Estimating causal effects for multivalued treatments: a comparison of approaches