Portfolio: Design Process Overview

The design process outlined here is intended to rapidly explore multiple solutions to design problems. The procedure is based on a description of the design space according based on the dimensions across which potential interfaces differ. Initial probe interfaces are designed and submitted to mixed-methods user tests in order to evaluate users' reaction to the different dimensions and identify optimal areas within the design space.

Background


This process was developed to address to design problems in my doctoral research:
  1. Restricted Selection: Designing a note-taking interface for selecting text that reduces copy-paste selection size while remaining user-friendly.
  2. Highlighting Interfaces: Designing a user-friendly highlighting interaction.

These are two different types of problems. In the first, I needed to tradeoff the desired behavioral outcome (shorter selections) with user satisfaction. In the second, I simply wanted to determine what highlighting interface would be most user-friendly. The procedure proved valuable for both problems.

Design Space and Interface Dimensions


The design space of potential interfaces that accomplish the same goal can be categorized according to the interface dimensions across which the interfaces differ. For example, highlighting interfaces can differ with regards to visibility and timing. With regards to visibility, some interfaces provide a visible button for students to click in order to highlight a selection, while others require students to press a keyboard button or click on the selection. With regards to timing, to highlight text, some interfaces require students to take an action (e.g. click a button) before making a selection, while others require the action to occur after making a selection.

Dimensions can be discovered through a competitive analysis of existing interfaces. For my highlighting study, I compared over 20 highlighting interfaces that implemented highlighting in a variety of ways. However, for my other study, I could find no interfaces intended to reduce selection size. In this case I first brainstormed a set of possible solutions with colleagues, and then evaluated the differences between these hypothetical solutions. I found this process actually served to inspire further novel solutions, as was also the case with the highlighting project.

User Tests


The user-test involves three methods of data collection. First, participants are asked to "think-aloud" while using the interfaces. They then fill out a survey containing Likert-scale measures of their response to the specific dimensions, as well as more general questions regarding their reactions to the interfaces. Finally, a semi-structured interview is conducted touching on more broad questions in the design space, such as what type of material a student focuses on while highlighting. Using multiple methods allows for stronger conclusions regarding the value of specific dimensions or interfaces, and identifies occasions when stated preference contrasts with observed frustrations.

Each participant in the user-test is given several interfaces to try out. Interfaces are assigned to users so that each participant gets experiences with as many levels of each interface dimension as possible. This ensures that their responses are grounded in knowledge rather than prediction.

Iteration/Hypothesis Testing


The first user-tests are intended to explore responses to the interface dimensions, not evaluate candidate interfaces. As data is gathered, certain levels or combinations of the dimensions are thrown out. For example, in the initial highlighting tests, it became apparent that users strongly preferred that the highlighting action be taken after a selection was made, not before. Hypotheses raised by the user data are also tested. For example, in the initial tests, visibility was a divisive dimension, with some users expressing a strong preference for an invisible interaction (e.g. clicking on the selection) while others strongly preferred a visible interaction (a button). The following phase explored interfaces that combined the two, allowing users to either click on a button or click on the selection, and found that solution satisfied all users.

Limitations


This design process produced guidelines for manipulating selections and implementing highlighting. The resultant interfaces proved effective experimentally. However, this process has several limitations. Two of the biggest are:

  • Cost of implementation: This process involves the design of many interfaces, which may require too much development time. This can be limited by using existing interfaces or prototypes.
  • Type of Interaction: I used this process to design two interactions that occurred within a short period of time. Interactions that took much longer, perhaps requiring coordination over time, would not be appropriate for this type of process.

References


The first version of this design study, which looked at restricting selection, was published at the Joint Conference on Digital Libraries 2008:
Bauer, A. and Koedinger, K. R. (2008) Note-taking, selecting, and choice: designing interfaces that encourage smaller selections. In Proceedings of the 8th ACM/IEEE-CS Joint Conference on Digital Libraries (JCDL '08). pp. 397-406. ACM Press
A more complete description of this design process, along with how it was used in my highlighting study, is available in my dissertation:
Bauer, A., Designing Note-Taking Interfaces for Learning. Carnegie Mellon University, Human-Computer Interaction Institute, Doctoral Dissertation - CMU-HCII-08-103.

Key Points

Interface Dimensions
Describing the dimensions across which the interfaces differ ensures that probe interfaces cover the range of potential solutions. It is not necessary to create all possible combinations, only enough so that each level of each variable is covered.
Mixed Methods
Coordinating data from multiple methods allows us to compare behavior and stated preference, and evaluate both interface dimensions and specific interfaces.
Iteration
Levels and combinations of interface dimensions are eliminated until all remaining interfaces can be included in a single user-test.