Details and Examples: PARETO Procedure

Constructing Effective Pareto Charts

The following are recommendations for improving the visual clarity of Pareto charts:

  • Decide carefully how the bars should be scaled. The default percent scale is not always the best choice. For instance, a count scale may be more appropriate in a comparative Pareto chart where the total count per cell varies widely from cell to cell and where you want to compare Pareto distributions on an absolute scale rather than a relative scale. You can request a count scale by specifying SCALE=COUNT. In other situations, it may be more appropriate to use a weighted percent scale or a weighted count scale (specify a WEIGHT= variable and either SCALE=PERCENT or SCALE=WEIGHT).

  • Use a weight variable if the counts are dependent on a factor such as exposure or opportunity that varies from one category to another. For instance, suppose that you are creating a Pareto chart for the number of medical claims submitted by company employees categorized by job title. The counts can be weighted to adjust for the fact that there are more individuals in some jobs than in others and for the fact that some jobs may be associated with greater health risks than others.

  • Use the NOCURVE option to eliminate the cumulative percent curve in situations where the curve reveals little information about the data. In general, the bars should be more prominent than the curve.

  • Maximize the space used for the bars by eliminating unnecessary labels and visual clutter. This is particularly important for comparative Pareto charts. The NOHLABEL and NOVLABEL options are useful for this purpose. You can also use the NOVLABEL2, NOVTICK, and NOVTICK2 options with a VBAR statement or the NOHLABEL2, NOHTICK and NOHTICK2 options with an HBAR statement.

  • Make legends more informative by specifying legend labels.

  • Avoid filling bars with multiple types of cross-hatched patterns; solid color fills are less distracting. Use color sparingly to emphasize important features (such as the vital few categories), and choose bar colors that provide good visual discrimination.

  • If you are working with a large data set involving many categories, limit the number displayed to achieve visual clarity.

  • If your application involves classification effects, construct more than one Pareto chart for the data using various combinations of classification variables (this approach is illustrated in Example 15.2.

  • Provide reference lines on comparative Pareto charts to aid visual comparison.

Refer to Chapter 2 of Cleveland (1985) for a general discussion of the principles of statistical graphics.