x

Guidelines

• The essential question in data analysis is: Compared to what? To understand your data, you need to place it in a context.
• Most interesting problems are essentially multivariate: there are a variety of factors that interact. Include many dimensions of data in your analysis.
• Follow your data when you are analyzing by changing the display, the data, or the charts to show you new relationships as you try to understand what you are seeing.
• A design strategy that is often effective is: Overview first, zoom and filter, then details-on-demand ([HCIL]). Provide an overall chart that covers the problem space. You should try to use a multidimensional chart for the overview (e.g., scatter plot, Multiscape, Parabox, Data Constellations, Map, Time Table). Use filtering and zooming to focus on a subset, then explore that subset by examining its details.

When you are designing a visual display of your data, consider these factors:

• Show as much data as you can: a chart can comfortably accommodate one to two orders of magnitude more data than a spreadsheet can.
• Extract data at the granularity of items that you want to understand. For example, if the raw data is transactions, aggregate to higher level entities (e.g., products, people, relationships).
• First choose charts that directly reflect relationships you are interested in:
  • Characteristics of a field: Bar, Pie, Line Chart, Histogram
  • Interaction of two things: Scatter Plot, Multiscape
  • Subsets over multiple fields: Parabox
  • Outliers over multiple fields: Parabox
  • Interactions between multiple fields: Data Sheet, Parabox
  • Relationships: Data Constellations
  • Events over time: Time Table
  • Geospatial data: Maps
  • Details about data or subsets (often used for the details on demand): Data Sheet
• Increase the dimensionality of your analysis by combining multiple charts into a page. Selection and coloring increase the number of dimensions that can be correlated. Filtering provides drill down to focus on a subset. (See Visual Discovery AE for more details.)

Too many Charts on a Page?

Here are things to try if you find your Page to be unable to fit all of the charts you think you need:

• Reexamine objective of the Page:
  • What questions/problems is it trying to address?
  • Are all charts needed in that Page?
  • Are there other Visuals that might be as good (or better) with less space usage?
• Use multiple Pages
• Use Data Hierarchies
• Orient bars in Bar Charts sideways (except for time-related dimensions)
• Reduce font sizes (especially titles)
• Eliminate titles if redundant w/axes labels
• Align filters in two columns on left and right or along left and bottom
• Use Full Screen button (client tool only)

x

Ranking with Filters. The primary focus is on a measure (like sales) broken down by a categorical grouping like product line. Typically, data is grouped in bar chart and/or pie chart. The Bar Chart or Pie Chart provide the grouping; other supportive components are included for slicing, filtering, and details.

Multiscape with filters. The primary focus is on a measure broken down by the cross-product of two groupings (example: quarterly sales by region). Data is grouped in a Multiscape chart. To see trends, you can rotate, zoom, or scale the Multiscape.

Maps. The focus of this page is where. Data points are plotted on an image or vector map with height representing a measure. Optional links highlight relationships between geographical entities. To see trends, you can rotate, tilt, zoom or scale the map as required.

Relationships with filters. Use this type of page to view abstract structures or relationships (examples: product affinities, organizational structures). Data points are displayed in a graph hierarchically or in a mesh. Node size/color/shape can represent different entity attributes. To see trends, you can rotate, tilt, zoom or scale the graph as required.

Correlations with Filters. Use this type of page to view correlations between data items with size, color, shape for displaying attributes and univariate displays for understanding other attributes and for filtering.

Advanced correlation with filters. This page type can be used to correlate multiple metrics and/or multiple dimensions in a single chart. A Parabox is used to display data.

Multiple relationships. Interactions of multiple fields are shown via a Parabox; geospatial aspects are shown via a map. Univariate displays provide filtering, and a Data Sheet provides details.

KPIs with filters. This page is very effective when you want to focus on KPIs across multiple dimensions. Charts used in this type of page include: Line Chart, Histogram, and Scatterplot.

Multiple univariate displays. Multiple univariate displays show multiple fields. Coloring and selection show interactions between dimensions. Use the normal bar charts to select a subpopulation. The Pie Chart and spine plot variety of Bar Chart provide comparison of that subset of the population to the whole population as a fraction of the whole.