1
$\begingroup$

I was hoping somebody could help with determining the correct statistical test to use.

Basically a teaching session was done which assessed confidence and perceptions of handling a situation using 3 questions. Each question had a Likert Scale (1-5). The same 3 questions on the pre and post questionnaire.

The problem is that not everybody filled out the pre/post questionnaire so the numbers are different and potentially more in one group than the other.

Is there any way to analyse this sort of data

$\endgroup$
4
  • 1
    $\begingroup$ Search on this site and elsewwhere for "missing data". There are many possibilities. After browsing around, you can give us a better sense of the exact problem. $\endgroup$
    – Peter Flom
    Commented Apr 14, 2013 at 21:54
  • $\begingroup$ It does not fully cover your problem but this question might be relevant: stats.stackexchange.com/questions/203/… $\endgroup$
    – Gala
    Commented May 15, 2013 at 12:26
  • $\begingroup$ Is each question intended to measure something different or all three meant to reflect the same construct? $\endgroup$
    – Gala
    Commented May 15, 2013 at 12:31
  • $\begingroup$ With a sample size of 30, how many are missing? Are single items missing or whole questionnaires? What exactly do you mean by 'the data are not linked'? These appear to be before-after data so they are dependent and not $iid$ - this might be the more important problem then some data missing hopefully at random. $\endgroup$
    – Bernhard
    Commented Aug 21, 2022 at 20:43

1 Answer 1

1
$\begingroup$

Classical test theory has a large problem with missing data. Modern test theory like Rasch or item response theory performs well with missing data and could be an excellent source of information about your data. There are some Rasch programs available for free if your sample size is small.

$\endgroup$
2
  • $\begingroup$ Thank you for your answers. I will have a read and get back to you. Whilst having missing data is an issue - the questionnaires aren't linked, so I have no way of knowing who answered which questionnaire. I also wondered which type of statistical test would work for the type of data. The sample size is around 30 people. $\endgroup$
    – T Yan
    Commented Apr 15, 2013 at 20:35
  • 1
    $\begingroup$ That's a small sample size and few questions. With only three questions, you may wish to keep the analysis very simple and only report the pre- and post-differences (with confidence intervals) for the three questions. Whether you report the mean or median will depend on the distribution. I wouldn't recommend getting too fancy. $\endgroup$ Commented Apr 17, 2013 at 20:07

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.