Analysis on Causal-Effect Relationship in Effort Metrics Using Bayesian LiNGAM

In the effort estimation studies, we can obtain open datasets from the past research. Those datasets are either within-company or cross-company dataset. On effort estimation, it was long discussed which dataset is appropriate for building accurate model. To find a new viewpoint in this discussion, we introduce the causal-effect relationship estimation technique. We use a simple Bayesian approach that is defined by the data generation model in a Linear Non-Gaussian Acyclic Model (LiNGAM). This model is applied to the function point and effort metrics in both within-company and cross-company datasets. We assume that if a dataset is appropriate for effort estimation, causal-effect relationships between metrics and effort will be extracted more. The result of case study shows that we can extract more causal-effect relationships from the cross-company dataset than that of from the within-company dataset.