An Empirical Evaluation of Undo Mechanisms

1
An Empirical Evaluation of Undo Mechanisms

• Aaron G. Cass
• Chris S.T. Fernandes
• Andrew Polidore
• Computer Science Dept.
• Union College
• Schenectady, NY, USA

2
Talk Outline

• Types of undo
• Problem statement
• Empirical evaluation design
• Pilot study
• Broad evaluation
• Results
• Future Work
• Conclusion

3
Linear Undo
4
Selective Undo

• Undo arbitrary action without requiring the
  undoing of subsequent steps Berlage 94

5
Selective Undo
Differing semantics

• Script model Archer, et al 84
• Removing action Ai results in state
• A1, A2, , Ai-1, Ai1, , An
• Cascading model Cass, Fernandes 05
• Accounts for dependencies between actions
• If Aj dependent on existence or result of Ai (j
  gt i), Aj also undone

6
Problem Statement

• Selective undo is believed to be superior to
  linear undo

• But
• Will it be easier for users to use?
• Is this a natural model for undo?
• Does it match users mental models?

7
Goal
Conduct a user study to find out!

• Compare three different models of undo
• Linear
• Script selective
• Cascade selective

8
The Study

• Want to determine users mental model of undo
• Design a task that elicits their mental model in
  the performance of that task

9
The Instrument

• Paper-based instrument
• Not associated with a computer application
• User freedom
• Selective undo not widely implemented

10
The Instrument
Steps taken to create this picture 1. Draw
Circle2. Draw Square3. Draw Triangle4. Color
Circle
Assume you have already done the steps above.
Draw what you think the outcome should be if you
were to undo step 4 (step 1).
11
Two Tasks
1. Draw Circle2. Draw Square3. Draw Triangle4.
Color Circle
12
Pilot Study

• Concern that word undo may trigger response
  based on past experience
• Compare with reverse the effects of
• 4 subjects

• Result kept undo in instrument wording

13
Evaluation Methods

• 29 subjects
• Undergraduate population
• Experience with widespread applications
• Little experience with specialized applications
• Within-subjects design
• Post-task questionnaire
• Single dependent variable
• ?2 analysis

14
Results (N28)
18 (64)
7 (25)
2 (7)
1 (4)
15
?2 Analysis

• Expected one model to be preferred
• plt0.05

• Expected cascade to be preferred over linear
• plt0.05

16
?2 Analysis

• Expected script to be preferred over linear
• ?2 not applicable
• Expected either script or cascade to be preferred
  over the other
• plt0.05

17
Conclusions

• In familiar applications
• Cascading selective is more natural than script
  selective
• Cascading selective is more natural than linear
• Script is more natural than linear(the data
  suggests)

18
Future Work

• Use a more complete application
• Pilot study with presentation software already
  complete
• Use a richer set of dependencies
• Implement selective undo in a representative
  application

19
Validity threats

• Validity threats
• Scalability
• To complex tasks
• To other applications