changes in awareness probability are phase locked to the dominance switches. We, therefore, analyzed the awareness probability as a function of normalized time in the two respective dominance epochs. In the resulting phase plots (
Figure 4B), negative phase values correspond with dominance of the motion direction in the ipsilateral eye, and positive phase values correspond with dominance of the other eye's motion direction. Note, there was a strong periodic modulation of the awareness probability, ANOVA, main effect of phase:
F(71, 4211) = 1.73,
p ≪ 0.001. Typically there was an abrupt drop in awareness probability just before the occurrence of a dominance switch, which was then followed by a somewhat more gradual increase in awareness probability over the course of the subsequent dominance epoch. The magnitude of this modulation varied with stimulus contrast, ANOVA, main effect,
F(2, 4211) = 3114.37,
p ≪ 0.001, and coherence, ANOVA, main effect,
F(3, 4211) = 42.65,
p ≪ 0.001, but the shape of the normalized time courses was remarkably similar, ANOVA, phase × contrast interaction,
F(142, 4211) = 0.46,
p > 0.99; phase × coherence interaction,
F(213, 4211) = 0.16,
p > 0.99; and phase × contrast × coherence interaction,
F(426, 4211) = 0.17,
p > 0.99. In fact, it appeared that even in the high-contrast condition, subjects generally claimed being unaware of visual motion at the time of a dominance switch. One might worry that this striking result could be due to some movement coordination problem, forcing our subjects to transiently release the middle mouse button to indicate a dominance switch on the other two buttons, but this was not the case: Our subjects did not report such coordination problems on debriefing, and for the vast majority of switches, releases of the middle button and presses of that same button did not coincide within a 2-s time window around the switch (not shown).