Posted on October 19, 2007
Ignoring interactions and not centering are two common mistakes in data analysis, states Helena Kraemer (bio).
I’ll give you one other example of I was thinking in terms of common error. I could keep on going on common errors for a long time. This has to do with regression analyses and something – it took me a long time to figure out. And that is trying to predict an outcome measure for multiple independent variables. Number one is people tend not to put interactions into the model and the argument I always thought was because there was no reason to believe there were interactions. Only belatedly I found out that in many cases the reasons for not putting them in is because they know exist but they are not interested in them. Now the problem with those interactions is that if they exist in the population and they’re not in the model, they don’t disappear just because you decided not to put them in the model.
And so those missing interactions get remapped to bias the whole analysis and possibly to compromise the power. So it’s one of those few situations where it compromises both type one and type two error. So, at that point I started really sort of pressing on getting those interactions in only to find that a number of people didn’t realize that when you have interactions in the model, how you code the data in, change all the conclusions. This is called centering. I mean you can find it in the literature but I could swear that 90% of the researchers I know who are using regression models do not know they have to center the data.
Excerpted from interview with researcher at the 2006 Career Development Institute for Psychiatry in Pittsburgh, PA.
Please note that the feedback is viewed only by 4researchers staff and is not intended for communication with individual contributors.
Use the form below to submit feedback about this article. If you would like a response, please be sure to include your e-mail address.