Experiments and quasi-experiments
29th November 2016 | Professor Nick Lee
For the past couple of columns, I’ve been discussing various issues around how to think scientifically and do experiments. In this column, what I want to do is bring a close to this series by presenting a couple of basic designs that could be used to generate knowledge of what works and what doesn’t in sales forces, including a “not-quite-experimental” design, which can be of use if true experiments are not possible for whatever reason.
A key principle of experiments, and the focus of my last column, was randomization. This is the process by which we select membership of the experimental group entirely at random, to avoid any other influences, or as we call them, confounds. Full details are in my last column, but I mention it here both because it leads into the next subject, which is control, and also because I recently experienced a wonderful example of how randomization can be problematic in business experiments.
More specifically, I was working with a very large multi-site organization, which wished to test whether a new management tool would work to encourage some behaviour. We’d managed to convince them to do an experiment, which was great. However, they came back to us with the idea that Site 1 would have the new tool and thus be the experimental group, and Site 2 would not, acting as a control group.