Imputation refers to the process of providing values for missing, erroneous, or inconsistent responses. For example, if a person's sex code is invalid (i.e., out of range or otherwise unacceptable) or missing; then an appropriate response should be substituted.
You may decide that for missing data, you'd rather just keep it "missing" and publish your tables with an extra column (or row) for unknown values. This is a very cumbersome method, however, as the number of missing values will vary for each data item, and so the number of missing entries will vary from table to table, making the data difficult to analyze.
Inconsistent responses occur when a response yields an impossible situation with respect to another response. For example, if a 5-year-old female reports having children, either her age is wrong, or her fertility data are wrong (i.e., that section should be blank). This type of error must be corrected, as your users will place very little faith in the quality of your data if this type of condition becomes evident in the tabulations. Many users also do not look kindly on "missing" or unreported data. Of course, nothing can correct for bad data, and if you find that a significant amount of your data are bad (poorly designed questionnaire, inadequate field procedures, inattentive coders and keyers, etc), you may want to reconsider whether the data should be released at all.
Procedures have been developed to provide the missing information, thereby avoiding discrepancies and the need to determine percentages twice (with and without unknowns). For a detailed discussion on using imputation and the methods available to you, please refer to the United Nations Handbook on Population and Housing Census Edits
Essentially, two methods of imputation are available: static