The design versus the analysis of observational studies for causal effects: parallels with the design of randomized trials