|Page tools: Print Page Print All RSS Search this Product|
Targeting non-response follow-up in a household survey – a geographic approach
Visiting households to obtain survey responses is very expensive, so it is appropriate to apply follow-up effort where it will have the most benefit. In a Special Social Survey (a large survey on a particular social topic), data collection extends over a long period, so it may be possible to use data from early-responding households to target further follow-up.
The responsive design team in Statistical Services Branch has considered and extended the existing methods for assessing bias reduction due to follow-up. One finding is that follow-up reduces bias when it improves the response rate for those types of dwellings or persons that would otherwise be under-represented in the sample. Conversely, increasing response among the well-represented types could actually increase bias. The team has proposed a strategy in which a basic level of follow-up is applied to all areas, but extended follow-up is targeted to avoid areas of types that are already well-represented after basic follow-up and that have few non-responding dwellings.
This approach is being tested using a case study approach based on data from the 2007-08 National Health Survey. In this study the impact on estimates of reducing the number of contact attempts or callbacks in various ways has been evaluated. The project demonstrates that there is a relationship between some survey estimates and the numbers of callbacks. It also shows that some characteristics of the areas being interviewed, as measured in a recent Census, are related to response in a way that is not adjusted for adequately by the survey weighting.
This leads to specific targeting strategies that favour extended follow-up in the low-response types of areas. The case study evaluates a number of such strategies and their effect on key estimates. The objective is to define a strategy that provides a sample that is similarly representative to that which is obtained by extended follow-up, or better, with a lower cost.
The next step for this work is to apply it to live surveys. This will involve working with survey areas and the population survey operations area to define appropriate follow-up scenarios for basic and extended follow-up, and to decide which scenario should be applied in each sampled area of the survey. We will also need to put in place monitoring to ensure that the resulting sample is representative. The aim will be to make major cost savings by reducing interviewer effort while producing valid and reliable estimates.
These documents will be presented in a new window.