Decision support algorithms in critical domains such as healthcare can be immensely beneficial, but they also pose risks, such as the propagation of historical biases. Causal prediction has been proposed as a solution to prevent algorithmic biases, but it requires strong assumptions about the data that were used to develop the algorithm.
This project aims to improve the transparency and trustworthiness of decision support algorithms by empowering end-users to evaluate their applicability. While causal assumptions are currently only evaluated globally, the project will investigate a method to assess them individually. By introducing the concept of causal effective sample size, end-users can be informed on whether an algorithm sufficiently applies to their individual situation.
The project will develop a method to calculate and communicate causal effective sample size and evaluate its value for end-users. The ultimate goal is to enable citizens to become active critical users of decision support algorithms, making informed choices about whether to use them in their individual decision process.
