Diagnostics (cont.)
Analysis of editing and imputation rules (cont.)
While the data not subjected to editing does indeed show a roughly uniform distribution, the data created by the dynamic imputation rule shows a clear bias against month of birth being allocated to November or December. This provides strong evidence that the algorithm used to allocate month of birth could be flawed, and should be investigated.
Would your interpretation of these results change if you were told that the census date for this census was in October 2001? Think about your answer before clicking to reveal the answer below.
Yes, one’s conclusion about the operation of the editing rule might change. It may have been the case, for example, that if month of birth was imputed using dynamic imputation, but constrained to ensure that month of birth was consistent with both the reported year of birth and reported age at the census date, then the results would follow this distribution. For example, if a woman reported her age to be 43, and she reported her year of birth to be 1958, then her birth month would have had to be chosen to be a month between January and October. Likewise, a woman reporting her age to be 42, with the same reported birth year would have to have her birth month allocated to November or December. The problem with such an approach, however – and manifested in these data – is that reported age and reported year of birth may not have been derived independently. A respondent may, for example, have ‘calculated’ her age or year of birth by differencing one or the other from the year in which she was being asked the question (in this case, 2001), and where a respondent (actual or proxy) did not know the woman’s month of birth.