I'm currently a fourth year university student. As part of my studies, I'm taking a class called Capstone, where students design and carry out a research project. An essential part of formulating this research is choosing a statistical procedure with which to analyze and present your results.

My study focuses on studying the increase of middle school students' awareness on the subject of bullying.

To do that, we will have a group of students who will take an initial questionnaire which has multiple choice questions about different situations and what type of bullying they represent. After that test, the same students will be giving a workshop where we will discuss bullying: the types that exist, how to recognize them and the negative impact they can have. After those workshops, the students will take another test, consisting of exactly the same questions as the first.

The goal is that, by comparing the answers on both tests, we will find that students answers on that second test correspond to a better identification and understanding of what bullying is.

My question the is: what type of statistical test would you recommend I use to sort and analyze the data I recollect?

  • 1
    $\begingroup$ What does the response consist of? Are you aggregating the answers into some single score, or a few scores, retaining answers along many dimenstions, or something else? Do you have other covariates to adjust for? $\endgroup$
    – Glen_b
    Nov 5, 2013 at 22:11
  • $\begingroup$ It may be that such a question might be better answered by estimation (point estimates, confidence intervals) than hypothesis tests. $\endgroup$
    – Glen_b
    Nov 5, 2013 at 22:14
  • $\begingroup$ Hi, @Melissa N. !!! I recommend you to create point scale for every possible answer (for each question in questionnaire). For example: Question no.1: five possible answers - (best answer 5 points .... worse answer 1 point). It is important (!) to gain information according which you will be able to "pair" tests of every single student (John Smith before workshop + John Smith after workshop). Create some ID's if questionnaires are anonymous. After data collection use paired t-test (or its non-parametric equivalent) in exploration for possible workshop effect. Good luck! $\endgroup$ Nov 5, 2013 at 23:06

1 Answer 1


I think you should do a simple pairwise difference comparison (before and after workshop) for each question separately.

Since you will probably use some Likert scale in your questionnaire (such as "Strongly agree", "Agree", etc.) your data will be ordinal.

You can use the Wilcoxon signed rank test, to estimate whether there was a significant change in the responses for each question after the workshop.

I think any serious statistical package will support it. I'm sure you will have for example SPSS at school.

If you want to be able to claim that it was the workshop that caused the change, I would recommend you to go a step further. Let the class fill out the questionnaire, then split the class in two parts randomly, and send only half of the class to the workshop. Then let the entire class repeat the questionnaire. The part of the class not taking your workshop will be your control group. You can check whether there will be significant difference even without the workshop.

(if the workshop offers some real additional value for the students, send the other half to the workshop after they have finished filling out the questionnaire a second time)


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.