Marco Russo and Alberto Ferrari ask who you believe, them or your lying eyes:
When looking at a report, it is natural to double-check the numbers produced. The simplest and most intuitive way is to verify whether the total equals the sum of individual rows. This behavior is extremely natural and mostly effective. Nonetheless, the total is the sum of rows only for additive measures, which are measures that are naturally computed as a sum.
When working with business intelligence solutions, sooner or later a developer will author a calculation that is non-additive. At that point, the total can no longer be computed by summing the rows for a very good reason: it would be inaccurate. When users complain about the fact that the rows do not sum up, seasoned BI developers offer a rational explanation of the reasons why the number are not summed: this process often provides a better understanding of how values are computed. Choosing the easy way out of introducing additivity in a naturally non-additive calculation means losing the opportunity to generate accurate calculations, and relying on inaccurate values.
Read on for examples and how to understand how to deal with non-additive or semi-additive features.