The balanced g factor loading is the average of the object test's conservatively estimated minimum g factor loadings on a number of categories of tests. This approach reduces the "contamination" of g that may take place if one estimates a test's g loading based on an unbalanced set of other tests; for instance, a disproportionate number of verbal tests would contaminate g with verbal ability, etcetera.
The categories are primarily homogeneous tests with respectively only verbal, numerical, spatial, or logical problems, as well as a category "heterogeneous non-compound", containing tests with a mixture of item types. By leaving out compound tests (that is, tests combining smaller tests that also give scores in their own right), overlap is avoided. The resulting g loading is balanced over the various categories as it is independent of the number of tests per category, and thus gives the categories equal weight.