Quantitative Analysis of Scrolling Techniques

1. Ken Hinckley Edward Cutrell Steve Bathiche Tim Muss Microsoft Research & Microsoft Hardware April 23, 2002 Quantitative Analysis of Scrolling Techniques

2. Motivating Questions: Product • Multi-Channel scrolling devices (1) save time to grab scrollbar (2) maintain visual focus on primary task • Can performance of the scrolling wheel be improved? • How does it compare? • How to evaluate & test our new scrolling products in general?

3. Accelerated Scrolling Wheel • Scroll further when you roll faster • Extend range of wheel • But… Is it really any better? • Possible loss of control / precision?

4. Motivating Questions: Research • How should one experimentally evaluate scrolling performance? Distance & Precision? • Which is fastest: blue or green (dotted) ?

5. D W Fitts’ Law: Cough Syrup for Input Devices • Widely used to study rapid, aimed movements (Fitts 1954) • Used in pointing device studies since 1978 • Task: Point at a target W wide at distance D • The Law: • MT = a + b log2(D/W + 1) • a, b fit by linear regression using observed MT • Never applied to scrolling

6. Frame Target line Scrolling Experimentfounded in Fitts’ Law • Scroll back & forth between 2 lines in a doc • Ex: comparing paragraphs • EachTrial had at least 10 Phasesof individual scrolling movements

7. Experimental Design • Device X D X W • ScrollPoint • Standard Wheel • 3 lines/notch • Accel. W1 • 1 line/notch • Accel. W3 • 3 lines/notch • nonsensical D X W’s • e.g. D=6, W=18

8. Note on Practice Effects • First 2 phases of each trial eliminated due to start-up effects

9. Results: Average Movement Times • Overall average Movement Time (MT) • ScrollPoint & Std. Wheel do not differ significantly • But what if we control for D/W ?

10. Results: By Distance (Raw Data) • Hard to see what’s going on in raw MT data • D: 3.5 cm to 2.25 meters; So MT also has wide range

11. Results: By Distance (Scaled) • Significant crossover interaction by Distance! • Std Wheel faster at small D, ScrollPoint for large D • Accel mappings improve performance

12. Results: By Width • No interesting crossover effects for Device X W • A faster device is faster across all W

13. … and Fitts’ Law Describes our Data • r ≥ 0.90 for all devices

14. Representative Tasks for Scrolling • We experimented with several tasks: • Scrolling while proofreading text for misspellings • Searching for highlighted line in document • [and following the link – Zhai] • Searching for highlighted target word in document, in presence of highlighted distracter words • Fitts’ task sensitive to subtle device diffs • Cognitive & visual search issues ignored

15. Design Insights • No one device or acceleration setting is “best” • Accel W1 vs. Std Wheel: faster + better resolution • Is it possible to combine Accel W1 / Accel W3 mappings to have optimal performance?

16. Qualitative Results • ScrollPoint: Most Ss preferred for long D • But in practice many would “just grab scrollbar” • “very ineffective in targeting lines” • “my hand didn’t get tired” • Standard Wheel: moved predictably • Fatigue / comfort frequent negative comment • Accel W3: “very easy to scroll long distances”, but most Ss disliked larger notches • Accel W1: liked finer notches, but still “tedious to scroll long distances”

17. Naturally Occurring Behaviors with the Wheel • How do users roll the wheel? • (1) trying to get somewhere fast, or • (2) reading fast reading

18. How Acceleration Works • Roll faster  move further • But do not change “reading” experience • For Δt < 0.1 notch/s Δy =K1(1+K2Δt)α • Otherwise Δy = 1 line • The user does nothave to learn anything new!

19. ProductVersion • To play with Accel. Scrolling, download IntelliPoint 4.0 • Differs slightly • e.g. no fractional lines • http://www.microsoft.com/ hardware/mouse/download.asp

20. Future Work • Apply Fitts approach to a scroll/select task • Scroll, then click on object of varying W • Two-handed scrolling: Current experiment can compare right- vs. left-handed devices, but not higher level benefits of 2h scrolling, e.g. • Anticipatory cursor motion • Avoid fatigue from single hand doing everything • Scrollbar: cost of moving mouse back and forth to scrollbar needs to be considered • More scrolling expt’s needed with Fitts’ Law

21. Thank You! • Questions? • kenh@microsoft.com • http://www.microsoft.com/ hardware/mouse/download.asp

23. Extra slides for questions etc.

24. Jellinek & Card 1991 • Gain theoretically does not affect performance • MT = a + b log2(D/W + 1); gD/gW = D/W • Observed MT almost unchanged for g = 1  10 • g Reduces footprint of device & reclutching • On Wheel, reduced footprint = faster MT

25. Principles of Bimanual Action • Yves Guiard, 1987. For right-handers: • Right-to-left reference: Action of the right hand occurs within the frame-of-reference defined by the left. • Scale Asymmetry: Movements of the right hand occur at higher spatial and temporal frequencies than the left • Left-hand Precedence: Action starts with the left hand.

26. Principles of (Bimanual) Scrolling • Scrolling is a background task that should be assigned to the nonpreferred hand. • Right-to-left reference: Movement of mouse cursor is within current document view. • Scale Asymmetry: Scrolling is a coarse task, cursor movement & selection are high-precision • Left-hand Precedence: Scrolling precedes detailed activity in the document. (MacKenzie 1998)

27. No switching between pointing & scrolling Overlapped action of the 2 hands Maintain visual focus & concentration on work Buxton & Myers 1986bimanual scrolling ~25% faster than scroll bar Bimanual Scrolling

28. Bimanual Controlon Office Kbd • Navigation controls on left • Scrolling [wide wheel] • Web [Forward / Back] • Windows [AppToggle] • Cut, Copy, Paste also well suited to left side • Compound selection [or placement of IP] + articulation of command

29.  OVERDRIVE • “Automatic transmission” for the wheel (+accel.) • Evaluating… Informally, seems to work great! • in a ~10 pg doc: IntelliPoint 4.0!  IP 5.0 (?)  • All of these have 1 line/notch precision