CMStatistics 2016: Start Registration
View Submission - CMStatistics
B1393
Title: Balancing scores in causal diagrams and causal estimates for different data contexts Authors:  Priyantha Wijayatunga - Umea University (Sweden) [presenting]
Abstract: Potential outcome model and graphical model are two major frameworks for causal inference tasks such as finding the effectiveness of a medicine for a given disease or that of a training program for unemployed people to find employments, etc. using past observed data. These two frameworks are related with each other and their logical equivalence is shown previously. We show how balancing scores found in the potential outcome model can be represented in the graphical models (causal diagrams). There have been some discussions on this so far, but we show that they are not quite correct. We discuss the diagrams and causal estimates for different contexts such as for matched data in cohort studies, case-control data, etc. These discussions that correct current literature show that one can define new causal estimates for matched case-control data. Note that in such contexts, due to data selection methods it is often not meaningful to use causal effect estimates used in, for example, cohort studies where different data selection methods are used. We also show how the estimates for matched data can be adjusted to obtain those for the population.