A short look at Census imputation rates.
This post is intended to simply show some information from the 2016 Census Table Builder product. A more detailed and much longer coverage of this topic, which includes justification for my conclusions is at this post. But only go there if you really like detail about statistics.
The release of Table Builder includes some Imputation Flags which are set for the core demographic variables (Age, Sex, Marital Status) when the Collector identifies an occupied dwelling for which no form is received. Thus it is effectively an indicator of one part of non-response.
The importance of imputation is made explicit in the report of the Census Independent Assurance Panel (CIAP) where it is shown (Table 3.2.2) that the final under-enumeration rate for the Census is 1.0% being a balance between a 4.3% gross undercount of people on Census forms, a 1.3% gross overcount of people on Census forms and a net overcount of 2.1% of persons imputed (there is obviously a rounding effect in that sum).
From Table Builder I have compiled charts of age-imputation rates (number of age records imputed x 100/total number of age records) for a range of areas. Let us first look at some larger areas.
The release of Table Builder includes some Imputation Flags which are set for the core demographic variables (Age, Sex, Marital Status) when the Collector identifies an occupied dwelling for which no form is received. Thus it is effectively an indicator of one part of non-response.
The importance of imputation is made explicit in the report of the Census Independent Assurance Panel (CIAP) where it is shown (Table 3.2.2) that the final under-enumeration rate for the Census is 1.0% being a balance between a 4.3% gross undercount of people on Census forms, a 1.3% gross overcount of people on Census forms and a net overcount of 2.1% of persons imputed (there is obviously a rounding effect in that sum).
From Table Builder I have compiled charts of age-imputation rates (number of age records imputed x 100/total number of age records) for a range of areas. Let us first look at some larger areas.
The set of selected suburbs are those I have blogged about in the past and are specified two graphs down. A point which interested me was the apparent increase in imputation as areas get more "rural". The ABS has not yet released the data which shows results classified by Major Urban, Other urban and Rural so I looked at the Capital City and Rest of State values for NSW and Victoria.
In both cases the imputation rates are higher for the ex-Metropolitan area than for the City. This suggests to me that the mail-out/internet completion approach was not a problem in this regard. For reasons explained in the longer post I believe a major reason for the issue is unoccupied dwellings being incorrectly classified as occupied.
For the local area I posted previously about the Stoney Creek Gazette catchment area (Carwoola, Hoskinstown and Primrose Valley) Captains Flat, Wamboin and Bywong. Here are the rates for them.
At first glance Hoskinstown needs to spend some time in the Naughty Corner while Primrose Valley/Urila and Bywong get a gold star! However this may simply reflect there being a few weekender residences in Hoskinstown.
My overall conclusion is that the overall Census data for our area is useful. The number of people and the age distribution are probably pretty good, but I still suspect the number of dwellings is somewhat low and there will be quite high not-stated rates for the variables for which values are not imputed.
Comments