|
Activity Number:
|
515
- Visualization for Distributions, Networks and Statistical Inference
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, July 31, 2019 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistical Graphics
|
|
Abstract #306679
|
Presentation
|
|
Title:
|
Lady Tasting Tea Lineups for Visual Statistical Inference
|
|
Author(s):
|
Karsten Maurer* and Seonjin Kim and George Woodbury
|
|
Companies:
|
Miami University and Miami University and Miami University
|
|
Keywords:
|
Visual Statistical Inference;
Design of Experiments;
Lineup Plots
|
|
Abstract:
|
Lineup protocols of Visual Statistical Inference were proposed by Baju et al. (2009) and validated by Majumder et al. (2013). The lineups construct many plots using data generated under the null, then randomly place a single copy of the plot constructed using the real data. When administered with individuals who have not seen the real data, the correct identification of the real plot are evidence that the data was not generated under the null. We propose the use of a variant on this testing procedure that follows the classic experiment of the Lady Tasting Tea (Fisher, 1935). In this setting the lineups are populated half by plots constructed with data generated under the null and other half with the real data after adding random perturbations that preserve plot structure. The task is no longer to pick out the most different plot, but to label the plots into the two sets that are most different. In this case the p-values can be calculated from the hyper-geometric distribution based on the number of correctly grouped plots. We evaluate this procedure empirically through a designed experiment of visual trials.
|
Authors who are presenting talks have a * after their name.
