Outliers

Popular Tests I Outliers 

• Population-Level Testing

  A common criterion for outliers is that the values are more than two standard deviations above the average for a given population. The Statistics command can quickly generate the mean and the standard deviation:

  

  To identify outliers, create a filter:
  Total_Cost > 1491 + (2 * 3155.49)

  
  
  

• Single-Category-Level Testing

  It's also possible to identify outliers at a category level, such as vendors, using the enhanced Summarize command:
  
  1. In the Summarize command, choose the Vendor field as the key field.
  2. Open the "Fields to process" dialog
  3. Select "Amount" twice
  4. Change the Type to AVG and STDDEV
  
  
  

  

  The output file contains the mean and the standard deviation for each vendor:
  5. Create a computed field for the 2 SD threshold:
  AVG_Amount + (2 * STDDEV_Amount)

  
  6. Open the transaction file.
  7. Join to the vendor threshold file and add the threshold field to the output.
  8. Filter for Amount > Vendor_Threshold
  

   

• Multiple-Category-Level Testing

  It's also possible to add multiple levels to the category testing. For example, you could test by Vendor-Product combinations to test for pricing consistency.
  
  Follow the same steps as in the previous example, using the Vendor and the Product ID as the key fields.