2
$\begingroup$

I am about to turn in a paper on research methods and need some guidance on how to analyze my data statistically.

The paper is about a hypnotized study where participants fill in a Likert type attitude questionnaire concerning certain types of behavior (cooperation, communication, leadership etc.).

The participants will then solve a team task whilst being observed and scored by an observer on those same types of behavior.

I wish to come out of this with a correlation between attitudes and behavior, but I am not sure how to do this. Can I plug it straight into a Pearson's r? or must I use Spearman rank order? What if the results does not allow for a parametric test?

Also any recommendations on how to analyze inter-rater validity would be great! would Cronbach's alpha be good?

Please be gentle in your suggestions, as I am a total novice when it comes to statistics:-)

Thanks!

Thanks for replies! Good karma to you! For simplicity I will presume there are no outliers. The Likert scale will be 1-5 for both attitudes and behavioral scores. There will be several attitudes questions and behavioral categories. Does that mean the data is on interval level? If so does that restrict me to Spearman's rank order?

$\endgroup$
4
  • $\begingroup$ Welcome to the site! Is this for a class project? If so, you should add the self-study tag, which will let other people know that they should structure their answers differently than for other questions. $\endgroup$ Commented May 2, 2013 at 18:19
  • $\begingroup$ How to correlate depends on the nature of the variables that you are correlating. Will these scales be interval level or ordinal level? Are there outliers? Also, you might consider regression models as well as correlation. $\endgroup$
    – Peter Flom
    Commented May 2, 2013 at 18:37
  • $\begingroup$ When you talk of “Likert type attitude questionnaire”, do you mean one question with a 1-to-5 or 1-to-7 response format for each behavior or scales composed of several such items? $\endgroup$
    – Gala
    Commented May 2, 2013 at 19:29
  • 1
    $\begingroup$ If you have added several Likert items into a single scale, you have already assumed an interval scale. The only remaining question is whether you particularly seek linear correlation or more general monotonic association $\endgroup$
    – Glen_b
    Commented Jun 2, 2013 at 8:43

1 Answer 1

1
$\begingroup$

What is a hypnotized study?

Sure can you assume that Likert-Scale items are interval-scaled. It may be a strong assumption, especially when respondents are heterogeneous and everyone has different intervals on that scale. It is a much softer assumption to treat it as ordinal; however, that might be beyond the scope of your homework. It would require either nonparametrics or threshold models.

Why don't you start with simple Pearson's r? If you don't see anything with it, using more advanced methods won't help anyway.

In a second step, work on the assumptions. Do you need to treat data as ordinal? Is there response bias, i.e. is there correlation between how people using a Likert scale in general (regardless of the content) and specific behavior? Some people do not like ambiguity; thus they just put their ratings on the extremes, although it is not their "true" answer. The same people may behave in a specific way. Thus you'd have inflated correlations.

Again, first things first.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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