Authors
 

Woei-Jyh Lee	adamlee@cs.umd.edu
Ser-Nam Lim	sernam@cs.umd.edu
Tzu-Ting Chen	jacktar@cs.umd.edu
Yu-Lin Wen	ylwen@cs.umd.edu

The widely accepted window environments of modern graphic user interfaces are largely two-dimensional, although windows can overlap.  Some innovators believe that three-dimensional interfaces can contribute to faster learning and improved user satisfaction, but they recognize that slower performance may be the price of a richer visual experience.  However, advocates of three-dimensional interfaces also suggest that retention may be better because spatial cognition is tied to 3D.  Our study gave 12 tasks to 12 subjects using Microsoft Windows Explorer (a 2D interface) and the same 12 tasks to 12 other subjects using ClockWise Win3D (a 3D interface).  For four tasks the 2D users had statistically significantly faster performance speeds than the 3D users.  No significant differences were found for errors, subjective satisfaction or retention.  We propose a predictive model based on clicks to complete tasks.

Review of commercial systems

In the recent years, there have been an abundance of operating systems that have opted towards the traditional two-dimensional windows environment.  Microsoft and Apple lead the industry with their own interfaces.

Microsoft started out with their command line interface in the early 80s with the MS-DOS system.  DOS was once widely used, but as time progressed and at the advent of a windows based interface, Microsoft saw it fit to switch all their products to a windows based environment.  Among these products include Windows 1.x/2.x/3.x, Windows 95, Windows 98, Windows NT and finally Windows 2000 and Windows Millennium edition.

Apple, with their own version of a windows based environment, launched its own line with their MAC OS built on the Macintosh.  The Macintosh has been using this windows based interface even before the wide spread use of Microsoft products.  In fact, many window interfaces can be traced back to Apple.  Pull down menus, window resizing are just some of the many tools that Apple pioneered.

What has been mentioned so far are PC based systems.  Systems like Unix and Linux, which use essentially command line interfaces for expert users, have readily been shifting to the use of a windows based interface.  Products like CDE (Common Desktop Environment), winLinux, and X-windows are all forms of a windows based front end interface that link to the original systems in the back end.

Internet browsers like Internet Explorer and Netscape Navigator are also commercial products that use this windows based interface.  These browsers have evolved from the command line interfaces like Lynx, found under the UNIX environment.

All the above mentioned are products that use two-dimensional window interfaces.  What about three-dimensional windows interfaces?  Commercial systems that use three-dimensional window environments are relatively fewer in number.  3DWM is a commercial three-dimensional tool that provides the necessary primitives for application developers [3DWM], although 3DWM is still under development.

Another, more significant three-dimensional window product is ClockWise Win3D.  This is the product that we have used in our experiments.  ClockWise Win3D is a three-dimensional front end to the of the Microsoft Windows operating system.  It has the usual features that can be found in any Microsoft Windows products, including icons, the start menu, and others except that everything that would be done in a two-dimensional environment, is now done in a three-dimensional environment.

The Clockwise environment shows these three dimensional rooms and users can move forward/backward and turn left/right. They cannot turn or move up/down.

Review of experiments:

[Ark et al. 98] studies the similarities and differences of two-dimensional icons and three-dimensional realistic GUIs.  The experimental results show that subjects located targets more quickly when using interfaces with three-dimensional objects and ecological layouts rather than with two-dimensional objects and simplistic layouts.

[Bowman et al. 99] is a study on the effect of various virtual travel techniques on the spatial orientation of users.  The results showed that techniques using virtual translation with physical rotation produces better user performance.  The implication of this study is the requirement for more comprehensive navigation techniques in a three-dimensional design.  This is because users of three-dimensional windows environments could easily get disoriented spatially.

[Cockburn et al. 01] is a similar study to ours - comparative evaluation of two document management systems that differ only in number of dimensions used for displaying and interacting with the data.  Results show that there is no significant difference between two-dimensional and three-dimensional windows environments.  Although this study is focused on document management system, it really provides an invaluable insight into the results that are achieved in our study.  On the other hand, our study involves tasks that are different from those in a document management system.  This could be a source of difference that we will evaluate later in the experiment section.

[Darken et al. 96] talks about the influence on user performance of the environmental cues suggested by the underlying design principles.  The experimental results show that subjects in the treatment, without any additional cues, were often disoriented and had extreme difficulty completing a given task.  This result suggests that providing environmental cues is critical to user performance.  In our case, the use of different window dimensions provides environmental cues of a different nature to users.

[Elvins et al. 97] compares two designs that provide virtual environment landmarks in a three-dimensional environment.  They are via landmark textual descriptions or thumbnail images and via three-dimensional thumbnail images called a worldlet respectively.  This study found that worldlets provide better user performance. 

[Elvins et al. 98] looks into the effect of navigating the internet via three-dimensional thumbnails and results are compared to navigating via two-dimensional thumbnail images.  The experimental results show that three-dimensional thumbnails improve landmark knowledge and expedite way finding in a virtual environment.  The intuitiveness of a three-dimensional environment, in this case, plays an important role in providing knowledge via close resemblance to real life objects.

[Risden et al. 00] studies the use of a three-dimensional design in web browsers.  The experimental results indicate that it enables the users to perform search tasks faster and more reliably when compared to a two-dimensional web browser design.

[Wiss et al. 98] did a comparative study of the differences between three three-dimensional visualization tools, namely, the CAM tree, Information Cube and Information Landscape.  The study found that there are inherent problems for each type of design when they are used for data sets that they are not intended for.  As a result, it is difficult to generalize their usage to other domains.  This study therefore is useful in pointing out that three-dimensional designs might not achieve wide spread acceptance because they might be useful only for certain domains. 

[TaskGallery] is a running 3D research prototype user interface that expands the desktop into an entire office with an unlimited number of desktops.  The screen becomes a long gallery with paintings on the walls that represent different tasks, and the user moves quickly and easily from one to another with a simple series of mouse and keyboard commands.  User studies were run to see whether the TaskGallery successfully engaged the knowledge and abilities people use to navigate physical space.  Study participants indicate that it seemed more natural to them to place tasks on the walls of the Gallery rather than on the floor or ceiling.  Users were also very adept at remembering the depth ordering of tasks.  This suggests that people feel compelled to treat the Gallery like a real-world environment, rather than just as an interesting desktop theme.  In general, users preferred the TaskGallery to the existing Windows interface.  The implication of this study on us is that the walls of a three-dimensional windows environment are the more appropriate places to display information.  Since Clockwise Win3D has a somewhat similar design, we did a user retention test at the end of each experiment to find out whether this is true.

As an overview, it seems that some previous studies indicate more than ever, that a three-dimensional interface design provides a more natural, realistic and intuitive manner of performing tasks, thereby improving user performance.  The objectiveness of these studies could however be clouded by the physical arrangements of the navigation structure.  When comparing two- and three-dimensional interfaces, it is important that the navigation complexity be as similar as possible.  Difference in navigation structure might as well be the underlying reason for superior performance instead of the window dimensions.  In our study, we are cautious about this possible bias.

Relevant psychological theories

User performance on different dimensional window environments is largely explainable if we have some understanding of user behavior.  As in one example, it could be touted that a three-dimensional window interfaces enables better user retention.  The possible reason is the object familiarity users experienced while using three-dimensional window interfaces.  In other words, better intuition enables better user performance in terms of user retention.

User performance in terms of speed of performing a task could similarly be explained psychologically. The OAI (Object Action Interface [Shneiderman]) model discusses the presentation of task and interface objects that are familiar to users.  The question, however, is whether a three-dimensional windows interface can fulfill the OAI model better than a two-dimensional windows interface.  Specifically, it voices back to the object familiarity.  More intuitive icons in three dimensions could provide a more object-oriented approach for users searching for a particular item.  Better encapsulation of relevant concepts is therefore necessary for such tasks.  Grouping internet relevant information into a three-dimensional "Internet Room" in ClockWise Win3D is a good example of an intuitive way of encapsulation.

Similar psychological arguments might be applicable to explaining error rates that users experience while using window interfaces.

Introduction and hypotheses

Figure 1: Starting screen for 2D experiment (click for larger image)

Figure 2: M-Media folder for 2D experiment (click for larger image)

Figure 3: Games folder for 2D experiment (click for larger image)

Figure 4: Office folder for 2D experiment (click for larger image)

Figure 5: Internet folder for 2D experiment (click for larger image)

Figure 6: Starting room for 3D experiment (click for larger image)

Figure 7: M-Media room for 3D experiment (click for larger image)

Figure 8: Games room for 3D experiment (click for larger image)

Figure 9: Office room for 3D experiment (click for larger image)

Figure 10: Internet room for 3D experiment (click for larger image)

This experiment compares the usability of a two-dimensional (Microsoft Windows Explorer) and a three-dimensional (Clockwise Win3D) desktop environment.  The figures above show the screen shots for the experiment setup for both 2D and 3D window environments.  The objective of the experiment includes understanding the following issues:

1. Compare the effect of spatial arrangement on both users' retention and ease of navigation.
2. Compare the amount of users' retention for both 2D and 3D window environments.
3. Compare the ease of navigation for both 2D and 3D window environments.

Another important issue could be (3) the ease with which the users can navigate.  Accordingly, the traditional 2D environment has a flat 2D tree/hierarchical navigation structure, as can be seen in Windows Explorer for an example.  On the other hand, a 3D window environment uses "rooms" as a logical component.  The issue therefore is which of the two designs improve the speed with which users can navigate.  Some possible arguments could include that since 2D environment is "flat" and items are clustered together, users could have more difficulty in distinguishing them at a glance and finding the item they are looking for.

Speed and error rate will be the measures in the experiment.  Speed is important for quantifying the ease of navigation and indirectly the users' retention.  We assume that the easier the navigation and the higher the users' retention, the faster it is for users to locate an item.  Error rate on the, other hand, is treated in the experiment as a secondary measure and will only be used to understand whether the aesthetic ecological effect of the 3D environment helps in identifying an item more accurately.

To create a fair experiment, there is a need for both windows environments to have similar layouts i.e. the only variance in the experiment should be the dimensions of the window environments.  Therefore, in the 2D window, there will be the exact same number of folders as in the 3D window's "rooms".  The items in each "room" in the 3D window will be replicated exactly in the corresponding folder in the 2D window.  This should ensure that there is only one variable in the experiment, which is the independent variable that the experiment is studying.

Tasks are generally to locate an item in both the 2D and 3D windows environment such as finding the solitare game or Excel spreadsheet.  The time with which the users perform the tasks will be recorded.  This is pertinent to the above-mentioned objectives since the speed of locating an item determines, in direct proportion, the ease of navigation and users' retention.  In addition, such tasks are also useful in determining the aesthetic effect of both window environments i.e. the ease with which the users can accurately determine the correct item.  There are a total of 12 tasks for each subject.

The independent variable is dimensions of the desktop environment and henceforth the experiment will be conducted in a 1x2 structure.  The dependent variables are:

1. The time taken by the subjects to complete the given task.  The time starts as soon as each subject is given the cue to start looking for the required items.  The time ends when the subjects finish all the tasks.
2. The error rates i.e. the number of wrong items that are selected by the subjects during the tasks.
3. Subjective satisfaction of the subjects.  This will be measured through the use of a survey that is given to the subjects at the end of the experiment.

Pilot study results

The pilot study was conducted with 2 subjects.  Both subjects did the 2D experiment and the 3D experiment.  The pilot experiments did not assess the user's retention, but we added it for the main experiment.

The pilot test also revealed that certain tasks to locate items which are placed in non-intuitive locations, produce a deterioration of performance.  This might be indicative (though not conclusive yet) that navigation structure is the underlying reason since in both 2D and 3D treatments, the subjects perform badly on these tasks.  There was also equipment limitation whereby the processing time of the tasks could be skewed due to CPU latency and internet related tasks being affected by network latency.

From the pilot experiments, some other observations were also made.  It was decided that there should be more tasks that require 3 or more mouse clicks and that the subjects should be required to start from the same location each time they perform a task.

Experimental subjects

The experiment was conducted using 24 subjects.  We selected these subjects from the undergraduate and graduate population of the University of Maryland, College Park.  The subjects varied from novice computer users to expert computer users, differentiated by the number of hours they use the computers in a week.  In order for the experiments to be fair, we used a similar distribution of novice to expert users in both window environments.

Due to the similarity in the navigation structure of both environments, it is necessary to use between subject experimentation.  This is because within subject experimentation would give subjects knowledge about the tasks they are required to perform in the second window environment.

Materials

Training materials that are provided to each subject before each experiment involved a number of dummy tasks.  In addition, the experimenter ran through the system features one by one with the subjects to familiarize them with the experiment platforms.  Before each experiment, each subject filled out a background survey form to identify their level of computer expertise and an experiment consent form.

We used a single PC running Windows 98.  All experiments are conducted on this machine to ensure that the system speed variable problem is contained.  The ClockWise Win3D system was installed and the information and tools provided in it were customized to provide a wide variety, ranging from Microsoft Office tools, games, and multimedia controls to internet information like weather and stock. We did not limit the training time for the subjects, instead we started the experiments when the subjects indicated that they were done with the training. There was also no finite time limit for each task. The subjects were given as much time as they needed to complete the tasks.

In addition, a two-dimensional window environment was setup on the same machine using the Windows Explorer.  This was set up to be exactly the same as ClockWise Win3D so that the only variance is the window dimensions.

After the subjects finished the tasks, they were required to fill in the subjective satisfaction form. nbsp; In addition, the subjects were required to fill out a user retention form that asked questions pertaining to their recollection of information presented in the systems.

Problems

One problem that was identified is that experiment machine was loaded with many processes, slowing the response time of the machine.  This may have influenced the results, as different subjects could face different response time and hence different user performance.

Another problem was with regard to the obscurity of ClockWise Win3D system.  Since this is not a very widely used software, many subjects have not seen it before.  On the other hand, the 2D experiment runs on familiar platforms (Windows Explorer on Windows 98).  This means that subjects might perform tasks faster in the 2D environment not because of the dimensions, but rather due to object familiarity.

Lastly, there was also the problem of subjects attempting to give up after unsuccessfully trying to perform a task for a considerable amount of time.  They were encouraged to carry on and hints were given to help them complete the tasks.  So this could also be another source of inaccuracies in the experiment results.

Figure 11: A look into the effect of navigation structure (2D)

Figure 12: A look into the effect of navigation structure (3D)

Figure 13: Comparing total number of errors for each subject

Figure 14: Comparing overall speed for each subject

Figure 15: Comparing overall speed between 2D and 3D

Figure 16: Comparing overall speed for each task (click for individual analysis)

Figure 17: Comparing overall error for each task

Results from the experiments are summarized in the charts above.  It is not difficult to see that 2D environment produces better speed for task completion.  This should be evident from figures 14, 15 and 16.  To explain this phenomenon, it is quite easily understood that since 2D experiments are all set up in a familiar environment (considering the wide spread use of Microsoft Windows Explorer), navigational familiarity played an important part.

T-test explanation

In this experiment, the T-test is performed on the two samples with unequal variance and two-tailed distribution.  A significance level alpha = 0.05 was used with 22 degrees of freedom.

Average data charts analysis

A T-test analysis of data showed statistically significant differences between the two treatments of window dimension in speed, but no statistically significant differences in accuracy factors.  An analysis of the average task completion time (speed) showed it to be significantly lower for the 2D treatment (M=25.92, SD=10.72) than the 3D treatment (M=37.43, SD=12.99), T-stat = -2.368, T-critical = 2.079, P = 0.027.  In the analysis of the average number of errors (accuracy) that subjects make to complete a task, subjects make less error in 2D environment (M=0.63) than in 3D environment (0.69), but without showing the statistically significant difference T-stat = -0.401, T-critical=2.080, P =0.693.

Total data charts analysis

A T-test analysis of data showed statistically significant differences between the two treatments of window dimension in speed, but no statistically significant differences in accuracy factors. An analysis of the total task completion time (speed) showed it to be significantly lower for the 2D treatment (M=311.00, SD=128.69) than the 3D treatment (M=449.17, SD=155.83), T-stat = -2.368, T-critical = 2.079, P = 0.027. As for the analysis of the total number of errors (accuracy) that subjects make to complete a task, showed no statistically significant difference between the 2D treatment and 3D treatment. The results were the same as above average data analysis.

Per task analysis

For the Excel, WinMine, Amazon and DJIA tasks, in figure 16, the T-test analysis of data showed statistically significant differences between the two treatments of window dimension in speed (task complete time).  The task completion times are significantly lower in the 2D environment than in the 3D environment.  However, for the CDPlay, Solitaire, NBA, Calculator, ICQ, WinZip, Display and RecentDoc tasks, in figure 16, the T-test analysis of data showed no statistically significant differences between the two treatments of window dimension in speed.  The raw data and T-test value can be found in Appendix Raw data.

Comparing these results to the difference in number of clicks between two environments, we believed that if users needed equal or fewer clicks in 2D environment than that in 3D environment, then it might be much easier to find a statistically significant difference between these two treatments in both speed and accuracy.  For example, in the WinMine and Amazon tasks, subjects only need to take two clicks in the 2D environment, but three clicks in the 3D environment. In this case, we always can find statistically significant differences between these two environments. However, we could not find a statistically significant difference between these two environments in the NBA task, in which subjects needed four clicks in the 2D environment, but only two clicks in the 3D environment. We believe that if the tasks took the same number of clicks to complete in both 2D and 3D environments, then we will find the task completion time (speed) are statistically lower in the 2D environment than in the 3D environment.

Time (seconds) vs. number of click(s) analysis

From the figure 11, the task complete time in 2D environment increases monotonously with the number of clicks. As we can see from the figure 12, in 3D environment, the task complete time still basically increases with the number of clicks, but not so obvious as increasing linearly in 2D environment. We believe that the time might increase more quickly than the increase in number of clicks, which means it would be more like increasing exponentially, instead of increasing linearly. 

Total time mean chart analysis

From figure 15, we see that subjects usually spend more time in the 3D environment than in the 2D environment for accomplishing a task.

User retention analysis

Figure 18: Comparing user retention

There is no statistically significant difference between the 2D and 3D environmentsts. But according to the raw data, for tasks such as finding a specific application or tool within a specific room, subjects on the 3D environment answered better than those on the 2D environment. Intuitively, it is because the visual icons and virtual desktop environment could help subjects memorize more easily with their environmental cues.  However, for tasks like finding the weather information or posted messages, the 2D subjects answered more correctly than the 3D ones.  This may be because in the 3D environment, subjects have to be close enough to see the weather information or posted messages.  The subjects on the 3D environment were not familiar with moving forward to enlarge the views.

Subjective satisfaction analysis

Figure 19: Comparing average of user rating for attraction and visual distraction

There is no statistically significant difference in terms of user's satisfaction.  The 3D environment attracted users at first sight because of new and fancy look. 

Figure 20: Comparing average of user rating for information arrangement and amount

The users' satisfaction rating towards the arrangement of information displayed on the screen in the 2D environment: the average is 5.75, and standard deviation is 2.83.  In the 3D environment, the average is 4.92, and standard deviation is 2.39.  There is no statistically significant difference between these scores. Most people are more used to Windows 2D environment than the Win3D so Win3D interface generated surprisingly good results. 

The users' satisfaction toward the amount of information displayed on the screen in the 2D environment: the average is 5.5, and standard deviation is 2.78.  In the 3D environment, the average is 5.75, and standard deviation is 1.81.  Again, there is no statistically significant difference between these scores.

In the 2D environment, all twelve subjects thought it was hard to find an item.  Two subjects thought the way to finding the way out of a folder was hard.  One subject thought the speed was too slow.  In the 3D environment, ten subjects among twelve thought it was hard to find an item.  Five subjects thought the speed was too slow here as well.  This shows that it might be easier to locate an item in the Win3D environment because of its direct manipulation, however because of the lack of resources used to run in the 3D environment, the speed tended to be much slower than the 2D environment, which led to the lower rating in speed.

The average users' satisfaction toward the initial instructions in the 2D environment was an 8.08 while it was a 7.17 with the 3D environment.  Although the training given to the subjects for both environments was the same, the Win3D environment still did not satisfy the needs of those who were taking the experiment in the 3D environment.  Many still felt awkward experiencing this new environment.

On the question "Would you consider using the pre-setup Windows Explorer instead of the regular Windows desktop? (2D)", eight subjects chose no but four subjects chose yes.  On the question "Would you consider purchasing a program such as ClockWise Win3D for your home PC to rearrange the desktop? (3D)", ten chose no but two choose yes.  Although the Win3D environment is appealing at first sight, it is too slow and requires a high performance computer to run on, as said by many of the subjects who took the test in the 3D environment.

Subjects' comments for 2D:

1. Better to include search and keyboard
2. Things are easier to find because they are placed in folders hat are descriptive
3. Some folders aren't placed properly, ex. The office folder had too many unrelated folders

1. Not used to the arrangement of information
2. Can't find an item quickly
3. To find an item is troublesome
4. Always find things not interested
5. Easy and attractive
6. The learning curve might be little
7. I am used to 2D and 2D is more efficient though for some people 3D might be easier to use

After testing 24 subjects, the authors became experts on using both 2D and 3D environments.  We practiced a few times before we did expert tests.  We took turns to record our best performance on twelve given tasks.  The detailed results of the expert tests could be found in Appendix Expert tests data.

Figure 21: Comparing best speed for experts and subjects (1/3)

Figure 22: Comparing best speed for experts and subjects (2/3)

Figure 23: Comparing best speed for experts and subjects (3/3)

Figures 21 to 23 list the shortest time to complete twelve given tasks for both experts and subjects on both 2D and 3D desktops.  In the majority of cases, the best expert beats the best subject by 40% of time usage.  This result supports the conjecture that familiarity was one of the main factors affecting speed to complete given tasks.  During the experiment, subjects wasted some time looking around folders or rooms.  However, experts planned a series of actions to locate the target.  When a subject took wrong actions, it took the subject a few seconds to correct them.  However, experts did not make mistakes in the tests.  Because there was no error in the expert tests, we could not compare the accuracy between experts and subjects.

The best expert used the same time to complete the tasks that require the same number of mouse clicks on both 2D and 3D, such as tasks RecentDoc, Excel, CdPlayer, Solitaire and Calculator.  Because the task NBA requires four mouse clicks and opens an external Internet Explorer on 2D environment, both experts and subjects responded faster on 3D than 2D.  Tasks WinMine, Amazon and Display requires moving actions on 3D environment, so they took both experts and subjects longer time to complete the tasks.

Figure 24: Best speed vs. number of click(s)

Best time usage vs. number of mouse clicks should be non-decreasing functions on both 2D and 3D.  However, the slope or the velocity on 3D environment should be higher than 2D environment.  For every number of mouse clicks, 3D cannot beat 2D on time usage.  If software designers can enhance the navigation methods in 3D environment, 3D might perform better than 2D on particular tasks.

Other analysis

An interesting observation can be seem from figure 11 and 12 whereby the number of clicks actually affect the speed with which a task can be performed.  Specifically, a task that requires more clicks requires a longer time to complete.  This is a simple idea but it also points out to us that navigational structure plays an important role in the user performance.

There is no statistically significant difference between 2D and 3D, in terms of accuracy. However, one merit about 3D is the attractive graphical environment. The intuitive environmental cues that can be found in 3D environment could be a worthy design guideline for 2D environment. From the understanding of how the navigational structure plays an important part in using the 3D environment due to better environment cues, we can see the importance of the navigational structure.

The question therefore remains: Which is more important to the user performance - dimensions of the window environment or navigational structure?  In addition, we could also ask: Does window dimensions really matter or is it actually the navigation ease that really is the underlying cause?

Impact for practitioners

The results from this study suggests that the familiarity of a software is important in achieving better user performance in terms of speed.  This is why a 2D environment outperforms a 3D environment in terms of speed due to widespread use of 2D window environments. Moreover, the results show that navigation ease is the key point to achieving better user performance and satisfaction.

The fact that the 2D is better in speed is also synchronous with better navigation ease since most users are better versed with the 2D environment and feel at ease with the associated navigation structure.  This leads to the conclusion that defining navigation ease would be instrumental in understanding the results.

Navigation ease is affected by 2 factors.  First, the amount of environmental cues is critical to users performance.  Providing better environmental cues as in 3D would lead users to a better navigational experience.  Second, familiarity with the navigational tools is equally important as evident in the 2D environment.  We therefore would recommend practitioners to consider merits of both 2D and 3D window environments and achieve a good balance.

Specifically, if software designers choose to use the 2D window environment, it is important that the designers should look into providing better environmental cues, so much so that borrowing design concepts from the 3D environments might perhaps be viable.  On the other hand, if designers choose to use the 3D objects in their design, it is important that the designers should strive to conform to current navigation tools that users are familiar with. Is navigation via moving a good design choice for ClockWise Win3D since it is not commonly used in most software?  In any case, extensive online training tools should be made available to create familiarity of navigation structures in users.

Suggestions for future researchers

Some possible future work could be:

1. Further study of the impact of window dimensions on user retention capabilities.
2. Further study of integrating the merits of 2D and 3D window environments.
3. Merits of using 3D graphical icons in a 2D window environment - the impact on user performance.
4. Investigate navigation methods in 3D window environment e.g. moving, jumping etc.

The earlier hypothesis suggests that the initial viewpoint is that the 2D is generally better than the 3D.  A refinement of this theory is warranted now that we have a better understanding.  Specifically, the impact of window dimensions on user performance lies in the navigational ease that can be achieved through different window dimensions.  Combinations of different features might be needed as a result.