Many practitioner researchers strive to understand which assessment practices have the best impact on learning, but in authentic educational settings, it can be difficult to determine whether one intervention, for example the introduction of an online quiz to a course studied by diverse students, is responsible for the observed effect. This paper uses examples to highlight some of the difficulties inherent in assessment research and suggests some ways to overcome them. Problems observed in the literature include: assuming that if two effects are correlated then one must have caused the other; confounding variables obscuring the true relationships; experimental approaches that are too far removed from reality; and the danger that self-reported behaviour and opinion is sometimes different from student's actual behaviour. Practical solutions include: the use of an experimental or pseudo-experimental approach; the use of mixed methods; and the use of meta-analysis.