Perhaps is because the dataset has been anonymized.
In general, datasets that contain personal information about individuals are anonymized before they are released.
That is, the records of the data base are stripped of any personal information (name, ID number, etc), and the demographic attributes (age, address, gender, country of origin, ethnicity) are distorted. By distorted I mean, that some of the attributes are modified (or generalized) such that no person can be undoubtedly linked to a single record in the data base.
Examples of modifying attributes could be remove the last digits of Zip Codes, or round the age to the closest multiple of 10.
An example to note the importance of anonymizing a dataset:
In [P1] it was shown in that 87% of the population in the United States may be unequivocally identified solely on the basis of the triple consisting of their date of birth, gender and 5-digit ZIP code, according to 1990 census data.
Those techniques are called statiscal disclousure control [P2], and there is a lot of literature in this regard. Notice that there is a tension between the protection of the privacy of the records in a dataset and the utility-loss of that dataset.
There are several privacy-measures that protect the privacy of the records of the data set against different kinds of attack that try to unambiguously identify records in the dataset.