Census PUMF Methodology Question

Question

I have received the following question regarding a few older Census Individuals PUMF methodology:

"The Canadian Census documentation mentions non-sampling errors. One of the sources of non-sampling errors being imputation for non-response. However, I could not find any info regarding imputation methods used for any of the variables in the documentation for Canada census in 1981, 1991, 2001.

Can one identify the observations for which data were imputed, and the imputation method used in Canadian census 1981, 1991 or 2001? How?"

I checked the 2001 Census Individuals PUMF users guide (second revision), page 197 (202 of 292), where it does briefly discuss Imputation Methodology. (A similar note shows up in the 1991 documentation, Wordperfect format). The example given is a geography variable, but there are no details of other possible affected variables, and nothing about identifying the affected observations. Any advice or explanation which I can pass on to the user will be greatly appreciated.

Answer

I consulted our census contacts and they indicated that it "is impossible to identify specific Census data that were imputed. The main reason is confidentiality and the fact that imputed data are incorporated in larger groups of data". They also indicated that "We have no way of knowing the number of variable affected by imputation. All of them could be affected or only a limited number. The impacts are generally greater when we look at income and less important when it comes to minor aspect like age." "After the imputation has been done initially on the census, no further imputation is done on the PUMF."

They directed us to the following general overview which may be of interest to your client: http://www.statcan.gc.ca/edu/power-pouvoir/ch3/imputation/5214784-eng.htm

While looking into your question, I also came across references to imputation in the 2001 Census technical reports (ex. http://www12.statcan.gc.ca/english/census01/Products/Reference/tech_rep/age_sex/processing.cfm#edit) and in the 1991 census dictionary (I searched the online version on the site at the University of Toronto).

I hope that this information is helpful.