A comparison of screening methods: Selecting important design variables for modeling product usability

Abstract

Product usability is affected by a large number of design variables. For example, to design an audio/visual product such as a video-cassette player, the design of controls, information displays, layout of controls, etc. could affect the user task performance. In addition, the product shape, color, material, etc. could also affect the subjective user satisfaction. Altogether, the number of design variables could easily go up to hundreds. To build a model describing the functional relationship between the product usability and the product design variables, it is very important to select only the important design variables for securing modeling efficiency and obtaining an effective model. Although some studies used expert opinions to select them, they have drawbacks such as lack of selection objectivity and expert availability. This study proposes statistical methods for screening important design variables to substitute the expert opinions. Three methods (i.e., principal component regression; cluster analysis; and a partial least squares) are applied to a set of design variables for audio/visual electronic products. Performances of the variable screening methods are examined by building usability models. That is, the variables screened by each method are used to build a usability model. The model performances are then compared to those screened by expert opinions to determine which method provides the best performance. The results show that all these three methods have better model performances than the expert screening in terms of R-2, the number of variables in the model, and PRESS.