limegimlet / data_viz

Course files from Udacity Data Visualization course

Visualizing major-league baseball stats with dimple.js

Course files & final project for the Udacity Data Visualization course, part of the Data Analyst nanodegree.

Summary

This plot highlights the difference in average homeruns and batting averages for right-, left-, and ambidextrous major-league baseball players, without any clear correlation to height or weight. It uses the dimple.js charting library.

Data

The data comes originally Udacity's baseball data set of stats for 1157 professional players.

The final visualizations use 2 datasets that are slightly modified versions of the original:

• Plot 1 uses a long-format version of the dataset, baseball_data_long.csv. The variable names were stacked into an attribute category column, and their values into a value column. This allows the creation of a categorical x-axis for all 4 baseball metrics.
• Plots 2 & 3 use the original wide-format baseball_data.csv with an additional counts column. It stores the total number of people in that player's handedness group, so it can be used to size the bubbles.
• Both .csv's had column names and values renamed, described in the Feedback details section.

Design

In retrospect, I focused too much initially on customizing look & feel instead of assessing the effectiveness of my design choices. On top of that, I ran into some showstopper bugs with customizing tooltips. In the end, I had to go back to the drawing board before I had something working well enough to seek outside feedback.

The evolution is documented below.

Initial design

Plot type: bubble chart

• Since there were major differences for 2 different variables (homeruns and batting average), the plotting on an x-y axis most clearly shows each group's relative position.

Colour scheme: black and white

• conveys pen & ink which again conveys the feeling baseball more than the muted pastels ubiquitous to corporate KPI dashboards.
• use cross-hatching to denote different groups: can use direction of cross-hatching to reinforce left vs right vs both.
• Inspiration is this plot

Axes: standardized around a mean of 0

• a standardized axes allows the same axes to compare different pairs of variables. Since I'd like to allow users to toggle between a plot of homeruns vs batting average and height vs weight, having common axes allows the user to focus only on the positional changes of bubbles.

Animation

• the user can click to comparing the other two numeric variables, height and weight, where there is nearly no difference between right- and left-handed players.
• inspriration is this plot.

Prototypes

Prototype 1a: Bubble plot with standardized data (z-scores)

Issues:

• Standardized axes over-complicates message
• The customized tooltip returns 0 for negative y-values and all x-values!
• It's statistically incorrect. Since HR and avg are not normally distributed, it's misleading to assume a mean of 0 with standardized values (this is the Central Limit Theorem).

Solution: use the original baseball data.

Prototype 1b: Bubble plot with non-standardized data

Issues:

• While bubbles are efficent in that they represent 3 variables at once, it also means there's more to absorb.
• The fill patterns are too noisy, especially when bubbles overlap
• Custom tooltip still returns 0 for x-values.

Solution: Instead, use a bar plot.

Prototype 1c: Animated bar plot

Issues:

• Unlike bubbles, the bars are side by side, so the fill pattern is hard on the eyes.
• Bars don't do as good a job as bubbles for contrasting performance differences vs size differences.

Solution: combine bar & bubble in one data story. And let go of the custom fill patterns.

Back to the drawing board!

Initially, I envisioned animating each of the 3 plots.

• The bar plot introduces to the data story. By displaying bars L, B, R for each statistic in turn, the viewer can more easily digest compare the handedness groups for each variable.
• The bubble plots provide the main point of the story. They would toggle to show the non-aggregated data as a scatterplot OR the means as bubbles.

Prototype 2a:'Overview' bar plot + 2 bubble/scatter plots

View on bl.ocks.org

Issue:

• In the end, the scatterplots didn't do anything to explain the differences between each handedness group, and would require extra dev time and clutter up the page with additional toggle buttons.

Solution: Keep plot 1 (the bar plot) animated, but plots 2 & 3 would be static bubble plots.

Prototype 2b:'Overview' bar plot + 2 bubble plots

View on bl.ocks.org

V1.0 - Ready for outside feedback

• Added titles to each plot that together formed a sentence. Aside from that, kept text minimal as the plots pretty much speak for themselves.
• Restored the handwriting-like font 'Shadow into Light'

View on bl.ocks.org

v1.1 & v1.2

These are interim versions I made after receiving feedback. See the 'Feedback details' section for a description of changes.

V2.0 - Final version

This is the version being submitted and included in this repo. See the 'Feedback details' section for a desciription of changes.

Feedback

4 people in total gave gave feedback on 3 different iterations of the data story.

To summarize:

• All 4 clearly understood the overall message.
• However, only 1 out of 4 liked the animation. The other 3 were puzzled why it couldn't be static.
• 1 person really liked the changing y-axis, another hated it.
• 1 person didn't understand that bubble size represented group size.
• Surprsingly (to me), only 1 person complained about the lag time after pressing "Play".

Feedback details

For v1.0

From Karsten, Director Solution Management, SAP Global GTM Analytics

For v1.1

Changes to address Karsten's feedback:

• Updated .csv file to make values more explicit on x-axis and legend
  • hr => homeruns
  • avg => batting avg
  • weight => weight (lbs)
  • height => height (in)
  • B => Both

From Paula, Director, Learning & Knowledge Management, Vision Critical

For v1.2

Changes to address Paula's feedback:

• Increased duration time for frames: story.FrameDuration=4000;
• Hid the y-axis
• Made the hover text lighter, so it was easier to see the button change to 'Pause' after clicking

From Shoko - Visual Artist / Writer / Translator

From Kinga - Mobile Application Developer, BAM software

v2.0 - Final version

Changes to address Shoko & Kinga's feedback:

• In plot 1, to reduce the 'noise' inherent with animations, I removed redundant tooltip text.
• In plot 3, added units of measure to axes labels

Additional changes based on my own assessments:

• To provide context, added a short description of dataset beneath plot 1.
• In plot 2 & 3, adjusted axes ranges to reinforce idea that bubbles represent means, i.e. towards the middle of plot, rather than upper right corner. Specifically:
  • plot 2: set x.overrideMax to the max value for batting avg. I did NOT do the same for y.overrideMax since it's such an extreme outlier.
  • plot 3: set both x.overrideMix & y.overrideMin to the minumum value for weight & height. This makes it more clear that the Both group shorter and lighter, although still close to the L & R means.

Justification for not addressing some feedback

Animation of plot 1

The main reason I animated plot 1 is that if it were static, it would show 4 groups of 3 bars. 12 bars is a lot to process.

Moreover, by forcing the viewer to press the Play/Pause button to see the bars and read the tooltips, it introduces 'cognitive disfluency': interacting with the visualization means the user needs to pay more attention to it.

It also solved the awkwardness of having batting average (a decimal) share the same y-axis with the other stats. However, even without batting average, I would have taken the same approach. A declarative title for a mostly-empty plot with a 'Play' button is an invitation to dive deeper.

It did make me uneasy ignoring such a common refrain, so I create a version with a static plot 1, and another that left out plot 1 altogether, to compare. Still preferred keeping an animated plot 1.

(I did NOT send them back to those who gave the feedback, because it's Christmas and I want to keep them as friends)

Lag time for 'Play' button

While only 1 person complained about it, it bothered me too. Unfortunately I could not figure out how to fix it. I tried to add a .transition.delay() to this series & stories variables, as well as through D3, but it didn't work.

Resources

• Dimple documentation https://github.com/PMSI-AlignAlytics/dimple/wiki
• Inspiration for the first design http://dimplejs.org/advanced_examples_viewer.html?id=advanced_bars_sketchy
• Example of storyboards in action, plus how to create onClick events: http://dimplejs.org/advanced_examples_viewer.html?id=advanced_storyboard_control
• bug with negative values in mouseover: PMSI-AlignAlytics/dimple#204
• Creating a play button https://bl.ocks.org/officeofjane/47d2b0bfeecfcb41d2212d06d095c763
• Information on overriding Dimple inline styles: https://stackoverflow.com/questions/27830560/custom-tooltip-color-and-font-size-in-dimple-js
• Writing reusable javascript: https://medium.com/@_kamerontanseli/writing-reusable-javascript-d220aa501e67
• Seeing what's available in an object: https://stackoverflow.com/questions/957537/how-can-i-display-a-javascript-object
• D3 transistions: https://bl.ocks.org/Kcnarf/9e4813ba03ef34beac6e