Task Workflow Design and its impact on performance and volunteers' subjective preference in Virtual Citizen Science

- Citizen science participants prefer task designs with greater variety and autonomy.
- This does not translate into better performance for either data quality or quantity.
- Simpler task designs led to a greater volume of data being produced.
- Interfaces that afforded the most direct capture of data led to better quality data.
- Trade-offs exist between data volume and accuracy and between preference and productivity.

Virtual citizen science platforms allow non-scientists to take part in scientific research across a range of disciplines. What they ask of volunteers varies considerably in terms of task type, variety, user judgement required and user freedom, which has received little direct investigation. A study was performed with the Planet Four: Craters project to investigate the effect of task workflow design on both volunteer experience and the scientific results they produce. Participants' feedback through questionnaire responses indicated a preference for interfaces providing greater autonomy and variety, with free-text responses suggesting that autonomy was the more important. This did not translate into improved performance however, with the most autonomous interface not resulting in significantly better performance in data volume, agreement or accuracy compared to other less autonomous interfaces. The interface with the least number of task types, variety and autonomy resulted in the greatest data coverage. Agreement, both between participants and with the expert equivalent, was significantly improved when the interface most directly afforded tasks that captured the required underlying data (i.e. crater position or diameter). The implications for the designers of virtual citizen science platforms is that they have a balancing act to perform, weighing up the importance of user satisfaction, the data needs of the science case and the resources that can be committed both in terms of time and data reduction.