The variability and non-dispatchability of wind power creates many challenges for the operators of electric transmission systems. Current U.S. wind energy policies are focused on encouraging quantities of wind power without much attention paid to quality of the power produced. Using detailed meteorological data from 113 different weather stations in Oklahoma, we simulate power production from a large number of interconnected wind farms and devise a variance-minimizing rule for successively adding farms over a wide geographic area. Our variance-minimizing rule reduces the standard deviation of five-minute averaged wind power output decreases by 27% after grouping of 8 stations. We compare our variance-minimizing decision rule with two other decision rules for incremental wind investment: a nearest-neighbor rule that has been suggested in previous literature and a profit-maximization rule that reflects decentralized decision-makers. All interconnection decision rules reduce the aggregate variance of wind power output, particularly after several stations are interconnected. We find that the nearest-neighbor rule reduces variance by less than half that of the varianceminimization rule. The profit-maximization rule achieves 75% of the variance reduction attained through the variance- minimization rule. We also evaluate and compare wind power variability over hourly and daily time scales and analyze the sensitivity of variance-minimizing wind energy investment patterns to wind-speed measurement frequency. This work is a first step in a larger project, in which we plan to compare the intermittency costs of wind power that arise from different siting policies or decision rules.