Consistency edits look at the individual data items within a questionnaire or case, and examine the validity of each item and the consistency of each item with respect to other related items. For example, the degree of educational attainment of an individual should have some predictable relationship with the person's age. This means that it is not expected that a 9-year-old child will have progressed much beyond the fourth or fifth year of elementary school, depending on local standards. If, in the unedited census data, a 9-year-old indicates educational attainment above that level, it is probable that either the age or the educational level have been incorrectly recorded. Whether the age or the educational level is modified to produce data consistency is a decision left to the subject-matter specialists, who should know whether age data or education data are more correctly reported. It is also important to take into account the method of data capture; some, like scanning, are prone to systemic error (that is, where a "0" is consistently read as a "9," for example). If such error is not recognized and taken into consideration when assessing the reliability of reported information, it can lead to even greater error in the final data.
At a minimum, consistency checks should seek to resolve all errors which might eventually lead to doubts about the quality of the data. That is, the subject-matter specialists must consider the types of tabulations to be produced and the uses to which the data may be put in the future. For example, if the plan includes tabulations of educational achievement by age and sex, the edit specifications must include a means of detecting and correcting the data for individuals whose declared educational achievement is not in line with their declared age (as in the example of the preceding paragraph). If a published table shows even one 9-year-old having completed secondary school, the quality of the data will be called into question, simply because (in that particular culture) it is not the norm that a child so young could be so advanced educationally. It makes no difference if, in fact, the child is a prodigy and did indeed complete secondary school; when it is a case of one or two "outliers" and the weight of probability is against them, the values must be changed and brought in line with reasonable expectations. It is not recommended that such changes be made only in the logic which generates the tables; if the data (even a subset) are to be given out to researchers or other users outside the statistical office, they must be clean and consistent throughout before distribution.