Evaluating the Effectiveness of Spatial memory in 2D and 3D Physical and Virtual environments

-Andy Cockburn

Bruce McKenzie

Presented By Viji Murugesan

Abstract

It investigates the effectiveness of spatial memory in real-world physical models and in equivalent computerbased virtual systems.

Introduction

Several experiments have shown that spatial organization of information allows efficient access to items in graphical user interfaces. For example, in evaluating their 3D Data Mountain, Czerwinski, van Dantzich, Robertson and Hoffman , found that spatial memory allowed rapid access to web-page thumbnails several months after the pages were originally organized. Spatial memory appears to be a valuable tool in supporting efficient information organization.

Contd..

This papers concern is the effectiveness of spatial memory as interfaces move from 2D toward 3D spatial organizations because there is increasing research and commercial interest in systems that provide 3D interfaces for standard file and document management tasks. Example systems include the Task Gallery and Win3D (www.clockwise3d.com).

Related Work

2 areas are relevant First, there has been extensive prior research comparing the effectiveness of 2D and 3D user interfaces. Second, several researchers have examined the role of spatial memory in predicting user performance with graphical user interfaces and in supporting information retrieval.

1)Comparing 2D and 3D interfaces

Contd..

The terrain comprehension result contradicts that of St. John, Oonk and Cowen [14], which shows better understanding of terrain shape using a 3D interface. However, another of St. John et al.s tasks showed better understanding of the vertical axis in 2D: subjects were better able to find the highest point on a map in the 2D condition. Further problems with altitude assessment in 3D are also reported in other papers.

Contd..

Outside aviation research there have been several evaluations showing no significant differences between 2D and 3D. Cockburn and McKenzie [2] compared 2D and 3D versions of the Data Mountain [11] and found no significant difference in performance, but a significantly higher subjective rating for 3D. Smallman, St. John, Oonk and Cowen [13] showed that 2D symbolic representations of military targets allowed faster and more accurate identification than 3D icons.

Contd..

Wickens, Olmos, Chudy and Davenport [19] provide a fitting summary for prior work on 2D versus 3D evaluations: whether the benefits of 3D displays outweigh their costs turns out to be a complex issue, depending upon the particular 3D rendering chosen, the nature of the task, and the structure of the information to be displayed.

2)Spatial memory and user interfaces

Several papers provide evidence that spatial aptitude is a strong predictor of performance with computer-based user interfaces. For example-The spatial arrangement of web page images provided by the Data Mountain allowed more rapid and accurate retrieval of pages (from sets of 100 pages) than the Favorites mechanism in Microsoft Internet Explorer. Jones and Dumais [8] provide some cautions on overreliance on spatial organization. Indicate combinations of semantic and spatial organization enhance performance.

2D,2 D and 3D Physical and Virtual Interfaces

Six interfaces were used in the evaluationthree physical interfaces and three computer-based virtual equivalents. 2D- X and Y coordinates manipulated. The occlusion algorithm for overlapping pages is deterministically applied so that pages with a lesser value on the y-axis (pages lower in the display) are placed in front of pages with higher y-axis values. 2 D-receding inclined plane on which pages can be located, closely analogous to Data Mountain. Pages recede in the virtual interface by dynamically reconfiguring the image size, providing a sense of perspective. 3D-allow x, y and z coordinates for each page to be specified.

2D Physical interface

Contd..

Webpage thumbnails-90x90mm photo quality printed images Mounted on stiff cardboard and covered in clear plastic for protection Title for page-overlaid on top of the Netscape window banner information Clips on the back of each card allowed them to hang from the fishing line. The 2D physical interface was constructed from chipboard and horizontal lines of taut fishing-line separated by 2cm in a single vertical plane. String marked the 900x710mm page placement boundaries.

2 D Physical interface

Contd..

Created by reclining the 2D interface to an angle of 25. The cards hung vertically off each fishing-line. The string placement boundaries were the same as the 2D interface (900x710mm).

3D Physical interface

Contd..

Constructed from painted steel rods and horizontal lines of taut fishing-line placed at 5cm intervals vertically and horizontally. The 3D structure allowed pages to be placed in a 900x900x750mm x, y and z space. Cards could be overlapped under similar conditions to those for the 2D and 2D interfaces.

Subjects position

In all 3 physical interfaces, fishing line caused the minimal occlusion of pages subjects sat on a height-adjustable chair set approximately 50cm from the front-edge of the interface. This gave an angle at the eye of approximately 84 between the left and right front edge of each interface and approximately 40 at the back of the 3D interface. Head positions were normally approximately mid-height in the 2D interface, towards the top edge of the 2D interface, and one-third height in the 3D interface. The subjects used a laser pointer to identify target pages when using the physical interfaces.

Virtual Interface

Written in Tcl/Tk and created windows of 800x600 pixels. The display resolution was 1024x768 with 79 dots-per-inch. Subjects sat approximately 50cm from the screen, giving a horizontal angle at the eye of roughly 30 for the front edge of each interface and approximately 18 and 9 for the back edge of the 2D and 3D interfaces. The thumbnails used in all three interfaces were miniaturized rendered web pages. The browsers banner information was not included in the thumbnail images because it was reported to detrimentally affect the subjects ability to visually identify pages. In all three interfaces, pressing and holding the right mouse button over any thumbnail magnifies it to 250x250 pixels, and reveals the page title information.

2D virtual interface

Contd..

All thumbnails are 85x85 pixels 63 pages can be placed in the display without any overlapping. The occlusion algorithm for overlapping pages is equivalent to that in the physical systems: when pages overlap, those with lesser y-axis values (lower in the display) are placed in front.

2 D virtual interface

Contd..

Pages are placed on receding plane-Adds perspective to the display Pages are positioned by dragging with the left mouse button. As pages are pushed up the plane, the thumbnail images diminish from a maximum size of 130x130 pixels at the bottom to a minimum of 40x40 pixels at the top. 71 images can be arranged in the 2D interface display before overlapping becomes essential. The 2D occlusion algorithm is identical to that in the 2D interface.

3D virtual interface

Contd..

x, y, and z coordinates for each thumbnail can be altered within a virtual cube. Thumbnails are the maximum size of 130x130 pixels at the front of the placement cube, and the minimum size of 40x40 pixels at the back. 69 images can be arranged in the display before overlapping becomes essential. Dragging with the left mouse button changes the x and y coordinates of thumbnails. Vertically dragging with the middle button changes the z coordinate, pushing pages further away or pulling them closer.

Contd..

To help overcome the problems of depth/altitude ambiguity whenever the user moves a thumbnail, ground-intercept information is revealed Natural occlusion rules apply, with pages at the front of the cube occluding those further away. There were no noticeable performance differences between the three virtual interfaces.

Ground-intercept information for the thumbnail being moved in the 3D virtual interface.

Experimental Method

Aim-investigate differences in peoples ability to use their spatial memory in physical and virtual systems, and to see what effects occur as richer levels of a third dimension are available. 3 between-subject factors factor realism has two levels, physical and virtual. factor dimension has three levels: 2D, 2D and 3D. The final factor density is within-subjects and has three levels: sparse(33), medium(66) and dense(99) pages.

Procedure

33 pages-one at a time, relocatable Same set of pages for all subjects but presented in random order Placed in physical interface and then virtual interface Then, retrieval task (time limit 100 sec) 69 subjects (computer science students) Training -10 min, each evaluation session-1 hour Repeated for 66 and 99 pages Answered 5 scale questions after the tasks

Results

2D users-static 2 D users-leaned forwards and upward in order to get a better view between pages. 3D users-used their upper body to move their head up to one foot to the left or right in order to look around occluding pages. Totally 2070 trials in 6 interfaces Mean time for retrieval-4.13 sec (low) 5 trials failed (one in each of the 3D-physical-dense, 2Dvirtual-dense, and 3D-virtual-dense conditions, and two in the 3D-virtual-medium condition. subjects commented that they were much faster at retrieving pages than they expected, indicating that their spatial memory was effective, but not trusted.

Retrieval Results

Physical interface meantime- 3.5 sec Virtual interface meantime -4.6 sec Mean time for 3 densities Sparse-3.2 sec Medium-4.2 sec Dense-5 sec Mean time for 3 dimensionalities has marginal difference 2D - 3.7sec 2 1/2 D 3.8sec 3D - 4.8sec

Mean page retrieval times. Error bars show one standard error above and below the mean.

Analysis of Results

Physical Interface-2 D had the highest time Virtual Interface-2 D had the lowest time (why?) No significant difference between 2D ,2 D virtual interface Comparison of 2D and 3D physical interface(2 D excluded)-big contrast(3D result in confusion) 2D-2.6 sec 3D-3.6 sec (interesting because the 3D interface allowed the subjects to organize images on a 2D plane that was slightly larger than that available in the 2D interface). 3D virtual interface more time(4.8 sec) subjects complained of clutter and protested that the window was too small when using the 3D virtual interface. This is negative reaction because 3D virtual system allowed more non overlapping pages to be placed in the display (69) than the 2D interface (63), and roughly the same number as the 2D interface (71).

Subjective measures

Mean response across all densities (disagree 1, agree 5). Q1 It was easy to place the pages Physical Interface-3.6 Virtual Interface-3.1 Q2 I will be able to quickly find pages no significant difference Q3 I was able to quickly find pages -subjects ability to retrieve pages with the physical interfaces as the number of dimensions increased (2D 4.2, 0.8; 2D 4.0, 1.0; 3D 3.7, 0.9): H=2.8, p=.06. The virtual interfaces showed no significant differences. Q4 The display is cluttered. Physical Interface-3.1 Virtual Interface-3.7

Contd..

Q5 Overall the interface/structure provides an effective way of organizing and retrieving web pages. Physical interface-4 virtual-3.3 Responses to Q2 were significantly lower than their responses to Q3, indicating that the subjects did not trust their spatial memory in each density Responses with the physical interfaces reliably decreased as the number of dimensions increased Responses across the three virtual interfaces were noticeably lower (worse) than the physical interfaces, but were not reliably different from each other

Discussion

Time taken increase through 2D,2 D and 3D interfaces subjects assessment of the effectiveness of the interfaces decreased through the 2D,2D and 3D conditions. Performance deteriorated as the number of pages in the displays increased. Results reported for the physical 2D interface are artificially poor because its implementation simply involved reclining the 2D interface to an angle of 25,maintaining the original 2D organization space of 900x710mm. The angle at the eye between the bottom and top of the 2D organization is therefore substantially less than that for the 2D interface.

Contd..

Although the fishing-line was successful at minimizing occlusion, it had two effects on the way the physical systems were used. First, fishing-line, like clothesline, provides a natural affordance of the way it should be used: items hang along it. Most subjects using the physical interfaces relied heavily on a horizontal grouping arrangement for grouping related pages along the lines. The second problem caused by the use of fishing-line is that it creates discrete placement locations on the y and z axes. In the 2D and 2D environments, cards cannot be placed less than 2cm vertically apart, and in the 3D environment they can be no closer than 5cm on the y and z axes.

Conclusion

Results show that our subjects ability to quickly locate web page images deteriorated as their freedom to use the third dimension increased. Their subjective responses also indicated that they found the 3D interfaces more cluttered and less efficient. Spatial memory clearly provides an effective aid to information retrieval but does 3D help? Results indicate that for relatively sparse information retrieval tasks (up to 99 data items), 3D hinders retrieval.

Future Work

investigate whether 3D spatial arrangements allow more effective retrieval than a series of 2D planes for larger data sets.

References

Barfield, W. and Rosenberg, C. Judgments of Azimuth and Elevation as a Function of Monoscopic and Binocular Depth Cues Using a Perspective Display.Human Factors 37, 1, 173-181. 1995. Cockburn, A. and McKenzie, B. 3D or Not 3D?Evaluating the Effect of the Third Dimension in a Document Management System. In Proceedings of CHI'2001, Seattle, April 2001, ACM Press, 434-441. Czerwinski, M., Dumais, S., Robertson, G., Dziadosz,S., Tiernan, S., and van Dantzich, M. Visualizing Implicit Queries for Information Management and Retrieval. In Proceedings of CHI'99, 1999, 560567. Czerwinski, M., van Dantzich, M, Robertson, G., and Hoffman, H. The Contribution of Thumbnail Image,Mouse-over Test and Spatial Location Memory to Web Page Retrieval in 3D. Proc. INTERACT1999. 163-170.

Contd..

8. Jones, WP. and Dumais, ST. The Spatial Metaphor for User Interfaces: Experimental Tests of Reference by Location versus Name. ACM Transactions on Office Information Systems 4, 1, 42-63. 1986. 11. Robertson, G., Czerwinski, M., Larson, K., Robbins, D.Thiel, D., and van Dantzich, M. Data Mountain: Using Spatial Memory for Document Management. In Proceedings of UIST'98, 1998, ACM Press. 153-162. 13. Smallman, H., St. John, M, Oonk, H. and Cowen, M. Track Recognition Using Two-Dimensional Symbols or Three-Dimensional Realistic Icons. SPAWAR Systems Center Technical Report Number 1815. 2000. www.nosc.mil/sti/publications/pubs/tr/1818. 14. St. John, M., Oonk, H. and Cowen, M. Using Two- Dimensional and Perspective Views of Terrain. SPAWAR Systems Center Technical Report Number 1815. 2000. www.nosc.mil/sti/publications/pubs/tr/1815. 17. Wickens, C., Liang, C., Prevett, T. and Olmos, O. Egocentric and Exocentric Displays for Terminal Area Navigation. In Proceedings of the Human Factors and Ergonomics Society 38th Annual Meeting, 1995, 16-20.