This study extends the framework of the Rubin causality model, one of the most important and applied models in statistical science in recent years, to develop a unified model for the heterogeneity of causal effects in hierarchical cluster data and individual-level effects where the intervention effect differs depending on some factors even for the same individual. We developed an efficient estimation method that avoids the bias introduced by existing methods. We also clarified that the presence of heterogeneity leads to the problem of inconsistent results across multiple RCTs, and developed a method to correct for selection bias that takes heterogeneity into account while partially utilizing population information. Simultaneously with the development of these methodologies, we conducted applied research in marketing, medicine, and education to demonstrate the validity of the proposed framework and methodologies.