Things aren’t always what they seem: Misinterpretation of observational data in software engineering