Paper Title: The Aligned Rank Transform for Nonparametric Factorial Analyses Using Only ANOVA Procedures
Authors: Jacob O. Wobbrock, Leah Findlater, Darren Gergle and James J. Higgins
Authors Bio:
Jacob O. Wobbrock is an Associate Professor in the Information School and an Adjunct Associate Professor in the Department of Computer Science & Engineering at the University of Washington. He works in the field of HCI, combining computer science, interaction design and psychology to investigate novel user interface technologies, input and interaction techniques, human performance with computing systems, and accessible, mobile & surface computing interfaces.
Leah Findlater is a postdoctoral researcher in The Information School, working with Dr. Jacob Wobbrock. Her research interests include personalization, accessibility, and information and communication technologies for development (ICTD). She is also a member of the AIM Research Group.
Darren Gergle is an Associate Professor in the departments of Communication Studies and Electrical Engineering & Computer Science at Northwestern University. He also directs the CollabLab: The Laboratory for Collaborative Technology. His research is in the fields of Human-Computer Interaction (HCI), Computer-Supported Cooperative Word (CSCW) and Computer-Mediated Communication (CMC) with an interest in developing a theoretical understanding of the role that visual information plays in supporting communication and small group interactions.
James J. Higgins is a Professor at the Kansas State University in the Department of Statistics.
Presentation Venue: CHI '11 Proceedings of the 2011 annual conference on Human factors in computing systems that took place at New York (ACM)
Summary:
Hypothesis: The authors of this paper present the Aligned Rank Transform for nonparametric analysis of factorial experiments using the familiar F-test. The ART offers advantages over more complex methods in its simplicity and usability. The authors offered the first generalized mathematics for an N-way ART and programs called ARTool and ARTweb to make alignment and ranking east. By providing three examples of published data re-examined using the ART, they exhibited its benefits.
The authors provide an example of Conover and Iman's Rank Transform (RT) that applies ranks, averaged in the case of ties, over a data set, and then uses the parametric F-test on the ranks, resulting in a nonparametric factorial procedure. They mention how it was discovered that this process produces inaccurate results for interaction effects and then provide a solution for it using the ART.
How the hypothesis was tested: The authors provide the steps of the ART procedure in five steps described below:
Step 1: This step shows how the residual is computed for each raw response Y
residual = Y - cell mean
Step 2: This step shows how the authors computed estimated effects for all main and interaction effects. They show multiple ways to estimate the main effect for a factor A with response Yi: one-way, two-way, three-way, four-way and ultimately N-way effects.
Step 3: This step shows how the authors compute the aligned response Y'
Step 4: This step assigns averaged ranks to a column of aligned observations Y' to create Y''
Step 5: This step shows how to perform a full-factorial ANOVA on Y''.
The authors also built the ARTool and ARTweb in order to make alignment and ranking easy, and to make sure no assumption of data is made in long-format respectively.
Discussion:
Effectiveness: This is a great method built by the authors in order to perform nonparametric analysis of factorial experiments using the familiar F-test. It is an excellent tool that can be used in almost any real time systems to make alignment and ranking easy and accurate.
Faults: The ART has some limitations. For data exhibiting very high proportions of ties, the ART simply replaces those ties with tied ranks. If data exhibits extreme skew, the ART will reduce that skew which may be undesirable if distributions are meaningful.
Authors Bio:
Jacob O. Wobbrock is an Associate Professor in the Information School and an Adjunct Associate Professor in the Department of Computer Science & Engineering at the University of Washington. He works in the field of HCI, combining computer science, interaction design and psychology to investigate novel user interface technologies, input and interaction techniques, human performance with computing systems, and accessible, mobile & surface computing interfaces.
Leah Findlater is a postdoctoral researcher in The Information School, working with Dr. Jacob Wobbrock. Her research interests include personalization, accessibility, and information and communication technologies for development (ICTD). She is also a member of the AIM Research Group.
Darren Gergle is an Associate Professor in the departments of Communication Studies and Electrical Engineering & Computer Science at Northwestern University. He also directs the CollabLab: The Laboratory for Collaborative Technology. His research is in the fields of Human-Computer Interaction (HCI), Computer-Supported Cooperative Word (CSCW) and Computer-Mediated Communication (CMC) with an interest in developing a theoretical understanding of the role that visual information plays in supporting communication and small group interactions.
James J. Higgins is a Professor at the Kansas State University in the Department of Statistics.
Presentation Venue: CHI '11 Proceedings of the 2011 annual conference on Human factors in computing systems that took place at New York (ACM)
Summary:
Hypothesis: The authors of this paper present the Aligned Rank Transform for nonparametric analysis of factorial experiments using the familiar F-test. The ART offers advantages over more complex methods in its simplicity and usability. The authors offered the first generalized mathematics for an N-way ART and programs called ARTool and ARTweb to make alignment and ranking east. By providing three examples of published data re-examined using the ART, they exhibited its benefits.
The authors provide an example of Conover and Iman's Rank Transform (RT) that applies ranks, averaged in the case of ties, over a data set, and then uses the parametric F-test on the ranks, resulting in a nonparametric factorial procedure. They mention how it was discovered that this process produces inaccurate results for interaction effects and then provide a solution for it using the ART.
How the hypothesis was tested: The authors provide the steps of the ART procedure in five steps described below:
Step 1: This step shows how the residual is computed for each raw response Y
residual = Y - cell meanStep 2: This step shows how the authors computed estimated effects for all main and interaction effects. They show multiple ways to estimate the main effect for a factor A with response Yi: one-way, two-way, three-way, four-way and ultimately N-way effects.
Step 3: This step shows how the authors compute the aligned response Y'
Step 4: This step assigns averaged ranks to a column of aligned observations Y' to create Y''
Step 5: This step shows how to perform a full-factorial ANOVA on Y''.
The authors also built the ARTool and ARTweb in order to make alignment and ranking easy, and to make sure no assumption of data is made in long-format respectively.
Discussion:
Effectiveness: This is a great method built by the authors in order to perform nonparametric analysis of factorial experiments using the familiar F-test. It is an excellent tool that can be used in almost any real time systems to make alignment and ranking easy and accurate.
Faults: The ART has some limitations. For data exhibiting very high proportions of ties, the ART simply replaces those ties with tied ranks. If data exhibits extreme skew, the ART will reduce that skew which may be undesirable if distributions are meaningful.
No comments:
Post a Comment