Using the simple method of correlation will normally allow us to find some pretty interesting results. One of the most useful things we can do as researchers is point out new and exciting interactions effects using simple statistical methods. What is an interaction effect?
We hear that term a lot in medicine. Interactions occur when an effect is NOT present, but suddenly IS present when conditions are met. With medicine, a side effect of taking Pill A with Pill B could be a feeling of dizziness, even though each pill by itself cannot cause that to occur. However, when both medicines are in your system together, the effect is made obvious.
Here’s my fictional example study. We took two sets of people and placed them into two groups. Each group could choose to practice how to play Chess with a coach for any number of hours they wanted. Later that month, the individuals would come back and take a test about asking them, “which of the following chess moves would be the best,” for a 100 question test. Group one was a group of people who had never played chess. Group two was a group of people who were on a college level chess team. Sounds like the college players will do well right? You’d be right.
What we find is that the more hours one spends in the training program, the fewer errors they make – at least for the novice group (r = -0.66). As we discussed last time, a negative correlation means as one score increases another decreases, thus more hours must mean less errors!
In our next graph, we find the training has no influence on the master player group ( r = -0.03, pretty much no relationship). Perhaps people who are already good at chess do not benifit from such rudimentary training! As we can see, most of them made very few errors in the first place.
This means we have an interaction effect. If a person is a novice, the training has an effect, if the person is a master, the training has no effect. So we would say there is an experience level by training hour interaction effect.
A special note for interaction effects – When we see them in graphs, we notice that the slopes will eventually intersect. If my graph went on further, it would clearly intersect around 30 errors or so. If you see intersection, you see interaction!
Now wait a moment… if we think about my example more, we could critique it like this, “If someone is a master of chess, clearly they have had more general training hours than a novice. This study is not measuring what you are intending to measure, as mediation is occurring.” This is a huge problem, if my study weren’t fictional! What does that critique mean? It means that another factor (overall lifetime chess training) is the real explanatory factor that my model is not correctly controlling. When another factor is effecting both of your variables and has the real explanatory power behind it, we call it a mediator.
In my next post, I will talk about mediators in depth, and give you some ideas about what they look like in psychology.