How can I show marking state (for individual data values) in aggregate representations?

Question

I am working with a data-visualization application where users can view individual data items and aggregate values. Users can also "mark" individual data items; the semantics of marking are simply "this user flagged this as interesting", and cannot otherwise be derived from the data. A user can choose a marking color and that color itself might have informal semantics (like red=bad).

When viewing an individual datum it is easy to convey the marking state visually. Now consider an aggregate view, such as a bar, mark, or piece of text (like a number) representing a collection of data. For example, consider the reputation bar chart on your Stack Exchange user page and imagine that you had marked some of your questions and answers and perhaps moderators had marked others. Suppose you want to know which bars (daily totals) those questions factored into.

What is the best way to convey, in an aggregate view, that some of the data contributing to a particular grapheme (bar, mark, text, line, etc) is marked?

If a grapheme represents a small number of data items, you could do some sort of proportional coloring. (This might not work so well for text and lines.) But what if a bar represents 1000 data items, two of which are marked, one each in two different colors? A proportional representation would get lost; there aren't enough pixels to show that. If only one marking color were in play then some annotation like outlining could work, but how would that work with multiple colors? (Is there a practical upper bound? Two is probably visible, ten probably aren't.) Is it better to leave the grapheme alone and use some sort of call-out scheme?

Answer 1

Dynamic brushing is one technique already used in data visualization. However, there are only two marker states: 'of interest' and 'not of interest'.

The representation of marked data in aggregate views depends on the aggregate view.

Pie charts, bar charts, and histograms, using shading in the slice/bar to show the proportion of that is selected in an individual datum view.

Answer 2

A large part of the correct answer depends on what you want the data visualizations to accomplish for the user. Usually a representation of large data is accompanied by labels that not only describe its contents but also serve as filters, allowing users to click through to data associated with a particular mark or status. Think of email. (This is a textual, rather than visual, example, but I think it applies.) When I log into Gmail I see:

• Inbox (89)
• Some Label (4)
• Another Label (3)

This lets me not only see a single representation of my inbox in numeric form, but it also shows me that my inbox is comprised of 4 messages and 3 messages associated with two different labels. I can click those to enter detail-view for just the messages belonging to those labels.

Is this getting at the problem at all?