The CSPro program should be tested using real data with perhaps 2,000 to 3,000 records. The edit statistics should be examined closely. The number of imputations should be a very small percentage of records. If a large number of imputations are being done for any particular edit test, the code should be examined, and the imputations accounted for before proceeding. The edit statistics should be examined by subject-matter specialists also, so they can determine whether or not the rate of imputation is too high. If appropriate coding techniques are used with the
impute statements, the imputations can easily be related back to the edit specification document.
Remember: Just because the CSPro program is free of syntax and logic errors is no guarantee that all the required editing is being done and is being done correctly! An error-free CSPro program cannot compensate for incomplete edit specifications. Among the test most commonly omitted tests when creating edit specifications are: making sure there is one and only one head of household, and making sure the head of household is of an acceptable age.