When looking at our members, we often have questions about how they break down among various dimensions: “Of all our users in India who have recently gotten free data, how many referred someone else?” “Of those, how many completed 3 or more offers?” “How does that compare to the same group in Brazil?”
We can pull the data from our analytics database, and create tables of percentages or rollup tables to query in various ways to answer these questions. But there’s something lost in percentages, and moreso in percentages of percentages. In an effort to let more people easily explore and get comfortable with our members, we hacked together a D3.js visualizer that would let you partition them visually. This gives you the ability to see immediately how many users meet any of a number of criteria, and then further partition those groups.
The above bl.ock is the visualization loaded with flag data for the countries of the world, rather than Jana members. It’s a small data set, but lets you play around with the visualization in a familiar domain. Click the link and follow along, if you like.
Let’s say you’re interested in the colors of African flags compared to the rest of the world. You can select “African country > 0” for the first criteria (boolean features are coded as 0/1, so >0 equates to True). You’ll see the box fill about 1/4 full, showing that 1/4 of all countries are African, while 3/4 are not. This represents that about a quarter of the countries in the world are in Africa.
Then if you select for the second criteria “Flag contains color green” you’ll see a much larger percentage of the blue “Africa” box get filled in than the grey, signifying that being in Africa is correlated with having green in your flag.
Change it to the “orange/brown” option, and you’ll see virtually no correlation– the likelihood of your flag having orange in it is independent of whether you’re African. Being able to visually explore these correlations can quickly shed some light on how different characteristics compare.
You might also discover interesting things in corners of your data you might not be actively investigating. For instance, let’s say we’re looking for flags similar to Brazil’s. How many have blue? Roughly 1/2. How many have blue and yellow? ~1/4. Blue, yellow, and green? ~1/8.
Looking at the look at the lower-left quadrant, you might notice that flags with blue but not yellow very rarely have green. Other combinations don’t seem to have as much impact.
Going a step further, if you further segment by whether a flag has text on it, you’ll see that while it’s not uncommon for Blue-Yellow-Green flags to have text, it’s almost nonexistent if those three colors aren’t present. Just five countries don’t have those 3 colors, and no country has writing on its flag without at least one of those colors.
Admittedly, this isn’t terribly exciting with the colors of countries’ flags. But discovering a correlation between characteristics and behavior of a segment of members can reveal interesting and often actionable details about how people use mCent.