I have been doing plenty of research into which test I should use on my Likert scale data and i am becoming more confused as there are so many people saying different things.

I am currently doing my dissertation on To what extent does music affect aggression whilst playing violent and nonviolent video games? I have use three different methods to collect data from 30 participants:

  1. An aggression scale before and after the participants played both violent and non-violent video games to measure the changes in aggression whilst listening to different genres of music

  2. Story completion tasks which give the participant a scenario and asks for ten responses i.e i would be angry, I'd hit the driver etc after playing both violent and non-violent video games

  3. A questionnaire to see what peoples' opinions are on violence in video games using a likert scale ( 5 point scale strongly agree - strongly disagree) and also open ended questions on preferences to music etc.

I have already completed my analysis for the aggression scales (1) but i am completely stuck on which test i should choose when trying to gather data from my likert scales. the questions were such as:

  • Video games are becoming too violent
  • Violence in video games are influencing people to become more aggressive
  • Video games are unsociable
  • Video games teach and encourage people's aggressive behaviour
  • Video games help people to forget about everyday life worries
  • Video games encourage people to be more aggressive in real life situations

which they then score on a scale of Strongly Agree, Agree, Neither Agree nor Disagree, Disagree, and Strongly Disagree for all statements (13 all together).

I want to know whether people believe games cause real life violence or whether they think games are too violent. Thanks in advance :)


For all practical purposes, you can use the usual statistical measures to convey the message of your statistical measures. Linear regression is a great tool to assess relationships between Likert outcomes, and so are proportional odds models with ordinal predictors. Using robust standard errors will protect you from having to make any parametric assumptions about the shape and distribution of the residuals in the model and the validity of the standard errors.

The benefit over rank-based statistics, such as Spearman correlation, is that you actually obtain an averaged difference in Likert responses comparing individuals differing by 1 unit in the "independent" response. (e.g. what is the expected difference in proportions of individuals who would elect to hit somebody comparing individuals who found that metal was "excellent" versus "good").

  • $\begingroup$ Thank you AdamO! your a star :) Just hope i get this right now! $\endgroup$ – vickie adshead Jul 14 '14 at 18:39

Also, see the assumption of equal intervals for this controversial point.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.