Theorising in HCI using Causal Models

Research Background and Problem

• Problem Identification:
  HCI literature contains a wealth of theories but lacks systematic tools for theorizing. Existing methods of applying theories are highly complex and diverse, making it difficult for researchers to build upon prior work. Additionally, researchers often struggle with deep application of theories and bridging the gap between design and theory.
• Significance:
  Theories in HCI are not only meant to explain phenomena but also to help understand causal mechanisms and control events in practice. The lack of theorizing tools limits the development of richer theories in the field, directly impacting the depth and rigor of research and practice.
• Research Motivation and Related Work:
  This study is inspired by Karl Weick's concept of "theorizing," which emphasizes the process of theory generation rather than focusing solely on the end state of a theory. The authors aim to provide a methodological tool through causal models, particularly Directed Acyclic Graphs (DAGs), enabling researchers to clarify theoretical assumptions, explore causal mechanisms, and effectively integrate domain knowledge.

Solution

• Proposed Solution:
  Employ Graphical Causal Modelling as an effective theorizing tool to help HCI researchers construct and analyze causal relationships. DAGs (Directed Acyclic Graphs) serve as the core tool to assist in theoretical modeling, analyzing potential confounding variables, identifying intervention points, and explaining data generation mechanisms.
• Innovations:
  1. Introduce DAGs as graphical tools for systematically modeling causal relationships in HCI, making theoretical assumptions explicit.
  2. Provide step-by-step guidance for constructing causal models and demonstrate the applicability of this method across different stages of the research lifecycle (e.g., literature review, hypothesis formulation, study design).
  3. Offer a universal structured language for theorizing, addressing the limitations of traditional theories in HCI (such as Activity Theory or Behavior Change Theory) in adapting to diverse research contexts.
• Implementation Steps and Key Techniques:
  1. Construction Steps:
     • Define the causal research question.
     • List theoretical concepts and operationalize them into measurable variables (distinguishing between latent and observable variables).
     • Represent causal relationships between variables using arrows, iteratively expanding and validating the causal network.
     • Document hypotheses and provide documentation to support the model.
  2. Key Techniques:
     • Use statistical tools (e.g., dagitty in R or DoWhy in Python) for causal graph analysis and visualization.
     • Integrate multidisciplinary knowledge (e.g., psychology, nutrition science) to ensure the models are both theoretical and actionable.

Research Outcomes

• Specific Outcomes:
  1. A comprehensive theorizing framework applicable across all stages of the HCI research lifecycle (e.g., literature review, study design, operationalization, intervention design).
  2. Demonstrations of how DAGs can clarify causal assumptions, identify optimal intervention points, and uncover relationships between variables in statistical models.
• Advantages Over Existing Solutions:
  • Enhance theoretical transparency by elucidating assumptions through causal graphs, enabling future researchers to critically evaluate them.
  • Provide flexible and intuitive modeling tools adaptable to both quantitative and qualitative analyses.
  • Support exploration of multivariable causal relationships, addressing issues like confounding and selection bias often overlooked in traditional models.
• Experimental and Evaluation Results:
  The authors demonstrated the application of DAGs in real-world cases, including modeling data generation mechanisms, designing intervention points, and evaluating the validity of statistical models. For instance, they illustrated causal pathways of different smartphone usage habits affecting text input performance, revealing how confounding variables might lead to inaccurate interpretations and proposing methods to control for these variables.
• Limitations and Future Directions:
  1. Limitations:
     • Causal models rely heavily on assumptions derived from external knowledge, and the model structure cannot be directly validated using data alone.
     • Models may oversimplify and overlook complex system interactions.
     • Researchers need a high level of scientific reasoning skills for effective causal modeling and theoretical assumption formulation.
  2. Future Directions:
     • Explore integration of causal modeling with other theoretical frameworks (e.g., Critical Theory, Aesthetic Theory).
     • Develop standardized guidelines for constructing and evaluating causal models within the HCI community to encourage consistent exploration.
     • Further integrate causal inference methods into machine learning and complex system modeling to accommodate larger-scale empirical research needs.

In summary, the causal modeling framework proposed by the authors provides a practical tool for systematic theorizing in HCI, while fostering interdisciplinary theoretical integration within the field.