The National Compensation Survey (NCS) currently uses Taylor series linearization to estimate variances of mean wages and total employment. Since Taylor series estimators depend on the form (ratio or total) of a parameter, they can become quite complex for certain other NCS parameters, such as pay relatives. Pay relatives are indexes relating locality to national pay: cells are weighted by national employment and hours. Consequently, future NCS variance estimation will be done using Fay's variation of balanced repeated replication (BRR). This paper presents BRR formulas and discusses how variance strata and PSUs are defined for different parameters. For means and totals, BRR is straightforward, except that locality and national estimates require different variance PSUs, variance strata, and Hadamard matrices. BRR is more complex for pay relatives, however, because they are functions of both locality and national parameters. Half-samples are defined using a hybrid Hadamard matrix that references both locality and national variance strata.