Selection mechanisms favour reproduction of better individuals imposing a direction on the search process. According to this it is expected that the effective number of offspring of an individual in the next generation would always agree with the algorithmic sampling frequencies. This does not happens due to sampling errors. Stochastic universal sampling is a method that tries to remedy this problem.
This presentation discusses performance results on evolutionary algorithms optimizing a set of highly multimodal functions and a hard unimodal function, under Proportional selection and stochastic universal sampling. Contrasting results are shown.