From explainability to trust

Abstract

This study investigates the effects of governmental algorithm registers on citizens’ trust in the Netherlands. Algorithms are increasingly used by governments to support decision-making, but their opaque nature can negatively affect public trust. While transparency is often promoted as a solution, the literature presents mixed evidence: some studies suggest transparency improves trust, whereas others indicate it may reduce it or even create confusion. This research addresses the knowledge gap concerning how governments can best explain their algorithmic decisions to citizens and how the design of algorithm registers influences trust.

The research adopts an exploratory, empirical approach, combining quantitative and qualitative methods. Conjoint analysis is employed to examine how different attributes of algorithm registers influence citizens’ trust. Attributes are identified from grey literature and grouped into three categories: intention (legal basis, impact, proportionality), operation (human interference, risks, detailed description), and technology (methods and models, source data, source code). A survey with 131 respondents was conducted via Qualtrics, including nine conjoint questions, a holdout question, and demographic inquiries. The sample shows an overrepresentation of men and highly educated respondents, and an underrepresentation of older age groups, which is considered when interpreting the results.

Regression analysis of the survey data reveals that not all attributes positively affect trust. The attribute “risks” has a negative coefficient, suggesting that highlighting risks may reduce citizens’ trust. Other attributes, such as legal basis, methods and models, and source data, show smaller coefficients with relatively high p-values, indicating limited influence. Overall, the explained variance of the model is low, implying that trust depends on factors beyond the register attributes, such as respondents’ general trust in government or the expertise of the algorithm controllers. Trust in the central government has a stronger effect than trust in local authorities. The results also suggest that understandability plays a crucial role: registers that are unclear or overly complex can confuse citizens and reduce trust.

To complement the quantitative findings, a focus group of digital transition consultants was consulted. Experts emphasized that registers should use clear, simple language, visualizations, and uniform formatting. They also highlighted the importance of involving citizens in the iterative design process to ensure that information is comprehensible and relevant. These insights align with the literature warning against information overload and misinterpretation.

The study concludes that governmental algorithm registers can both positively and negatively affect citizens’ trust. Transparent registers do not automatically result in higher trust; the content, clarity, and contextual relevance are critical. Moreover, registers are only one component of building trust—general trust in government and effective communication also play significant roles. Policymakers should carefully design registers, balance transparency with comprehensibility, and engage citizens in dialogue about algorithmic decision-making. Future research should include larger, more representative samples, explore additional attributes and algorithm types, and consider alternative analytical methods beyond linear regression to capture the full complexity of trust dynamics.

In summary, the study highlights that an appropriate design of governmental algorithm registers is essential for supporting trust, but transparency alone is insufficient. Effective registers must be clear, relevant, and citizen-centered, while governments must also foster broader trust and communication to ensure algorithmic decisions are understood, accepted, and perceived as legitimate.