On the Representation and Aggregation of Evidence in Software Engineering: A Theory and Belief-based Perspective

An adequate representation and a feasible aggregation procedure of evidence represents a challenging problem in many disciplines. The right representation can help scientists discuss and present the results of their findings and, if it is simple enough, it can be useful for practitioners to base their decisions on improvement implementations. The aggregation strengthens confidence in comparison to single evidence and is an important contribution to the body of knowledge. In this paper, we present a preliminary proposal to use empirically-based theories and belief functions as a means to represent and aggregate evidence. By having evidence explained by the same theory, we used belief functions to combine them in a way that the theory propositions (cause-effect values) result from combined evidence. We suggest this can be an useful way to obtain a good estimate of multiple evidence combination. In addition, we indicate its possible usefulness for practitioners to formalize and reuse their experiences. A real-case application of the approach is presented by formulating a theory for Usage-Based Reading inspection technique and aggregating the evidence acquired in three related empirical studies. This application indicated that the approach can give compatible results with the aggregated evidence.