AN EXTENSIBLE AND INTERACTIVE RESEARCH PLATFORM FOR EXPLORING FITTS LAW

Size: px

Start display at page:

Download "AN EXTENSIBLE AND INTERACTIVE RESEARCH PLATFORM FOR EXPLORING FITTS LAW"

David Wilcox
6 years ago
Views:

1 AN EXTENSIBLE AND INTERACTIVE RESEARCH PLATFORM FOR EXPLORING FITTS LAW Schedlbauer, Martin, University of Massachusetts Lowell, Department of Computer Science, Lowell, MA 01854, USA, Pastel, Robert, Michigan Technological University, Department of Computer Science, Houghton, MI 49931, Heines, Jesse, University of Massachusetts Lowell, Department of Computer Science, Lowell, MA 01854, USA, Abstract We demonstrate the utility of a new software platform for the interactive exploration of movement time models such as Fitts law. The software is written in Java and provides a flexible environment for HCI research, input device evaluation, throughput comparison and education. It is specifically designed using a patterns-based object model to provide maximum extensibility for new models, task types, and selection modes. The platform can be used to share experimental configurations for more meaningfully comparable experiments. We demonstrate the platform s utility for HCI research by measuring the effects of posture on movement time. The experimental results show a significant increase in mean movement time when standing compared to sitting while using a trackball. Keywords: Fitts law, human performance modelling, input device evaluation, completion time predictions.

2 1 INTRODUCTION While many ad hoc utilities have been developed for evaluating input devices and predicting user s interaction times, except for Soukoreff and MacKenzie s (1995) Fitts law model builder, few of these utilities are research platforms. A standardized research platform that is extensible would allow human-computer interaction (HCI) researchers to compare experimental results meaningfully and share experimental configurations. In this paper we demonstrate the Movement Time Evaluator (MTE), which is an interactive and extensible research platform. MTE presents a configurable environment for exploring input device characteristics, as well as a tool for rapidly evaluating performance models. MTE is written in Java and is constructed on an extensible object-oriented framework. Data files generated by MTE are openly sharable and exportable in formats compatible with many statistical analysis packages. MTE also has a comprehensive graphical user interface for researchers to configure their experiments and interpret results. Using MTE, researchers can immediately compare Fitts law and its variations (Accot and Zhai, 2003; Fitts, 1954; Kvålseth, 1980; MacKenzie, 1995; Meyer et al., 1988; Oel et al., 2001). MTE extends Soukoreff and MacKenzie s (1995) platform by adding bivariate pointing tasks, probe corrections, additional performance models, target motion, and trajectory evaluation. We believe that MTE will allow user interface device manufacturers to conduct standardized experiments and to directly compare device characteristics, performance and usability. 2 MOVEMENT TIME MODELS Movement Time (MT) models allow HCI researchers to understand and predict human aiming performance, which are abundant in user interfaces. The most popular psychological model for human aiming is Fitts law (Fitts, 1954). Card, English, and Burr (1978) were the first scientists to demonstrate the applicability of Fitts law in HCI. Many other scientists have extended Fitts law to a variety of HCI domains and tasks (Accot and Zhai, 1997; Accot and Zhai, 2003; Radix, Robinson, and Nurse, 1999; Murata, Fujii, Arima, Iwase, 1999). However, Fitts law has not been extended to pointing tasks that require the user to stand or walk. We demonstrate MTE s utility by extending Fitts law to pointing with a trackball while standing, and compare the trackball pointing performances while sitting versus standing. The general Fitts law is a logarithmic relationship between MT and the ratio of target distance (A), and target width (W): A MT a blog 2 (1) W where the value of ε {½, 1, A/W}, and a and b are experimentally derived regression coefficients. The logarithmic term in equation (1) is commonly referred to as the movement Index of Difficulty (ID). Typically the target width, W, is the smaller of the horizontal or vertical extent of the target for univariate (one dimensional) pointing or the width along the approach vector for bivariate (two dimensional) pointing, although other values of W have been proposed, such as target area and the sum of width and height (Accot and Zhai, 2003; MacKenzie, 1995). These variations of Fitts law by MacKenzie (1995), as well as the alternate models by Meyer et al. (1988), Accot and Zhai (2003) and Kvålseth (1980), have been shown to be superior to Fitts original formulation in specific testing configurations. MTE allows researchers to test conformance with all of the above models. The throughput (TP) of an input device is a measure of the device s efficiency (MacKenzie, Kauppinen and Silfverberg, 2001; Zhai, 2004) and is the ratio of the mean ID and the mean MT.

3 3 PLATFORM DESIGN MTE s architecture is based upon an extensible object-oriented design implemented in Java. The platform can be controlled remotely, allowing a researcher to observe and direct the experiment from one workstation while the participant interacts with the software on another workstation. Connectivity is through a standard TCP/IP socket, and therefore the researcher and the participant can be separated by any distance or any number of walls. 3.1 Configurability The configuration of experiments is a two-step process. First, the researcher specifics and saves an experimental configuration file, which specifies session invariant parameters: target extent, target shape (rectangular, oval, text, vibrating, virtual keypad), home region placement (center of screen, random, coordinate system origin, or none), target position distribution (uniformly random, static, or reciprocal), feedback (auditory, visual), number of repetitions, movement type (point-and-click versus drag), and information recorded (movement velocity and selection errors) Second, before the researcher runs an experiment, a configuration file is selected from a list of stored setups. Prior to the beginning of an experimental trial, the researcher can enter trial-specific factors, such as subject information, environmental factors, or any relevant ad hoc information. In addition, demographic data points (age, gender, height, and handedness) are saved for each participant. 3.2 Evaluation and Analysis Basic statistical analysis built into the platform facilitates rapid evaluation of experimental results and interactive exploration of models. MTE implements Pearson moment correlation analysis and linear regression, as well as configurable scatter plots and distribution graphs. In addition, the raw data and the trajectory of the individual acquisition movements during a session can be viewed so that the movement patterns for different input devices can be studied. Also, the researcher can view a table comparing the correlation coefficients of various movement time models, including the variations of Fitts law by Meyer et al. (1988), Kvålseth (1980), and Accot and Zhai (2003). The linear regression coefficients and the correlation coefficient (R) can be computed either on the observed or averaged MT values over any ID binning interval. 3.3 Extension and Customization MTE can be extended by adding new static or dynamic (moving) shape types, target position distributions, MT models, and data export formats. Many of these extensions require minimal integration. HCI researchers should find the easy integration of new MT models to facilitate quick hypotheses testing and to encourage the process of discovery in scientific enquiries.

4 3.4 Exporting to Other Tools MTE records the cursor position, time, and movement velocity approximately every 16 msec using object serialization. The data can be exported as CSV files for importing into spreadsheets, relational databases, and statistical analysis packages. The format of the exported data can be customized. 3.5 Availability The MTE platform is available under the GNU Public License (GPL), and an executable version that runs under Java and later on Linux and Windows 98/NT/XP is available on line (Schedlbauer 2005). 4 EXPERIMENTAL RESULTS 4.1 Configuration In this section we illustrate the utility of MTE and its effectiveness for interactive data exploration by determining the effect of posture on MT in Fitts law. We configured several experiments, which were carried out using a trackball, first sitting and then standing. 9 men and 3 women participated in the experiment. The test platform was a Gateway M275 Tablet PC (1.6GHz CPU, 512MB RAM) with a 14.1 TFT Stylus Touch Sensitive LCD (resolution 1024 by 768 pixel) running Windows XP and Java A Kensington Turboball (Model 64227) was connected to the computer via USB. Each participant was sequentially presented 80 circular targets that had to be acquired after clicking on the home region; clicking on the home region started the timing of the trial. The experiment varied target-to-home distance and four target diameters (15, 30, 45, and 60 pixels). We used a fixed-random target positioning distribution configuration, which generates a set of randomly positioned targets that were used for each participant. Alternatively, we could have selected pseudo-random placement with a variable seed, reciprocal placement where the target moves back and forth between the left and the right side of the home position, and a static distribution where the targets are displayed at fixed and pre-programmable positions. The target order can be randomly varied to mitigate the effects of learning in repeated trials. 4.2 Exploration and Initial Evaluation The individual participant data sets can be merged into a single data set for analysis, or a summarized data set that combines the mean MT of the individual experiments. Data sets can be explored in their raw form or binned (grouped and averaged), and outliers can be removed. Figure 1 shows a scatter plot of the raw data for one participant s experimental session along with the linear regression equation for the Shannon formulation (MacKenzie, 1995) of Fitts law. MT is the ordinate and the ID is the abscissa. The platform allows any recorded parameter to be plotted against any other recorded parameter, e.g., amplitude against movement time or target angle against ID. To facilitate the comparison of different experiments, multiple data sets can be plotted side-by-side or overlaid. The trackball movements while standing appear to be slower compared to a sitting posture. In addition, MTE can produce plots of the cursor s trajectory tracks and time vs. velocity plots (Figure 2.) Similar to the Generalized Fitts Law Model Builder by Soukoreff and MacKenzie (1995), MTE can summarize the correlation and throughput for several MT models for all the trials as shown in Figure 3. The summary shows the correlation coefficient (R) and the coefficient of determination (R 2 ) for each model using the Euclidean distance (A) between the targets as the amplitude as well as the length

5 of the cursor trajectory (D). Note that the correlation is consistently higher for trajectory distance than for Euclidean distance. The simple and Kvålseth models have larger TP than the other models because their ID are larger, but all model s TP is consistently larger for sitting than standing. Also, the mean MT is larger for standing than for sitting. A statistics package should be used to determine if the difference in TP and MT is significant. Figure 1. Scatter plot of ID (Shannon formulation) versus MT of two runs of the same experiment, one while standing (diamonds) and while sitting (squares). Note that MT appears to be higher when standing. Regression line, correlation coefficient (Pearson moment) and linear regression equation are shown. The values of ID are scaled by a factor of 100. Figure 2. Graph of time since start of movement versus speed (ds/dt). The graph starts at an offset which represents the reaction time before movement towards the target commences. An initial ballistic movement towards the target is followed by a series of small corrective submovements until the target is selected.

To test this hypothesis, the data collected using MTE was exported to R, a statistical package (Gentleman and Ihaka, 1997).

6 (a) (b) Figure 3. Table comparing several movement time models for the results of the experiment while sitting (a) and standing (b). 4.3 Further Analysis Initial analysis of the experimental data and the plot of Figure 1 suggest that MT using a trackball increases when standing compared to sitting. To test this hypothesis, the data collected using MTE was exported to R, a statistical package (Gentleman and Ihaka, 1997). Even though the movement time shows an increase, Shannon s formulation of Fitts law (MacKenzie, 1995) remains a robust predictor of movement time. Correlation analysis between MT and ID results in R = 0.54 (p < ) for the raw data and R = 0.98 when using mean MT over twenty ID ranges. Figure 3 shows strong correlations with the other movement time models as well. A Wilcoxon sum-of-ranks significance test on the unmatched measurements shows a significant interaction (p < 0.003). In addition, the mean error rate while standing was 7.5% (σ=0.301) compared to an error rate of 4.1% (σ=0.220) while sitting, an increase of 80%. The error rate while sitting is consistent with those reported by MacKenzie (1995) and Card et al. (1978). The Wilcoxon sum-ofranks test shows the increase of the error rate while standing to be significant (p < 0.002). Figure 4 illustrates the interaction between target size, posture, mean MT, and error rate. The movement time while standing is considerably greater than sitting, but the difference is less as target size increases, i.e., as ID decreases. A similar trend is apparent for the mean error rate.

7 Movement Time (ms) MT ( sitting) MT ( standing) Error Rate ER (sitting) ER (standing) Target Diameter (pixel) Target Diameter (pixel) Figure 4. (a) Mean MTs (a) and error rates (b) for each posture by target diameter. The increase in movement time while standing is more pronounced as the target size decreases and consequently the difficulty of the movement increases. (b) 5 PLATFORM PERFORMANCE The software performed without incident during the collection of the measurements. A few modifications were made to the target feedback parameters after initial trials. In particular, some participants reported that they were unsure when the trial started, so the home region was hidden after its successful acquisition. When it was discovered that occasionally participants paused during trials, a mechanism for excluding compromised data was added to MTE. Finally, the platform s display of the total time taken for a series of trials prompted many participants to treat the experiment like a game. We felt that this might affect the accuracy of the movement and the total time feedback was omitted from the present version of MTE. We are currently expanding the capabilities of MTE by adding more movement time models. Supplementary performance models, such as those for measuring reaction time and learning, are being considered for inclusion into MTE. In addition, the availability of a web-accessible repository of experimental configurations and results is being explored. 6 CONCLUSION MTE is an effective research platform for the exploration of MT models, and we have demonstrated its effectiveness as a tool for evaluating input devices. We have also demonstrated that MTE s interactive features provide immediate feedback to the researcher and can illuminate the factors influencing participants MT. Specifically, we have demonstrated the utility of MTE by showing the effect of standing on MT while using a trackball. The experimental results show a significant increase in mean MT and mean error rate using a trackball while standing compared to sitting at all levels of movement difficulty, although more difficult movements have a higher increase. MTE is an open source research platform for conducting Fitts law tests. Although the statistical calculations built into MTE are useful, they are not intended to be a substitute for more rigorous statistical analysis provided by specialized statistics packages.

8 REFERENCES Accot, J, Zhai, S. (1997). Beyond Fitts law: Models for trajectory-based HCI tasks. In Proceedings of the ACM Conference on Human Factors in Computing Systems CHI 1997, March 1997, , Atlanta, GA: ACM. Accot, J, Zhai, S. (2003). Refining Fitts law models for bivariate pointing. In Proceedings of the ACM Conference on Human Factors in Computing Systems CHI 2003, April 2003, , Ft. Lauderdale, FL: ACM. Card, S., English, W., & Burr, B. (1978). Evaluation of mouse, rate-controlled isometric joystick, step keys, and text keys for text selection o n a CRT. Ergonomics, 21, Fitts, P. M. (1954). The information capacity of the human motor system in controlling the amplitude of movement. Journal of Experimental Psychology, 47, Gentleman,R. and Ihaka, R. (1997). R Project. Hoffmann, E. R., Sheikh, I. (1991). Finger width corrections in Fitts Law: Implications for speedaccuracy research. Journal of Motor Behavior, 23(4), pp Kvålseth, T. (1980). An alternative to Fitts law. Bulletin of the Psychonomic Society, 16(5), MacKenzie, S. (1995). Movement time predictions in human-computer interfaces. In Readings in Human-Computer Interaction, 2nd Edition, , Los Altos, CA: Morgan Kaufman. MacKenzie, I. S., Kauppinen, T., & Silfverberg, M. (2001). Accuracy measures for evaluating computer pointing devices. In Proceedings of the ACM Conference on Human Factors in Computing Systems - CHI 2001, New York: ACM. Meyer, D., Abrams, R., Kornblum, S., Wright, C., & Smith, J. (1988). Optimality in human motor performance: Ideal control of rapid aimed movements. Psychological Review, 95:3, Murata, A., Fujii, M., Arima, Y., Iwase, H. (1999). Extending Effective Target Width in Fitts' Law to a Two-Dimensional Pointing Task. International Journal of Human-Computer Interaction. 11(2), Oel, P, Schmidt, P., A. Schmitt (2001). Time prediction of mouse-based cursor movements. In Proceedings of Joint AFIHM-BCS Conference on Human-Computer Interaction IHM-HCI September 2001, Lille, France, Volume II, Radix, C., Robinson, P., Nurse, P. (1999). Extension of Fitts' Law to modeling motion performance in man-machine interfaces. IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans, 29(2), Schedlbauer, M. (2005). Movement Time Evaluator, Soukoreff, W., & MacKenzie, I. S. (1995). Generalized Fitts' law model builder. In Proceedings of the ACM Conference on Human Factors in Computing Systems CHI 95, Denver, CO, Soukoreff, W., & MacKenzie, I. S. (2004). Towards a standard for pointing device evaluation, perspectives on 27 years of Fitts law research in HCI. International Journal of Human-Computer Studies, 61(6), Thompson, S., Slocum, J., Bohan, M. (2004). Gain and angle of approach effects on cursorpositioning time with a mouse in consideration of Fitts law. In Proceedings of the Human Factors and Ergonomics Society 48th Annual Meeting. New Orleans, LA: Human Factors and Ergonomics Society, Zhai, S. (2004). Characterizing computer input with Fitts law parameters the information and noninformation aspects of pointing. International Journal of Human-Computer Studies, 61(6),

Brain Computer Interface Cursor Measures for Motionimpaired and Able-bodied Users

Brain Computer Interface Cursor Measures for Motionimpaired and Able-bodied Users Alexandros Pino, Eleftherios Kalogeros, Elias Salemis and Georgios Kouroupetroglou Department of Informatics and Telecommunications