Should I split my sample?

For my master thesis I have conducted a survey with Likert-scales. But now I'm sure which test to use to verify or falsify my hypotheses.

For example, I have a hypothesis that states that:

Citizens who have a negative perception towards European politics, think of lobbying of less democratically legitimate than citizens who have a neutral or positive attitude towards politics.

I measured, with Likert-scales, both these attitudes (the perception towards European politics and the perceived democratic legitimacy of lobbying) in a scale from one to five. Edit: I have 170 respondents for this survey.

My question is, can I just use linear regression analysis to find a correlation, or do I have to split the sample into the group that is negative (and thus score a 1 or 2) and neutral or positive (who score 2,3,4)? Any help would be greatly appreciated. Thank you in advance.


Active Member
If both variables are ordinal or continuous, you should not split the sample. You should run Kendall's tau and / or Spearman's rho on the whole sample. This will pick up any linear or non-linear monotonic relationship between the two variables.