Date of Completion

2018

Document Type

Honors College Thesis

Department

Computer Science

Type of Thesis

Honors College

First Advisor

James P. Bagrow

Keywords

crowdsourcing, causality, priming, inter-task effects, mechanical turk

Abstract

Online microtask crowdsourcing platforms act as efficient resources for delegating small units of work, gathering data, generating ideas, and more. Members of research and business communities have incorporated crowdsourcing into problem-solving processes. When human workers contribute to a crowdsourcing task, they are subject to various stimuli as a result of task design. Inter-task priming effects - through which work is nonconsciously, yet significantly, influenced by exposure to certain stimuli - have been shown to affect microtask crowdsourcing responses in a variety of ways. Instead of simply being wary of the potential for priming effects to skew results, task administrators can utilize proven priming procedures in order to promote project goals. In a series of three experiments conducted on Amazon’s Mechanical Turk, we investigated the effects of proposed priming treatments on worker retention, response quality, and response diversity. In our first two experiments, we studied the effect of initial response freedom on sustained worker participation and response quality. We expected that workers who were granted greater levels of freedom in an initial response would be stimulated to complete more work and deliver higher quality work than workers originally constrained in their initial response possibilities. We found no significant relationship between the initial response freedom granted to workers and the amount of optional work they completed. The degree of initial response freedom also did not have a significant impact on subsequent response quality. However, the influence of inter-task effects were evident based on response tendencies for different question types. We found evidence that consistency in task structure may play a stronger role in promoting response quality than proposed priming procedures. In our final experiment, we studied the influence of a group-level priming treatment on response diversity. Instead of varying task structure for different workers, we varied the degree of overlap in question content distributed to different workers in a group. We expected groups of workers that were exposed to more diverse preliminary question sets to offer greater diversity in response to a subsequent question. Although differences in response diversity were revealed, no consistent trend between question content overlap and response diversity prevailed. Nevertheless, combining consistent task structure with crowd-level priming procedures - to encourage diversity in inter-task effects across the crowd - offers an exciting path for future study.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.

Share

COinS