Automatic Allocation of Interviewer Work Using Operations Research Approach
Population Survey Operations maintain a workforce of field interviewers to approach Australian householders to conduct surveys. Each month, PSO allocate each interviewer a "workload" of household locations for personal and/or telephone interviews. Household surveys can either be regular (Monthly Population Survey MPS) or irregular one-off and lasting for one or several months (Special Social Surveys SSS).
Interviewer workload allocation is subject to hard restrictions such as number of available interviewers, dates when interviewers are available for duty, geographical location of interviewers and workloads, training for particular surveys, and the horizon period within which a survey collection must occur, and soft restrictions such as giving an interviewer a sufficiently large workload to make it worth their while. Currently, a laborious and manual one-by-one allocation process is employed by well-practiced PSO staff. After the first cut allocation is set, there is also a reallocation task to be performed regularly when a condition changes (eg an interviewer drops out), as well as to allocate follow-up visits for non-respondents, until the collection period is finished. Some agencies such as Statistics Netherlands employ computer tools to aid officers doing manual allocation (e.g. the Address X Interviewer System - AXIS).
Operations Research and Process Improvement (ORPI), in partnership with Population Survey Operations, have developed a prototype Automatic Workload Allocation Tool (AWAT). Olena Gavriliouk (CSIRO) took an integer programming approach to minimise cost (distance travelled). Olena modelled the workload allocation procedure as a generalised assignment problem (GAP) with capacity constraints. AWAT currently contains three modules ("First Cut" FCWAP, "Reallocation" WRAP, and "Combined" WAM) which can be used for different types of workload allocation performed by PSO. Olena implemented AWAT in C++ and employs a commercial optimizer Gurobi to find (near)-optimal workload allocations.
The AWAT has been tested on selected SSSs, with the resulting allocations being at least as cost-effective as manual allocations. The AWAT runs in seconds, or up to minutes for the highest-dimension cases involving the most interviewers and household locations. Manual allocations for surveys with smaller samples are already near-optimal, and so the major advantage of using AWAT is for speed. However, for larger surveys which are more complex to allocate efficiently, AWAT saves on the order of several thousand kilometres in aggregate distance travelled for "First Cut" allocations.
AWAT field trials will continue to test out additional features, such as indicating which interviewers to train for which survey, and test out complex situations where multiple different surveys are running concurrently. The field trials will establish impacts on business processes and uncover hidden benefits or costs, with a view to an automated workload allocation future.
If you would like more information about the AWAT and its modules, Gurobi, or field testing, please contact firstname.lastname@example.org or email@example.com