We demonstrate the use of stochastic collocation to assess the performance of photonic devices under the effect of uncertainty. This approach combines high accuracy and efficiency in analyzing device variability with the ease of implementation of sampling-based methods. Its flexibility makes it suitable to be applied to a large range of photonic devices. We compare the stochastic collocation method with a Monte Carlo technique on a numerical analysis of the variability in silicon directional couplers. This demonstrates the high performance and efficiency in analyzing device variability with the ease of implementation of sampling-based methods. Its flexibility makes it suitable to be applied to a large range of photonic devices. We compare the stochastic collocation method with a Monte Carlo technique on a numerical analysis of the variability in silicon directional couplers.