In FEA programs, stress is calculated in so called integration points. These points are inside each element. But after the analysis user wants to know the distribution of the stress over entire model. That’s why the values from integration points are extrapolated to nodes. Most nodes are common to two or more elements so they will receive several values from extrapolation.
With the default “Compute scalars before averaging” option Abaqus first calculates scalar values (like Mises stress) from stress tensor components. Then averaging is used to obtain single scalar value for each node. When this default type of averaging is selected user can specify the averaging threshold in %. This controls the extent of averaging between neighboring elements. How it works ? In case of 75%, if the relative difference between contributions that specific node gets from neighboring elements is greater than 75% , these contributing values are not averaged. If it’s less than 75%, values are averaged. In case of 0% there’s no averaging and in case of 100% all results are averaged.