In the U.S. Consumer Expenditure (CE) Interview Survey, variance estimates are obtained by the balanced repeated replication (BRR) method. In the traditional sampling literature, justification for this approach uses the (approximate) independence of sample selection across strata and PSUs, and focuses only on the sampling error component of survey error. However, in the CE Interview Survey, we often need to have variance estimators that account for both sampling and measurement error. Consequently, one must consider modification of standard replicate-based variance estimators that will account appropriately for the correlation across strata and primary sample units (PSUs) induced by interviewer level measurement error. This paper considers simple variance estimators based on a collapsed-stratum approach. The collapse procedure is intended to ensure the newly paired pseudo-PSUs do not share a common interviewer, but have similar population characteristics. Specific matching algorithms are developed and applied to data from the CE Interview Survey. These algorithms use stratum and primary-unit level variables like population size and interviewer characteristics.