As anyone who does extensive Census research will attest, there are a number of "errors" on the sheets. Since humans do the enumeration, there are all sorts of errors of hearing names, writing names, transcribing names, remembering dates, etc.
As to the "undercount".... an attempt was made for the 1940 Census to estimate how many people were not counted in a specific age group...namely the sample of men that could join the armed forces. The estimate was that 2.8 to 3.1% were missed, but jumped to 13% for blacks. See pages 103 and 104 of:http://www.ipums.umn.edu/usa/voliii/enumproc1940.pdf