Benchmarking is used often in establishment surveys to adjust sample weights to match the current distribution of population of interest. In the National Compensation Survey, the weight of each establishment in the sample is adjusted to match the distribution of current employment by industry from the Quarterly Census of Employment and Wages program. The process involves calculating a benchmark factor for each cell and multiplying the establishment weight by the calculated factor. In cases where there are fewer than three responding sample establishments or the factor is larger than 4.00, two or more cells are collapsed. The question is which cells should be collapsed so that the effect on the mean square error is minimized. This paper presents the current collapse pattern and several other collapse patterns and evaluates their impact on mean square error of earnings estimates.