1504.0 - Methodological News, Sep 2009  
ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 18/09/2009   
   Page tools: Print Print Page Print all pages in this productPrint All

Modelling Probability of Response in ABS Household Surveys

The Operations Research Unit (ORU) has conducted a study to look at the effects of area characteristics, household characteristics, interviewer characteristics and survey design features on response rates, by modelling the probability of response using paradata from the ABS Monthly Population Survey (MPS). This study is part of the work on developing a non-response framework for ABS household surveys (an article on this framework was published in the December 2008 edition of Methodological News).

The recent availability of paradata (i.e. data about the process) from ABS surveys allows the analysis of a richer set of information about statistical data collection activities, which can then be used to help make informed decisions about operational efficiencies. The ORU developed a logistic regression model based on interviewer call record data to predict the likelihood of response at each call. The model can be used to identify improvements to follow-up strategies of non-respondents. For example, results from the model can be used to design effective and efficient interviewer calling strategies.

Our research included households selected in the MPS during September to December 2007 (our model was tested on January and February 2008 MPS data). The MPS is conducted by face-to-face and telephone interviewing and the available paradata included records of calls (such as, day and time of call and the outcome of the call) and information about the interviewers making those calls.

A household was considered to have responded if it was a fully responding household. Probabilities of response were modelled separately for face-to-face households, telephone interview households and combined face-to-face and telephone interview households and for initial and follow-up workloads. The factors found to have a significant effect on the probability of response were state/territory, region (met/ex-met), whether the household was first month in sample, nth call attempt, workload size, age of interviewer, interviewer experience, interviewer performance (previous individual response rate), day of call and time of call.

Findings from our response model indicate that:

    • in initial face-to-face workloads, households in all states/territories (except for ACT) are more likely to respond than households in NSW;
    • higher performing interviewers (i.e. higher individual response rates) are more likely to obtain a response than lower performing interviewers;
    • Monday is the most successful day in obtaining a response;
    • the best time for the second call is a weekday morning, regardless of the time of the first call; and
    • while interviewers seem to make the majority of their calls in the afternoon, the highest response probabilities can be found for morning calls during weekdays and for afternoon calls during weekends.

Our research has given us a better understanding of the relationship between operational procedures and response rates. Our aim now is to understand how changes in operational procedures will impact on costs, response rates and the quality of survey outputs, so as to lead to making informed decisions about operational efficiencies.

For further information, please contact Rosslyn Starick on (03) 9615 705 or rosslyn.starick@abs.gov.au.