Browsing by Subject "algorithmic reactions"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Algorithmic Reactions and Task Characteristics in College Admission
    (2022-04) Tran, Khue
    The main objective of this dissertation was to investigate factors that affect decision-makers' trust in and reliance on algorithmic predictions as decision aids in the context of college admission prediction tasks. College admission officers often made predictions about the applicants' future success based on multiple pieces of available information, such as applicants' standardized test scores and high school GPA. While research showed that admission officers could benefit from algorithmic decision aids to make admission decisions, it remained unclear how people perceived the competence of the advice from a mechanical approach in this type of prediction task. The dissertation examined the interplay between sources of advice (whether advice comes from a college admission expert or a statistical model) and the characteristics of the prediction task (i.e., the task complexity and the cost of prediction failures). The results from the two studies suggested that people trusted the advice coming from a human expert more than the same advice from a statistical model. When people received more information about the candidates, they tended to rely more on the advice. In Study 1, participants in the High-cost condition, who read the scenario that failure to select the right students would have detrimental effects on the University, were less likely to trust the advice. Study 2 explored people's perceptions of algorithmic judgments – judgments made from a statistical model. People across conditions preferred human judgments to statistical models or algorithmic recommendations in selection and admission situations. The findings suggested that people tended to rely on some advice as a decision aid when predicting future college GPA based on multiple pieces of information. However, they remained somewhat skeptical of advice coming from a statistical model and preferred human experts making decisions in college admission or personnel selection contexts.