|Page tools: Print Page Print All RSS Search this Product|
Modelling Business Provider Response Behaviour: A Survival Analysis Approach
Understanding business survey data provider's response behaviour is an important consideration for selecting an efficient data collection procedure. If providers are contacted too infrequently, they may not respond; however if they are contacted too frequently, the data collection procedure may be inefficient. In more extreme cases, contacting a provider too many times (or via inappropriate channels) leads to excessive respondent burden and, potentially, provider complaints and non-response. Therefore, in order to allocate resources effectively, we need to understand a provider's reaction to our attempts to obtain their cooperation.
ABS business surveys are typically conducted as mail-out, mail-back collections, which are supplemented by both written reminder letters and telephone follow-up calls. While every provider who has not yet responded will typically receive a reminder letter at the same time, telephone contact is more costly and therefore is prioritised on the basis of the provider's significance to the estimate and on their expected level of cooperation. For example, a very large business with a poor response history is much more likely to be contacted than a smaller business with a good response history.
The survey response process can be regarded as a survival process to attempt to answer: what is the fraction of the whole sample of businesses who will response before a certain time? We commence by posting out forms, which the provider is able to return at any time. After a certain period of time, if they are still not responding, we begin intervention - telephone and written reminder contact. The probability of response, and time to response, are dependent on both these interventions (which are time-dependent covariates, as they change value through the collection period) and on demographic information such as business size (which are generally fixed). We cannot simply ignore the impact of time, since we are interested in selecting appropriate timing for the follow-up, and also because we have censored data. In our case, censored data arises because although a response would eventually occur, we do not observe the response time, either because we have ceased follow-up, or because the covariate values have changed. Therefore, we use a survival analysis model where we are modelling the time to response.
Some key results of this analysis (for a selection of ABS business surveys) include:
Further research is underway to refine the results and apply these to selecting the most appropriate forms of follow-up procedures.
For further information, please contact Melanie Black on (02) 6252 7241 or email@example.com.
These documents will be presented in a new window.