I'm writing my thesis in psychological research. On one variable I did a tercile split because in other empirical work it is often done so and literature says that most likely there will be effects on the extremer parts of the distribution. In analyses then, I will use a dummy-coded variable.

In which section I should report the tercile split? Directly in methods section when I describe the measure itself? Or little bit later in analysis plan where I mention the statistical methods (e.g. regression) used?

Edit: I have to mention that the scale was a 10 point scale but in sample only values between 3 to 7 were observed, so this was also an argument for categorizing the variable.

  • Have you discussed this with your advisor? Commented May 12 at 19:52
  • Unfortunately, the adviser isn't available. Do you ask because splitting continuous vars may be an issue for itself? I have to mention that the scale was a 10 point scale but in sample only values between 3 to 7 were observed.
    – Madamadam
    Commented May 12 at 20:02
  • 2
    I’m voting to close this question because it's probably better asked on stats.stackexchange.com. Commented May 12 at 22:44
  • 1
    I agree that this may get better answers at CrossValidated, and have flagged and asked the moderators to migrate it. Commented May 13 at 7:08

2 Answers 2


Don't split your variable. This loses information for no reason. Instead, use a spline transform if you believe the response might be nonlinear. (If you believe you need to use splits because previous work did so, then consider running an analysis with a split and reporting it in an appendix. That other people ran bad analyses is not a reason to continue running bad analyses.)

In the "Methods" section, I would recommend you just explain the measure used, together with the relevant literature. Discuss any statistical transformations - whether splitting or spline transformation - in the "Statistical analysis" section.

This is at least what I would recommend. Alternatively, you could take a look at a couple of papers in the journal you plan on submitting to, and see how they dealt with similar transformations.


I agree with Stephen's answer and comment. I think you may have added your edit after his answer, which is OK, but what I'm writing about is related to that.

Just adding that you say your variable is an integer from 1 to 10 but you only have 3 through 7 in your data. Then, splitting really is odd. You are going from a variable with 5 levels to one with 3. Also, you say that literature says the effects are in the more extreme parts of the variable -- in that literature, was the same scale used? Were the more extreme parts 1 and 2 and 8, 9 and 10? Or was it a different variable?

Finally, that sounds like it could be a count variable, or perhaps a Likert-type variable; let us know which. And, is that variable the dependent variable or an independent variable?

Not the answer you're looking for? Browse other questions tagged .