1Fall 2005 6.831 UI Design and Implementation 1 Fall 2005 6.831 UI Design and Implementation 2 Fall 2005 6.831 UI Design and Implementation 3 Fall 2005 6.831 UI Design and Implementation 4 !" Experiment design2Fall 2005 6.831 UI Design and Implementation 5# Start with a testable hypothesis e.g. Mac menu bar is faster than Windows menu bar Manipulate independent variables different interfaces, user classes, tasks in this case, y-position of menubar Measure dependent variables times, errors, satisfaction Use statistical tests to accept or reject the hypothesisFall 2005 6.831 UI Design and Implementation 6$ % Users Windows users or Mac users? Age, handedness? How to sample them? Within- or between-subjects? Implementation Real Windows vs. real Mac Artificial window manager that lets us control menu bar position Tasks Realistic: word processing, email, web browsing Artificial: repeatedly pointing at fake menu bar Measurement When does movement start and end? Ordering of tasks and interface conditions Hardware mouse, trackball, touchpad, joystick? PC or Mac? which particular machine?Fall 2005 6.831 UI Design and Implementation 7 &' Processindependentvariablesdependentvariablesunknown/uncontrolledvariablesXYY = f(X) + Fall 2005 6.831 UI Design and Implementation 8# ( Internal validity Are observed results actually caused by the independent variables? External validity Can observed results be generalized to the world outside the lab? Reliability Will consistent results be obtained by repeating the experiment?3Fall 2005 6.831 UI Design and Implementation 9& ! Ordering effects People learn, and people get tired Don t present tasks or interfaces in same order for all users Randomize or counterbalance the ordering Selection effects Don t use pre-existing groups (unless group is an independent variable) Randomly assign users to independent variables Experimenter bias Experimenter may be enthusiastic about interface X but not Y Give training and briefings on paper, not in person Provide equivalent training for every interface Double-blind experiments prevent both subject and experimenter from knowing if it s condition X or Y Essential if measurement of dependent variables requires judgementFall 2005 6.831 UI Design and Implementation 10& ! Population Draw a random sample from your real target population Ecological Make lab conditions as realistic as possible in important respects Training Training should mimic how real interface would be encountered and learned Task Base your tasks on task analysisFall 2005 6.831 UI Design and Implementation 11)%! Uncontrolled variation Previous experience Novices and experts: separate into different classes, or use only one class User differences Fastest users are 10 times faster than slowest users Task design Do tasks measure what you re trying to measure? Measurement error Time on task may include coughing, scratching, distractions Solutions Eliminate uncontrolled variation Select users for certain experience (or lack thereof) Give all users the same training Measure dependent variables precisely Repetition Many users, many trials Standard deviation of the mean shrinks like the square root of N (i.e., quadrupling users makes the mean twice as accurate)Fall 2005 6.831 UI Design and Implementation 12* + Divide samples into subsets which are more homogeneous than the whole set Lots of variation between feet of different kids But the feet on the same kid are far more homogeneous Each child is a block Apply all conditions within each block Put material A on one foot, material B on the other Measure difference within block Wear(A) Wear(B) Randomize within the block to eliminate internal validity threats Randomly put A on left foot or right foot4Fall 2005 6.831 UI Design and Implementation 13*' %, (-. %, Between subjects design Users are divided into two groups: One group sees only interface X Other group sees only interface Y Results are compared between different groups Is mean(xi) > mean(yj)? Eliminates variation due to ordering effects User can t learn from one interface to do better on the other Within subjects design Each user sees both interface X and Y (in random order) Results are compared within each user For user i, compute the difference xi-yi Is mean(xi-yi) > 0? Eliminates variation due to user differences User only compared with
View Full Document