Statistical choices can affect inferences about treatment efficacy: A case study from obsessive–compulsive disorder research