CS-TR-2268
`CAR-TR-450
`
`Revised January 1989
`
`High Precision Touchscreens:
`Design Strategies and Comparisons with a Mouse
`
`Andrew Sears
`Ben Shneiderman *
`
`Department of Computer Science
`Human-Computer Interaction Laboratory
`University of Maryland
`College Park, MD 20742
`
`Abstract
`
`Three studies were conducted comparing speed of performance. error rates. and user
`preference ratings for three selection devices. The devices tested were a
`touchscreen, a touchscreen with stabilization (stabilization software filters and
`smooths raw data from hardware), and a mouse. The task was the selection of
`rectangular targets 1, 4, 16, and 32 pixels per side (0.4x0.6, 1.7x2.2, 6.9x9.0,
`13.8x17.9 mm respectively). Touchscreen users were able to point at single pixel
`targets, thereby countering widespread expectations of poor touchscreen resolution.
`The results show no difference in performance between the mouse and touchscreen
`for targets ranging from 32 to 4 pixels per side.
`In addition, stabilization
`significantly reduced the error rates for the touchscreen when selecting small targets.
`These results imply that touchscreens, when properly used, have attractive
`advantages in selecting targets as small as 4 pixels per size (approximately one-
`quaner of the size of a single character). A variant of Fitts' Law is proposed to
`predict touchscreen pointing times. Ideas for future research are also presented.
`
`* Address correspondence to Ben Shneiderman
`To appear in the International Journal of Man Machine Studies
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018—00025
`Page 1 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 1 of 22
`
`

`

`Int. J. Man—Machine Studies (1991) 34, 593—6”
`
`,
`
`High precision touchscreens: design strategies and
`comparisons with a mouse
`
`ANDREW SEARS AND BEN SHNEIDERMAN
`Department of Computer Science and Human -C0mputer Interaction Laboratory,
`University of Maryland. College Park, MD 20742 USA
`
`(Received 4 August 1989 and accepted in revised form 29 January 1990)
`
`Three studies were conducted comparing speed of performance, error rates and user
`preference ratings for three selection devices. The devices tested were a touch-
`screen, a touchscreen with stabilization (stabilization software filters and smooths
`raw data from hardware), and a mouse. The task was the selection of rectangular
`targets 1, 4, 16 and 32 pixels per side (04 x 0-6, 1-7 x 22, 6‘9 x 9-0, 13-8 x 17-9 mm
`respectively). Touchscreen users were able to point at single pixel targets, thereby
`countering widespread expectations of poor touchscreen resolution. The results
`show no difference in performance between the mouse and touchscreen for targets
`ranging from 32 to 4 pixels per side. In addition, stabilization significantly reduced
`the error rates for the touchscreen when selecting small targets. These results imply
`that
`touchscreens, when properly used, have attractive advantages in selecting
`targets as small as 4 pixels per size (approximately one—quarter of the size of a single
`character). A variant of Fitts‘ Law is proposed to predict touchscreen pointing times.
`Ideas for future research are also presented.
`
`Introduction
`OVERVIEW
`Many pointing devices are available for use with computers, but none are as natural
`to use as the touchscreen. Pointing at an item or touching it, is one of the most
`natural ways
`to select
`it. Touchscreens allow the software designer
`to take
`advantage of this convenient selection method by having the users simply touch the
`item they are interested in.
`Touchscreens are easy to learn to use, require no additional work space, have no
`moving parts, and are very durable (Pickering, 1986; Shneiderman, 1987; Stone,
`1987; Muratore, 1987; Potter, Weldon & Shneiderman, 1988). Durability has made
`touchscreens popular in many applications, including kiosks at airports, shopping
`malls, amusement parks and home automation. Even with these positive features,
`the touchscreen’s reputation for a lack of precision, high error rates, arm fatigue,
`and smudging the screen have resulted in limited use (Pickering, 1986; Shneider-
`man, 1987). Current touchscreen implementations do not include tasks requiring
`high resolution or tasks that are performed by frequent or experienced users. An
`adequate reduction in error rates, combined with the speed of the touchscreen may
`help expand this relatively limited use.
`
`PREVIOUS EXPERIMENTS
`Many studies have compared touchscreens with other selection devices for various
`tasks. Our summary motivates our experiments. First, studies that compared the
`593
`
`0020-7373/91/040593 + 213010010
`
`© 1991 Academic Press Limited
`
`
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018-00025
`Page 2 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 2 of 22
`
`

`

`I
`
`594
`
`A. SEARS AND a. SHNEIDERMAN
`
`touchscreen with other selection devices are reviewed. Then several studies that
`explored the use of alternative selection strategies. an error reduction method we
`employ, will also be summarized.
`Muratore (1987) did an extensive literature survey, reviewing 14 studies that
`compared various cursor control devices. Her interpretation of these results implies
`that. the touchscreen was the fastest but least accurate of the devices studied. Hall,
`Cunningham, Roache and Cox (1988) investigated the effects of various factors on
`touchscreen performance. The display was an IBM InfoWindow color terminal with
`a piezoelectric touchscreen using the land-on selection strategy forcing the selection
`at the location of the initial touch. Feedback was not provided about the accuracy of
`selections. They reported that accuracy varied from 66-7% for targets 10 mm per
`side, to 99-2% for targets 26 mm per side, and that accuracy was maximized once
`targets were approximately 26 mm per side. Ostroff and Shneidennan (1988)
`compared a touchscreen, mouse, number keys and arrow keys. The touchscreen was
`a Carroll Touch infra—red touchscreen using the land-on strategy. The study involved
`selecting words from an interactive encyclopedia (Hypertiesm).‘i‘ The results were
`similar to those of most other studies comparing the touchscreen and the mouse,
`indicating that the touchscreen was faster. They found no significant difference
`between error rates for the mouse and the touchscreen. This finding may be due in
`part to the relatively large size of the targets used and the rapid but awkward form
`of the jump mouse. (A jump mouse moves the cursor from one target to the next,
`skipping the space between them.) Ahlstrom and Lenman (1987) compared a
`conductive touchscreen using the land-on strategy and mouse for the selection of a
`six character word from a list of words. This study indicated that the touchscreen
`was faster. but
`resulted in much higher error
`rates. Karat, McDonald and
`Anderson (1986) compared a touchscreen, mouse and keyboard for selection tasks.
`The touchscreen used was an Elographics analog membrane touchscreen using the
`land-on strategy. The task involved selecting items from a menu in a calendar
`program and a telephone directory. Some tasks also involved a typing sub-task. The
`results indicated that the touchscreen was the preferred device for the task without
`the typing sub-task, while the keyboard was preferred when the sub-task was
`included. The touchscreen was the fastest for both tasks.
`
`These studies have been limited to relatively large targets for selection tasks, but
`they do give some insight into the potential use of touchscreens. It is clear that a
`touchscreen can be used for rapidly selecting relatively large targets. Unfortunately,
`most of these studies also indicate that error rates were significantly higher for
`touchscreens. There are two explanations that may account for the majority of these
`errors,
`the inability of the touchscreens used in these studies to provide precise
`information about the location of a touch, and inadequate selection strategies for
`the tasks studied.
`
`The inability of the touchscreen hardware to provide precise information may be
`due to a lack of resolution or the result of multiple pixel locations. possibly as many
`as 20 or more, being returned for a touch in a single location. While research by
`touchscreen manufacturers has dramatically increased the resolution of touchscreens,
`the problem of returning multiple pixel
`locations for a single touch remains.
`
`t Hyperties is a trademark of Cogneties Corporation.
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018—00025
`Page 3 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 3 of 22
`
`

`

`HIGH PRECISION TOUCHSCREENS
`
`595
`
`The extent to which this is a problem depends on both the touchscreen technology
`and manufacturer. Carroll Touch has published a Touch Handbook which provides
`a brief review of current touchscreen technologies including resolution, response
`time, and environmental resistance (Carroll Touch, 1989). Stabilization of the
`touchscreen will allow a single touch to result in the selection of a single pixel,
`possibly resulting in a significant reduction in errors, primarily for small targets.
`Ideally stabilization would be accomplished at the hardware level, but canyalso be
`done in software. Our studies will use software stabilization to filter and smooth raw
`
`data from the touchscreen hardware. Stabilization is an important idea that can be
`applied to many technologies including touchscreens, data gloves and light pens, but
`has never been tested with touchscreens.
`
`Many alternative selection strategies have been suggested to help reduce errors
`including take-off, first-contact, land-on, and others requiring a second touch. The
`land-on strategy uses the location of the initial touch for the selection. If the initial
`touch corresponds to a selectable region,
`that region is selected, otherwise no
`selection is made. The first-contact strategy results in the selection of the first
`selectable region the finger comes into contact with. With this strategy the users
`move their fingers on the screen until a selectable region is touched, this region is
`then selected and the appropriate process is initiated. Once again, all additional
`contact is ignored until the finger is removed from the screen. The take-off strategy
`allows users to place their fingers on the screen and move to the desired region on
`the screen before a selection is made. A cursor is placed slightly above the users
`fingers when they touch the screen indicating the exact location of where a selection
`would be made. Users can then drag the cursor to the desired region, and lift their
`fingers frOm the screen to select it. A selection is made only if there is a selectable
`region under the cursor when users lift their fingers.
`Several studies have been conducted to compare alternate selection strategies.
`The results indicate that some strategies may be promising for a wide range of tasks,
`and a significant reduction in error rates is possible (Murphy, 1987; Potter et al.,
`1988; Potter, Berman & Shneiderman, 1989). Murphy (1987) compared seven
`selection strategies. He conducted an experiment that involved selecting targets that
`were 19 mm2 from a matrix of 60 targets. His results indicated few significant
`differences among the selection strategies, making it difficult to promote any single
`strategy as the best with respect to either selection time or error rates for this target
`srze.
`
`Researchers at the University of Maryland Human—Computer Interaction Labo-
`ratory have performed two experiments comparing the land-on, first contact and
`take-01f strategies. The first experiment involved the selection of a two character
`state abbreviation from a S x 10 matrix. This study indicated that the first-contact
`strategy was the fastest, while the take-off strategy produced the fewest errors. The
`second experiment
`involved the traversal of a hypertext database by selecting
`highlighted words. There were no significant differences in the time needed to
`perform the task, while the first-contact and take-off strategies produced fewer
`errors than land-on (Potter et 01., 1988; Potter er al., 1989).
`selection
`These experiments
`indicate that
`first-contact may be the fastest
`strategy, while the results pertaining to error rates did not consistently favor one
`strategy over the others. While these studies do provide a comparison of the
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018-00025
`Page 4 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 4 of 22
`
`

`

`5%
`
`A. SEARS AND a SHNEII)ERMAN
`
`indicate how well a touchscreen using these
`they do not
`selection strategies.
`strategies will perform compared with other selection devices.
`Some researchers have claimed that the current touchscreen technology would not
`allow high-resolution selection, saying that selection of a single character with a
`touchscreen would be slow even if it were possible (Sherr, 1988; Greenstein &
`Arnaut, 1988). Others have blamed the size of the human finger for the lack of
`precision, claiming that the size of the user’s finger limits the size of selectable
`regions (Beringer, 1985; Sherr, 1988; Greenstein & Arnaut, 1988). Previous studies
`have made no attempt at evaluating a touchscreen for high resolution tasks,
`restricting targets to relatively large sizes ranging from a square that is 6-4 mm per
`side, to targets that were approximately 25.4 x 40-6 mm. In addition, many of these
`studies have indicated that touchscreens result in significantly higher error rates than
`many other selection devices, including the mouse. Our experiments studied the
`selection of small targets with the touchscreen as compared with the mouse. We also
`studied the effects that stabilization and the use of an alternative selection strategy
`have on these selections. Error rates and selection speed were measured. User
`preference data were also collected.
`
`Experiment one: stabilized touchscreen, non-stabilized touchscreen,
`and mouse
`INTRODUCTION
`
`The main purpose of the first experiment was to provide the comparison of a
`touchscreen with a mouse, using an improved selection strategy for high resolution
`tasks. The secondary purpose was to investigate the effect stabilization has on speed
`of performance. error rates and user preference for selection tasks when using a
`touchscreen.
`
`Due to the difficulty involved in modifying hardware, stabilization was accompl-
`ished using software that
`filters and smooths raw data from the touchscreen
`hardware. These results should generalize to stabilization performed by either
`hardware or software.
`
`The first step was to determine which selection strategy should be tested. To do
`this, we must understand the requirements of the task being evaluated. A typical
`high resolution task may be the selection of the start and stop points for a line in a
`graphics package, or possibly the selection of a character in a word processing
`program. Since it is difficult to touch a single character accurately, let alone a single
`pixel on the first attempt, the land-on strategy is not adequate. In addition, many
`high resolution tasks involve the selection of targets that are not defined before the
`selection is made, such as the starting point of a line which makes the first-contact
`strategy inappropriate. On the other hand, the take-off strategy provides continuous
`feedback about cursor location, allowing the user to position the cursor before a
`selection is made by lifting the finger. This makes take-off the best candidate for
`many high resolution tasks.
`
`PILOT STUDY RESULTS
`
`A pilot study helped determine that the original target sizes (16. 8, 4 and 2 pixels
`per side) were inappropriate. We decided that a larger range of target sizes would
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018-00025
`Page 5 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 5 of 22
`
`

`

`HIGH PRECISION TOUCHSCREENS
`
`597
`
`make the trends in selection time and error rate data more apparent. The addition
`of a single pixel target allows predictions to be made about the selection of the
`smallest possible target we could represent.
`
`SUBJECTS
`
`Thirty-six subjects volunteered from the Psychology Department subject pool of the
`University of Maryland. The amount of computer experience the subjects had was
`not controlled. Three subjects had used a touchscreen one time. while the remaining
`subjects had no experience. Experience using a mouse ranged from none to every
`day, with the majority of the subjects using the mouse very infrequently.
`
`EQUIPMENT
`
`All tasks were performed on an IBM PC—AT with an IBM Enhanced Color Display
`and a MicroTouch touchscreen. The monitor was placed on the desk in the normal
`monitor position with the keyboard placed in front of it. The monitor measured
`27-6 x 19-5 cm and was used in EGA mode (640 X 350 pixels) resulting in pixels that
`were 0-4 X 06 mm. The MicroTouch touchscreen is a capacitive touchscreen that
`provides continuous information about the location of a touch on a 1024 X 1024 grid.
`It requires only a light
`touch to be activated and averages the location of all
`simultaneous touches and returns a centroid location. The touchscreen was cleaned
`
`once before the first subject began the experiment, and was not cleaned again until
`the last subject had completed the experiment. Software was written to convert the
`touchscreen coordinates to pixel coordinates and to stabilize the resulting pixel
`coordinates. A MouseSystems Optical PC-Mouse with three buttons was used with a
`mouse pad that measured 22-9 X 197 cm. The mouse was calibrated so that a single
`pass horizontally on the pad resulted in the cursor moving the width of the screen,
`and a single pass vertically on the pad resulted in the cursor moving the height of the
`screen. Users were free to place the mouse pad anywhere they wanted.
`After the experiment was completed it was discovered that the software provided
`with the mouse only allowed the cursor to be moved in two pixel
`increments
`horizontally. This did not impair the selection of targets, however, the resolution of
`the screen for the mouse tasks was essentially half (320 x 350) that of the screen for
`tasks using the touchscreen, possibly influencing the results in favor of the mouse.
`New mouse software was obtained for the second and third experiments to correct
`these problems.
`
`Stabilization software
`Stabilization allowed a touch to result in a single pixel coordinate. The first attempt
`at stabilizing the touchscreen used running-means of the last 20 x and y coordinates.
`Although stability was improved dramatically, the selection of a single pixel was still
`not reliable. and the cursor lagged far behind the user’s finger. Several additional
`steps were necessary to solve these problems.
`First, a small
`region (0-9 X 1-7 mm) around the current cursor location was
`deactivated, requiring the user‘s finger to move beyond this region before the cursor
`moves (Figure 1, Region A). The second step was to define a larger region
`(8-6X 16-8 mm) around the cursor that resulted in a movement that was only a
`fraction of the actual distance between the cursor and finger locations (Figure 1,
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018—00025
`Page 6 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 6 of 22
`
`

`

`598
`
`A. SEARS AND B. SHNEIDERMAN
`
`Current cursor
`
`location
`
`Region A
`
`4—_— Region B
`
`
`
`Region C includes all the
`space outside of Region 3
`
`FIGURE 1. Regions defined for stabilizing the touchscreen. If the current touch is within Region A, the
`cursor does not move. If the touch is in Region B, then the cursor moves a percentage of the distance
`between the current touch and the current cursor position. if the current touch is in Region C, then the
`cursor moves to the location of the current touch.
`
`Region B). For instance, if the user‘s finger was at point X (Figure 1), the cursor
`would only move to point Y. In this way, it was possible to perform very precise
`movements by dragging a finger on the screen.
`The steps allowed the selection of a single pixel, but resulted in a significant delay
`between a movement of the finger and the movement of the cursor. One additional
`step was necessary to eliminatethis delay. Whenever the location of the current
`touch was far enough from the current cursor location, the cursor moved directly to
`the location of the touch (Figure 1, Region C). In this way, the cursor could be
`dragged across the screen very rapidly without a significant distance between the
`cursor and the user’s finger.
`Although it may appear that stabilization will lead to a loss of directness between
`the movement of the finger and the movement of the cursor, careful manipulation of
`the size of Regions A and B allows stabilization without a loss of directness.
`
`DESIGN AND PROCEDURE
`
`Selection device and target size were within subject variables. There were three
`selection devices, a mouse, a non-stabilized touchscreen, and a stabilized touch-
`screen. There were four rectangular targets: 1, 4, 16 and 32 pixels per side
`(0-4 x 0-6, 1-7 x 2-2, 6-9 x 9.0, 13-8 X 17-9 mm respectively). The four pixel target
`was approximately one-quarter of the size of a character which is 9 X 7 pixels. With
`this range of target sizes the results will be applicable to many practical tasks.
`Each subject was tested with all selection devices and target sizes, resulting in
`three groups of four tasks for each subject. Each task required the selection of a
`series of six targets that were presented on the screen. Targets appeared in one of
`four positions, about 2-5 cm from each corner of the screen (Figure 2). Each subject
`had one practice trial for each task.
`Selection device was held constant in each group of tasks, and target size was held
`constant within a task. Within each group of tasks the target size decreased, in order
`from 32 pixels per side down to a single pixel. We chose to provide decreasing target
`sizes to facilitate the subjects’ skill acquisition as they moved to smaller and more
`difficult targets. We recognized the disadvantages of non-random ordering, but we
`felt the additional experience was important. Each subject performed a set of
`selection tasks similar to the following list.
`
`
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018—00025
`Page 7 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 7 of 22
`
`

`

`599
`
`mou PRECISION Tournscaems
`________________.____——————
`Task number
`Device
`Target size (pixels per side)
`___________————————-
`l
`Non-stable touchscreen
`32
`2
`16
`3
`4
`4
`1
`5
`32
`6
`16
`7
`4
`8
`1
`
`Mouse
`
`Stable touchscreen
`
`9
`10
`11
`12
`
`32
`16
`4
`1
`
`The order in which devices were used was randomized among subjects to prevent
`any possible bias. The order that the six targets within each task were presented
`was also varied to prevent subjects from anticipating the correct location for the
`next target.
`Instructions were presented on the computer screen. Before each task a short
`message was presented telling the subject which device would be used. Subjects then
`pressed ENTER to begin the task. A target was presented and subjects had to select
`it with the appropriate device. When the target was successfully selected, or five
`errors were made on the current target, a tone sounded and the next target was
`presented. An error occurred each time subjects lifted their fingers without making
`a successful selection. A maximum of five errors was allowed per target to prevent
`subjects from getting stuck indefinitely on a target if they were not able to select it.
`Six targets were presented for each task. After the sixth target was selected, a
`message indicating the number of errors and time taken was presented. Subjects
`then pressed ENTER to continue to the next task.
`When using the mouse, selections were made by moving the mouse until the
`center of the cursor was on the target and clicking any of the mouse buttons.
`Selection using both touchscreens, involved touching the computer screen, dragging
`the cursor until the center of the cursor was on the target, and then lifting the finger
`from the screen. In all cases the cursor was a plus sign (+), made by five pixels
`
`
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018—00025
`Page 8 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 8 of 22
`
`

`

`(100
`
`A. SEARS AND H, SHNEIDERMAN
`
`TABLE I
`
`Mean selecrion lime (in seconds) per target (SD.
`in parentheses)
`
`Target size (pixels per side)
`16
`4
`
`32
`
`1
`
`3- 13
`(1-28)
`1 -83
`(0-37)
`1-86
`(045)
`
`3-47‘i~
`(1 -60)
`1-98
`(0-33)
`1-93
`(0-47)
`
`4-97
`(1-98)
`4-27
`(1-27)
`4-57
`(1 ~65)
`
`6-081
`(1 -87)
`11 -78
`(4-42)
`12-28
`(4-95)
`
`Mouse
`
`Stabilized
`touchscreen
`Non-stabilized
`touchscreen
`
`1 p = <o-os
`
`vertically, and five pixels horizontally and was presented approximately 6 mm above
`the subject’s finger to allow the subject to view both the cursor and target when
`selecting small targets. The cursor was blue and targets were red; when the cursor
`and target overlapped, the intersection became white making it easier to know when
`the cursor was correctly positioned.
`The time to select each group of six targets and the number of errors per group
`were recorded for each task.
`In additon, subjects were asked to indicate their
`preference for each device on a scale from 1 to 9 (1 being strongly disliked, 9 being
`strongly liked). All data were recorded on the computer.
`
`RESULTS
`
`Selection times
`
`The mean time from the initial presentation of a target until either successful
`selection or until five errors occurred appear with standard deviations in Table 1 and
`
`(seconds)
`
`
`
`Selectiontime
`
`FIGURE 3. Selection time for four target sizes and three selection devices. key: -~. non-stabilized
`touchscreen; ----. Stabilized touchscreen; —. Mouse.
`
`Target size (pixels per side)
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018—00025
`Page 9 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 9 of 22
`
`

`

`HIGH PRECSION TOUCHSCREENS
`
`60]
`
`TABLE 2
`
`Mean number of errors per large! (5.0. in parentheses)
`
`Target size (pixels per side)
` 32 16 4 1
`
`
`
`
`
`Mouse
`
`0-50
`0-08
`0-06
`0-08
`(0-68)
`(0-18)
`(0-12)
`(0-15)
`153'?
`0-35
`0-05
`0-03
`Stabilized
`(1 ~08)
`(0-58)
`(0- 10)
`(0.06)
`touchscreen
`4-381'
`0771’
`006
`0-02
`Non-stabilized
`
`touchscreen (0-62) (0-06) (0' 15) (0-60)
`
`
`
`
`‘l‘ p = <0-05
`
`the means are plotted in Figure 3. An ANOVA with repeated measures for
`selection device and target size showed significant main effects for selection device,
`F(2,70) =5-0, p = <00], and target size, F(3, 105) =232-5, p = <0-001. A sig-
`nificant
`interaction between selection device and target size. F(6.210)=50-0,
`p = <0-001, was also found. Tukey’s post hoc HSD test showed that both
`touchscreens are faster than a mouse for targets 16 pixels per side (p = <0-05), and
`the mouse is faster than both touchscreens for a single pixel (p = <0'05). There
`were no other significant differences across the devices.
`
`Error rates
`
`The mean error rate per target and standard deviations appear in Table 2 and the
`means are plotted in Figure 4. An ANOVA with repeated measures for selection
`device and target
`size showed significant main effects
`for selection device,
`
`target
`
`Errorsper
`
`32
`
`16
`
`4
`
`1
`
`Target size (pixels per side)
`
`FIGURE 4. Error
`
`target sizes and three selection devices. key: —--. non-stabilized
`four
`rates for
`touchscreen; -‘-. stabilized touchscreen; —, mouse.
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018—00025
`Page 10 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 10 of 22
`
`

`

`602
`
`A. SEARS AND a. SHNEIDERMAN
`
`TABLE 3
`
`Mean user preference ratings for three selection devices (5.0.
`in parentheses)
`_________——-————-——-——
`
`Non-stabilized
`Stabilized
`touchscreen
`touchscreen
`Mouse
`________._____.—_—
`
`1-9'l'
`6-7
`7-5
`Mean user
`preference rating
`(1 7)
`(1-9)
`(1 -S)
`____—.————_
`
`Tp = <0‘05
`
`F(2, 70) = 186-4, p = <0-001, and target size, F(3, 105) = 356-6, p = <O-001. A
`significant interaction between selection device and target size, F(6, 210) = 177-44,
`p=<0~001, was also found. Tukey’s post hoc HSD test showed that the non-
`stabilized touchscreen resulted in more errors than either of the other devices for
`the 4 x 4 pixel target (p = <0-05). For the single pixel target, the mouse resulted in
`fewer errors than either of the other devices, and the stabilized touchscreen resulted
`in fewer errors than the non—stabilized touchscreen (p = <0-05).
`
`User preference
`User preference means and standard deviations appear in Table 3. A one-way
`repeated measures ANOVA on selection device showed an effect for selection
`device, F(2, 70) = 106-9, p = <0~001. Tukey’s post hoc HSD test showed that the
`non-stabilized touchscreen received lower preference ratings than either of the other
`devices (p = <0-05).
`
`DISCUSSION
`
`The stabilized touchscreen was as fast or faster than the mouse while making no
`more errors for targets as small as four pixels per side. This indicates that a
`touchscreen can be used for selection of single characters, which are 9 X 7 pixels, in
`many applications that currently use a mouse. There were no differences in selection
`times between the two touchscreen implementations, and stabilization resulted in a
`significant reduction in errors for the two smaller targets.
`The results indicate that it was possible to select a single pixel with a touchscreen
`although the mouse resulted in faster, more accurate selections than either
`touchscreen. The significant increase in selection time and error rates from the four
`pixel
`targets to the single pixel
`indicates that none of the selection devices, as
`currently implemented, are appropriate for the selection of the single pixel targets.
`Additional work must be done to improve the input devices if they are to be used
`for selecting single pixel targets without a zooming feature.
`User preference ratings indicate that the stabilized touchscreen was preferred over
`the non-stabilized touchscreen. Since subjects are using their fingers to move the
`cursor on the screen.
`it seems reasonable to expect
`the cursor movements to
`correspond directly to movements of their fingers. When using the non-stabilized
`touchscreen,
`the jitter caused by the lack of stability violates this expectation,
`possibly resulting in lower preference ratings. When stabilization is added,
`the
`cursor tracks their fingers accurately, resulting in both higher preference ratings and
`lower error rates.
`
`
`
`Microsoft Ex. 1012
`
`Microsoft v. Philips -
`
`|PR2018-00025
`Page 11 of 22
`
`Microsoft Ex. 1012
`Microsoft v. Philips - IPR2018-00025
`Page 11 of 22
`
`

`

`HIGH PRECISION TOUCHSC‘REENS
`
`603
`
`The targets explored in this experiment allow predictions to he made about a wide
`range of practical target sizes. Considering that the majority of high resolution tasks
`are performed by experienced users, studies that
`include additional practice, or
`instructions for selecting small targets, may prove useful. Several subjects devised
`strategies for selecting the single pixel targets. Two subjects learned that they could
`position the cursor near the target and then simply roll their fingers up and down or
`left and right to make fine manipulations. Subjects that were observed using this
`strategy on the stabilized touchscreen had a mean error rate of only 0-25 when
`selecting six single pixel targets. When this mean is compared with the overall mean
`of 1-53, it becomes apparent that this strategy is very successful in reducing errors. If
`all subjects were exposed to this method of selecting small targets, the error rates
`might decrease. The second and third experiments incorporated this idea, presenting
`brief instructions to subjects before the experiment.
`Although many people have claimed that smearing will be a significant problem
`when using touchscreens this problem did not occur in the office-like conditions of
`this experiment. The t0uchscreen surface is lightly ground, rather than polished,
`thereby reducing the glare and impact of fingerprints. The touchscreen used for this
`study was cleaned once before the experiment began and was not cleaned again.
`Small amounts of oil and dust accumulated on the screen but the accumulation was
`similar to that on standard monitors. Actually, less dust appears to collect on the
`touchscreen used in this experiment than on many standard monitors. No subjects
`complained that the accumulation affected their performance, and the experimen-
`ters did not notice a difference in performance between the early subjects and those
`at the end of the study.
`
`Experiment two: stabilized vs non-stabilized touchscreen
`INTRODUCTION
`
`The major purpose of the second experiment was to eliminate potential problems
`with the first experiment. Comparisons were limited to the stabilized and non-
`stabilized touchscreens. In the first study the targets were presented in one of four
`loc