The paper addresses the need for statistical models to accurately reflect an analyst’s domain knowledge. To this end, the authors conduct an exploratory study using a domain-specific language (DSL) to express conceptual models. The study reveals a preference among analysts for detailing variable relationships and a desire to resolve ambiguity in their models. Based on these findings, the authors develop rTisane, a DSL that allows analysts to express their conceptual models and engage more accurately with their assumptions. The use of rTisane results in statistical models that better fit the data and maintain the analyst’s intent.
Publication date: 25 Oct 2023
Project Page: https://doi.org/10.1145/nnnnnnn.nnnnnnn
Paper: https://arxiv.org/pdf/2310.16262
