Distance Estimation Experiment 1

Distance Estimation Experiment #1

Immersion, perspective cues, barriers, and gender.

Method


Participants. Twenty people (10 men) ranging in age from 18 to 25 (M = 19.81, SD = 1.60) were recruited from the University of Washington department of Psychology human subject pool. Participants were paid $10 per hour.

Stimuli and apparatus. The virtual environment was a dark gray cubic room measuring 300 world-units on each side. On half of the trials, a rectangular grid was superimposed on the walls, floor, and ceiling of the room. At the beginning of each trial, two small boxes, one red and the other green, were placed at random locations in the room (see fig DE1-1).


Figure DE1-1



This environment was modeled in World Up by SENSE8 and run on a Pentium Pro 200 using an Oxygen 102 graphics accelerator board. For the desktop condition, participants viewed the scene while sitting 38 cm from a 35 cm x 26.5 cm monitor with a resolution of 1152 x 900 (32K colors, 76 Hz refresh). The GFOV of the world was set at 80 degrees. Viewing distance was controlled by a hood placed around the monitor that also served to shield participants’ peripheral view. The frame rate for this system was approximately 8.22 frames per second.

Participants in the immersive condition interacted with the VE by moving in a 6’ x 6’ curtained enclosure in the real world. Head movements were tracked with a six degree-of-freedom tracker (Polhemus Fastrak). The sensitivity of the tracker was calibrated so that collisions between the virtual viewpoint and the room’s walls corresponded to collisions between the participant’s body and the curtains of the real room. These collisions were signaled with a tone. Immersed participants viewed the VE in a VR4 HMD from Virtual Research, which offers a 60 degree horizontal field of view and 742 x 230 resolution. These participants told their distance estimates to the experimenter who entered them into the computer. Participants in the desktop condition navigated by means of mouse commands that provided six degrees-of-freedom of movement. Desktop participants entered their distance estimates by themselves.

Procedure. Each participant was randomly assigned to either the immersed or desktop condition (subject to the constraint that equal numbers of participants represent both groups). Participants took part in three one-hour experimental sessions conducted on three different days. Each session was composed of five ten-minute blocks of trials separated by a two-minute break period.

At the beginning of the experiment, participants were told that they would be repeatedly placed in a virtual room that measured 300 x 300 x 300 units and to scale their distance estimates accordingly. They were told that the researchers were interested in their initial estimates of the distances between the two blocks and were instructed not to spend too much time with any one trial. At the beginning of each trial, the participant’s virtual viewpoint was set in the corner of the room at a height that modeled their real-world height. The participant then navigated through the room, examining the two blocks from as many vantage points as he or she felt necessary to make an accurate estimate of the distance between them. When ready, the participant made his or her estimate and then was given feedback about the correct distance, whereupon the next trial began.

Results

Participants completed an average of 139 (SD = 44.08) trials over all three sessions. For the analyses that follow, accuracy (mean percent error), estimated model parameters, and model fits (R^2) were calculated separately for each participant in each trial type (grid present and grid absent). These dependent variables were then subject to separate 2 (grid present/absent) x 2 (desktop/immersed) x 2 (male/female) mixed effects univariate ANOVA’s. To ensure comparability between model fits, all models were fit using non-linear regression. All statistical tests employ a two-tailed alpha level of .05.

Percent overestimation. Across all twenty participants, overall percent distance errors ranged from -9.23 to 9.82 (M = 0.31, SD = 4.78) and were not significantly different from zero (t(19) = 0.298, p = .78). The presence of the grid had a significant, though slight, effect on overestimation. Participants tended to overestimate distances more when the grid was absent (M = 3.26, SD = 3.79) than when it was present (M = 1.91; SD = 3.44). The between-subject variables -- display mode (desktop vs. immersed) and gender -- had a less powerful effect on accuracy. A 2 (grid present/absent) x 2 (desktop/immersed) x 2 (gender) mixed effects ANOVA revealed a significant main effect of the grid (F(1, 16) = 5.73, p = .029). No other main effects or interactions were significant at the .05 level, although the difference in overestimation between immersed (M = 1.55, SD = 5.03) and desktop (M = 3.63, SD = 4.12) participants was notable (F(1, 16) = 2.06, p = .17). The effects of grid and immersion are evident in figure DE1-2. Effect sizes (eta-squared) of grid presence, gender, and immersion on overestimation were estimated at .26, .13, and .11 respectively.


Figure DE1-2



Steven’s exponents and power model fit. Exponent estimates for the twenty participants (collapsed over grid conditions) ranged from 0.71 to 1.02 (M = 0.88, SD = 0.08) and were significantly less than one (t(19) = 6.44, p < .0001). For ten (50%) of the participants, the 95% confidence interval of the exponent estimate did not contain one. A 2 (grid) x 2 (display mode) x 2 (gender) mixed effects ANOVA revealed a significant main effect of gender (F(1,16) = 5.31 , p = .04), such that exponents for men (M = 0.92, SD = 0.07) were significantly higher than those for women (M = 0.85, SD = 0.07). No other main effects or interactions were significant. The effect of display mode approached significance (F(1,16) = 4.27, p = .055), with exponents estimated from desktop performance (M = 0.85, SD = 0.08) being slightly lower than those from the immersed condition (M = 0.91, SD = 0.08). Figure DE1-3 illustrates the effects of gender and immersion on exponent estimates.


Figure DE1-3



Averaged over all participants, the estimated modulus (M = 2.03; SD = 0.96) was significantly different from one (t (19) = 4.81, p < .001). Estimates of the modulus (k) were highly correlated with exponent estimates(r(18) = .97, p < .001), and the effects of the independent variables on the modulus was the same as for the exponent.

R^2 values were not significantly affected by gender, immersion, or the presence of the grid. A 2 (grid) x 2 (display mode) x 2 (gender) mixed effects ANOVA revealed no significant effects or interactions on participants’ R^2 values. However, again, the effect of immersion approached significance (F(1,16) = 3.87, p = .07) with the power model fitting better for immersed participants (M = .74, SD = 0.09) than for desktop participants (M = .65, SD = 0.10).

[ return to top ]