1.) What is household non-response: I would have thought it to be 100 minus the survey response rate?
2.) If the global non-response combines both household non-response and item non-response, how can it be smaller than the survey non-response rate? Is this a function of the weighting done (and, if so, how was the weighting done)?"
Answer
1.) Response from Census: "Yes, that’s true. For the inscope population (dwellings occupied by people who should be counted) – the response rate is the percentage of those dwellings who respond, and the household non-response is (100 – the response rate).
It should be noted that we provided different values of the response rate – see the following:
<http://www12.statcan.gc.ca/NHS-ENM/2011/ref/about-apropos/nhs-enm_r012.cfm?Lang=E> and the important value for our purposes is the weighted response rate.
Our raw (unweighted) response rate was indeed 68.6%, but the weighted response rate (after sub-sampling and non-response follow up) was 77.2%. This leads to a non-response rate of 22.8%. The weighted one is the more important one. "
2.) Response from Census:
"The issue that seemed of primary confusion to the questioner would be how our response rate can be 68.6%, (which obviously implies a non-response rate of 31.4%) and how our GNR (which includes that 31.4% non-response + further item non-response) could be smaller, at 26.1%. It can’t, obviously. And it isn’t. He needs to compare weighted response rate to GNR. The survey non-response rate he’s referring is unweighted in his email.
The 68.6% is the unweighted response rate (or 31.4% unweighted non-response rate). The weighted response rate is 77.2% (or 22.8% weighted non-response rate). The GNR is weighted. This 22.8% is less than the GNR of 26.1%. Users should compare weighted response rates (or weighted non-response rates) to GNR.
Please note– the linked page, particularly the footnote, describes in very rough terms the weighting, in particular the impact of sub-sampling. <http://www12.statcan.gc.ca/NHS-ENM/2011/ref/about-apropos/nhs-enm_r012.cfm?Lang=E>
The footnote reads as follows:
Unweighted response rates are calculated from final data, following the completion of data processing and data quality verification. The final status of a dwelling as respondent, non-respondent, or out of scope for the National Household Survey (NHS) is done as part of data processing. The unweighted response rates are then calculated as the number of sampled private dwellings that returned a questionnaire divided by the number of sampled private dwellings classified as occupied by field staff.
Weighted response rates are based on final design weights of NHS. Dwellings that responded to the NHS prior to a fixed date in collection have their initial design weight based on the sampling fraction in their area. After this fixed date, in order to limit the non-response bias as much as possible with the available resources, the NHS focused collection operations on a subsample of remaining non respondents. Within this subsample, the design weights were increased to reflect this change. The weighted response rates are then calculated as the weighted number of sampled private dwellings that returned a questionnaire divided by the weighted number of sampled private dwellings classified as occupied by field staff."
"The issue that seemed of primary confusion to the questioner would be how our response rate can be 68.6%, (which obviously implies a non-response rate of 31.4%) and how our GNR (which includes that 31.4% non-response + further item non-response) could be smaller, at 26.1%. It can’t, obviously. And it isn’t. He needs to compare weighted response rate to GNR. The survey non-response rate he’s referring is unweighted in his email.
The 68.6% is the unweighted response rate (or 31.4% unweighted non-response rate). The weighted response rate is 77.2% (or 22.8% weighted non-response rate). The GNR is weighted. This 22.8% is less than the GNR of 26.1%. Users should compare weighted response rates (or weighted non-response rates) to GNR.
Please note– the linked page, particularly the footnote, describes in very rough terms the weighting, in particular the impact of sub-sampling. <http://www12.statcan.gc.ca/NHS-ENM/2011/ref/about-apropos/nhs-enm_r012.cfm?Lang=E>
The footnote reads as follows:
Unweighted response rates are calculated from final data, following the completion of data processing and data quality verification. The final status of a dwelling as respondent, non-respondent, or out of scope for the National Household Survey (NHS) is done as part of data processing. The unweighted response rates are then calculated as the number of sampled private dwellings that returned a questionnaire divided by the number of sampled private dwellings classified as occupied by field staff.
Weighted response rates are based on final design weights of NHS. Dwellings that responded to the NHS prior to a fixed date in collection have their initial design weight based on the sampling fraction in their area. After this fixed date, in order to limit the non-response bias as much as possible with the available resources, the NHS focused collection operations on a subsample of remaining non respondents. Within this subsample, the design weights were increased to reflect this change. The weighted response rates are then calculated as the weighted number of sampled private dwellings that returned a questionnaire divided by the weighted number of sampled private dwellings classified as occupied by field staff."
