Circumventing multiple testing: A multilocus Monte Carlo approach to testing for association