III: Medium: Counterfactual-Based Supports For Visual Causal Inference

数据可视化是一种重要且无处不在的工具，用于支持跨各种领域的数据分析任务。可视化的价值在于它们能够以图形方式“显示数据”，而不是使用字母和数字，使用户能够为他们所看到的内容分配意义。这反过来又可以帮助用户分析复杂的数据、发现新的见解、做出数据驱动的决策，并与其他人就他们的发现进行沟通。因此，这些发现的正确性显然取决于用户在查看或与数据可视化工具交互时所做推断的正确性。然而，最近的研究表明，人们经常将可视化模式解释为数据中变量之间因果关系的指标，即使不存在因果关系。其结果是，可视化可以极大地误导用户得出错误的结论。该项目开发了一种新的可视化方法，基于反事实推理的概念，旨在帮助用户在使用可视化工具分析数据时得出更准确和更普遍的推论。该项目的成果，包括开放源码软件，旨在广泛适用于各个领域。此外，该项目还将使用人口健康领域的数据和用户进行评估，这些数据和用户有可能为改善人类健康做出贡献。更具体地说，该项目将开发一套创新的以反事实为中心的可视化方法。在认识到用户在查看数据可视化时对数据进行因果推断的自然倾向时，这些方法将直接旨在减轻得出错误结论的风险，同时增强用户稳健地发现更可能是统计支持的因果交互的指标的模式的能力。基于反事实推理的原则，这个项目将实现三个关键目标。首先，将开发方法来增强传统的过滤器驱动的可视化与反事实子集的比较。目标是为用户提供所需的信息，以便从可视化数据中得出更可靠的结论。其次，将开发方法来利用来自这些反事实子集的统计数据，以帮助指导用户的探索活动，目的是提高发现的效率。第三，将开发一个工作流程，用于识别和解释与用于反事实比较的变量相关的次要变量。该项目将导致新的计算方法和用户工作流程的设计和开发，实现这些贡献的开源软件，以及将表征这些基于反事实的技术的有效性的评估研究。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估来支持。

Data visualization is a critical and ubiquitous tool used to support data analysis tasks across a variety of domains. Visualizations are valued for their ability to “show the data” graphically, rather than using letters and numbers, in a way that enables users to assign meaning to what they see. This in turn helps users analyze complex data, discover new insights, make data-driven decisions, and communicate with other people about their findings. The correctness of these findings is therefore clearly contingent upon the correctness of the inferences that users make when viewing or interacting with a data visualization tool. However, recent studies have shown that people often interpret visualized patterns as indicators of causal relationships between variables in their data even when no causal relationships exist. The result is that visualizations can dramatically mislead users into drawing erroneous conclusions. This project develops a new approach to visualization, based on the concept of counterfactual reasoning, designed to help users draw more accurate and generalizable inferences when analyzing data using visualization tools. The project's results, including open-source software, are intended to be broadly applicable across domains. In addition, the project will be evaluated with data and users in the population health domain with the potential to contribute to improvements to human health.More specifically, this project will develop a set of innovative counterfactual-centered methods for visualization. In recognition of users' natural tendency to draw causal inferences about data while looking at data visualizations, these methods will directly aim to mitigate risks of drawing erroneous conclusions while amplifying users' ability to robustly discover patterns that are more likely to be indicators of statistically supported causal interactions. Building upon the principles of counterfactual reasoning, this project will achieve three key aims. First, methods will be developed to enhance traditional filter-driven visualizations with comparisons against counterfactual subsets. The goal is to provide users with the information required to make more robust conclusions from visualizing data. Second, methods will be developed to leverage statistics derived from these counterfactual subsets to help guide user's exploratory activity with the aim of increasing efficiency of discovery. Third, a workflow for identifying and accounting for secondary variables that correlate with those used for counterfactual comparison will be developed. The project will result in the design and development of new computational methods and user workflows, open-source software implementing these contributions, and evaluation studies that will characterize the efficacy of these counterfactual-based techniques.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.