Semi-Additive Metrics

Semi-additive metrics are those whose values can be summarized for some dimensions of a model, but not all. Ratios such as Average are also considered semi-additive metrics.

For example, suppose you have a column that tracks employee headcount. The headcount yesterday was 10 employees, today it is 11, so does this make the total for the month 21? Definitely not! This is an example of a semi-additive metric. While headcounts of two regions or two departments can be summed up, the changes across time for a single region or department should not be summed up. So, headcount is a semi-additive metric across time. In this example, we want to look at the last non-empty value we have in a given month to determine our total headcount.

For this use case, you'd define the semi_additive property in the metric's SML, specifying the dimension over which the values should not be summarized (usually a date or time dimension). Instead, the query engine returns either the First Non-empty or Last Non-empty value in the result set when the results are sorted by time. For more information, see Add Semi-Additive Metrics.

Note: In Tableau, semi-additive metrics will appear as CNTD (Count Distinct) aggregations in the client application if the return value is an integer and AVG if the value is not an integer. However, the actual query will use whatever aggregate calculation you specified when the metric was defined in AtScale. This is because SQL clients do not understand the concept of semi-additive metrics, and this is how AtScale informs the client that the memetricasure is non-additive. Users should not change the aggregation function on the client-side in Tableau for these semi-additive metrics.