So creating a model is a good idea, but it is not the magic bullet. >If you really want test data in the same proportions and distribution as the >production database, it is best to make a copy of the production database and >desensititize it--scramble the SSAN or other private information in ways >that do not affect the distribution.