Interpreting sequential proﬁling data via animated box-and-whisker plots Michael Plater Findlay, Alan Leonard, John C. Castura • Compusense Inc., Guelph, Ontario, Canada Objective This work explores approaches to visualizations of temporal data through the use of web-based animation of boxplots. Interactive animations of temporal data help to tell the story of the product experience. References Bostock, M., Ogievetsky, V. & Heer, J. (2011.) D3: Data-Driven Documents. IEEE Transactions on Visualization and Computer Graphics, IEEE Press. Box-and-whisker plot • Univariate data • Box is the height of the Interquartile range (IQR), with a crossbar at the median • IQR is a robust measure containing the middle 50% of the data and calculated by Q 3 −Q 1 • Upper fence calculated by Q 3 + 1.5 IQR, lower fence calculated by Q 1 − 1.5 IQR • Whisker length is determined by the most extreme value inside the fence • Outlying observations fall outside the fence: mild and extreme outliers occur more (less) than 1.5 IQR above (below) Q 3 (Q 1 ) Visualization in Practice • Trained sensory panel evaluates four snack bars in triplicate using sequential profiling. • Outlying observation for one panellist on the Sweetness attribute easily identified at time point 4. • One product stands out with intense Caramelized Flavour and Sweetness attributes. • Post-hoc letters above each sample box-and-whisker plot indicate differences between samples at each time point. • Animated box-and-whisker plots provide a snapshot view at each timepoint that incorporate univariate comparisons across samples, as well as a sense of intensity changes at adjacent time points. Web-based Animation • JavaScript framework “D3” (Data Driven Documents) used to create and animate the visualizations • Uses HTML, CSS and SVG (Scalable Vector Graphics) • The viewer is able to control the playback of the animation of the data, and can jump to specific time points Data • The visualizations lend themselves well to scale or ordered categorical data which are collected either at discrete timepoints (e.g. sequential profiling) or continuously (e.g. time intensity) Benefits • A flexible and portable means of creating engaging, visually appealing data visualizations for explaning complex sensory occurrences • Using an easy-to-understand animated data-driven visualization, get a sense of how product properties change over time, differences between products, and correlation among attributes and timepoints. Q 3 Q 1 Median Time Point 1 Time Point 2 Time Point 3 Time Point 4 Outlying observation Time Point 1 Time Point 2 Time Point 3 Time Point 4

Interpreting sequential profiling data via animated box-and-whisker plots

Download PDF Report

Upload
compusense-inc
View
89
Download
0

Embed Size (px)

DESCRIPTION

Sequential profiling extends traditional descriptive sensory analysis by collecting data at several pre-established times during evaluation. Intensities in product properties are quantified at each time point, permitting product changes to be captured. Evaluation is conducted with panellists who are trained to identify and scale sensory attributes, which removes ambiguity. We use animated box-and-whisker plots to visualize sequential profiling results. Animated box-and-whisker plots are created using Data-Driven Documents. Essentially animated box-and-whisker plots provide transitions between static boxplots constructed for different time points. Like static box-and-whisker plots, animated box-and-whisker plots allow for outlier detection and provide a useful summary of the distribution of the data. Non-parametric tests can be conducted to facilitate comparisons of products, accompanied by a visual presentation of pairwise product differences, with respect to particular attributes. This information provides the viewer with information that can either confirm or disconfirm a pattern that suggests itself. Animation can either be controlled by the viewer actively or run automatically. The visualization sequence can also be paused to permit a closer inspection. In this study, trained sensory panel evaluate four snack bars in triplicate using sequential profiling. Animated box-and-whisker plots reveal an outlying observation related to one of the panellists for the attribute Sweetness, with an intensity value so far beyond any other value in the dataset that we conclude it must have been submitted in error. Several other outliers were noted. The visualization is robust in that it is not influenced by such outliers. One products stands out as having an intense Caramelized Flavour and Sweetness, another for evolving Grain Flavour and Nutty Flavour. Animated box-and-whisker plots provide a snapshot views at each timepoint that incorporate univariate comparisons across samples, as well as a sense of intensity changes at adjacent time points.

Citation preview

Interpreting sequential profiling data via animated box-and-whisker plots

Michael Plater Findlay, Alan Leonard, John C. Castura • Compusense Inc., Guelph, Ontario, Canada

ObjectiveThis work explores approaches to visualizations of temporal data through the use of web-based animation of boxplots. Interactive animations of temporal data help to tell the story of the product experience.

ReferencesBostock, M., Ogievetsky, V. & Heer, J. (2011.) D3: Data-Driven Documents. IEEE Transactions on Visualization and Computer Graphics, IEEE Press.

Box-and-whisker plot • Univariate data

• Box is the height of the Interquartile range (IQR), with a crossbar at the median

• IQR is a robust measure containing the middle 50% of the data and calculated by Q3−Q1

• Upper fence calculated by Q3 + 1.5 IQR, lower fence calculated by Q1 − 1.5 IQR

• Whisker length is determined by the most extreme value inside the fence

• Outlying observations fall outside the fence: mild and extreme outliers occur more (less) than 1.5 IQR above (below) Q3 (Q1)

Visualization in Practice • Trained sensory panel evaluates four snack bars in triplicate using sequential profiling.

• Outlying observation for one panellist on the Sweetness attribute easily identified at time point 4.

• One product stands out with intense Caramelized Flavour and Sweetness attributes.

• Post-hoc letters above each sample box-and-whisker plot indicate differences between samples at each time point.

• Animated box-and-whisker plots provide a snapshot view at each timepoint that incorporate univariate comparisons across samples, as well as a sense of intensity changes at adjacent time points.

Web-based Animation • JavaScript framework “D3” (Data Driven Documents) used to create and animate the visualizations

• Uses HTML, CSS and SVG (Scalable Vector Graphics)

• The viewer is able to control the playback of the animation of the data, and can jump to specific time points

Data • The visualizations lend themselves well to scale or ordered categorical data which are collected either at discrete timepoints (e.g. sequential profiling) or continuously (e.g. time intensity)

Benefits • A flexible and portable means of creating engaging, visually appealing data visualizations for explaning complex sensory occurrences

• Using an easy-to-understand animated data-driven visualization, get a sense of how product properties change over time, differences between products, and correlation among attributes and timepoints.

Median

Time Point 1 Time Point 2 Time Point 3 Time Point 4

Outlying observation

Time Point 1 Time Point 2 Time Point 3 Time Point 4