Keep it Simple!


Posted January 27, 2017

In the sciences and in statistics, people may sometimes believe that more complicated theories, models, or analyses are better, or somehow more correct. This is not always the case! In fact, more complicated theories, models, or analyses can be more trouble than they are worth, necessitating more pages of writing, more research, a larger sample size, and more complex results that may befuddle you and your readers more than necessary. In short, overly complicating your research can make an already arduous process even longer and more frustrating.

Let’s consider statistical analyses as an example. Most of us know the simple t-test, also known as the Student’s t-test. It is one of the most basic significance tests you can conduct. As such, some eschew it as elementary. However, this is not the case! It is based on mathematics and theory just as much as any other statistical test. It was, in fact, not developed for the express purpose of student’s statistical homework—it was developed by William Sealy Gosset under the pseudonym of “Student”, hence the name. If your study calls for just one comparison (such as between a dichotomous independent variable and a continuous dependent variable), make just that one comparison using a t-test!

On the other hand, sometimes your study may call for the comparison of more than two groups. In this case, it would not be best practice to use multiple t-tests, as this will inflate your risk of Type I error (i.e., finding a significant difference when no difference actually exists in the population, or a “false positive”). Here, keeping it simple would involve the use of just one ANOVA (if your variables are appropriate), rather than multiple t-tests. Similarly, rather than using multiple ANOVAs for multiple dependent variables, consider using one MANOVA.

When it comes to regressions, you might be tempted to include as many predictor variables or covariates as possible. You want to know all you can, of course! However, with every predictor or covariate you add to your model, more and more participants will be required in order to achieve appropriate statistical power.

Finally, keep in mind the big picture of your study and try not to bog down your narrative with too many superfluous details. And, of course, avoid using anything in your dissertation if you are not confident in your mastery of it. You will be expected to know everything about your dissertation inside and out. So, do not get in over your head with an overly complex model or an advanced analysis that you will not be able to explain. When in doubt, keep it simple!


Pin It on Pinterest

Shares
Share This