Following my last post - Word Clouds - love 'em or hate 'em?
The New York Times shared a text visualization that inspired much conversation here at VC. You can view it here.
(Works best on a browser with SVG support, such as Chrome, Firefox, Safari or IE 9.)
There are a number of improvements that this graphic makes over the traditional word cloud that I wrote about. Here is a summary of why we here at VC think this is great and why a similar style could help a researcher provide insight.
- Augmenting text size with background circles mitigates the potentially skewed perception of longer/shorter words in a word cloud.
- The background circles additionally show the proportion of the two groups using the word (blue and red as well as in text) adding to the analysis. If this concept was applied to text responses from community panel members - instead of Democrats and Republicans, it could easily be buyers and non-buyers or favourable and unfavourable opinions of a concept/brand/product.
Using pie charts could even show usage among more than two market segments.
- By arranging these circles along the horizontal axis representing usage frequency between the two groups amplifies this effect.
- While selected words are preset to display, the graphic includes the ability to add words interactively to the visualization to test a hypothesis or drill into themes.
- The excerpts below the visualization show the words in context allowing a researcher to better understand the sentiment. Clicking on a word highlights and filters the excerpts adding to the speed of analysis.
We did notice at least one drawback. This concept works well as long as the exact positioning of the words doesn't matter. The underlying forces that pull the circles to one anchor or another on the horizontal scale are not exact. For example Jobs and Hope line up horizontally and are positioned closer to the right which implies that Republicans tend to use these words more than Democrats. However, when comparing the percentages (Jobs: 52.4% vs 47.6% and Hope: 47.2% vs 52.8%) we see that Jobs should be positioned more toward the left since more Democrats use the word Jobs than Republicans.
We also noted that the positioning of the circles will get even trickier with more anchor points (as in market segments or in a plane with two scales) and could muddy interpretation if not addressed carefully.
This example clearly takes us a step closer to a text visualization that gives insight instead of just making open ends pretty. If you run or own a community panel, can you see yourself using a visualization like this? What else would you like it to do for you?