I created a survey collecting perceptions of players of a mobile game for learning. There were 24 questions and each one consisted of a 7-point Likert scale. Part 1 (16 questions) was about their perceptions of the usefulness of various features of the game, and Part 2 (8 questions) was about their degree of satisfaction of various features of the game. For Part 1, questions ranged from 1(totally agree) to 7 (totally disagree). For Part 2, questions ranged from 1 (extremely dissatisfied) to 7 (extremely satisfied). There were 15 participants.
Among the 3 options below, I am wondering what is the best statistical test to discover correlations in the data:
Option 1: Spearman's rho, because Likert scales contain ordinal data
Option 2: Kendall's tau, because Likert scales contain ordinal data and it seems that Kendall's tau is more robust even though Spearman's rho is used more often?
Option 3: Pearson' r, because we can consider data from a 7-point Likert scale as ordinal data?
Sorry, for option 3, I meant to say 'interval data', not 'ordinal data'.
Hello Diane,
It wasn't clear to me whether you intend to evaluate relationships between scores on pairs of items/questions from your survey, or whether you first were going to generate one or more composite scores from the first part and one (or more) from the second part and determine the relationship of those instead.
In the latter case, any number of mechanisms (e.g., factor analysis, item response theory, structural equation models) could be used to create composites that were defensible as interval strength. In the former case, ordinal it is!
So, I'd try Spearman for item pairs; Pearson for (appropriately formed) composites.
By the way, a Spearman correlation is nothing but a Pearson correlation run on the ranked values of the scores (so, I guess the answer is always "Pearson"), and has the advantage of dealing with tied ranks in a suitable fashion (whereas how to do so in Spearman is sometimes a point of contention).
Good luck with your work.
Thank you David, I'll be evaluating relationships between scores, not finding composites, so I'll go with Spearman, then. Much appreciated!
Dear Diane Rozells ,
Your data set is really small so it is unlikely that you will obtain a significant result unless there is a stro relationship between two variables. You should be guided by your research questions in choosing which pairs to compare with the Spearman correlation. You really have far too many questions compared with your sample size. You also need to manage the risk of Tyoe I errors in your statistical testing if you are running several tests.
Article The Bayes Factor, a Suitable Complement beyond Values of p
- Badges
- Science topic