JohnM: Likert analysis help


I was hoping you could give me some guidance on 5 point Likert-type scale analysis. I've read all the threads about it but none fully address my situation. I'll try and give you the information that you've requested of others however if I miss anything please let me know - my future is riding on how well I do this!

I've collected 150 questionnaires which assessed both the attitudes towards first-cousin marriage and beliefs about the genetic risks (recent culturally-insensitive comments by a Labour MP prompted this work). The final two questions are the most important to my study. The eighth question asks whether we should stop cousin marriage and the ninth (final) question asks this once again whilst providing some statistics about the increased risk. The hypothesis being that once people are aware of the risks, their response will favour stopping consanguineous marriage more than the response in question 8. I would like to assess whether my results here are statistically significant in both the entire dataset and also two sub-groups within the dataset.

I've uploaded the questionnaire if you want a clearer picture.

While I'm assuming that there will be a 'direction' to the results, I haven't started writing this section of the study. As there is very little research into this area I'm thinking that instead of locking myself down into a one-tailed hypothesis I would be better treating it as an exploratory study and develop a hypothesis/prediction from the results to be tested in further research. What impact would this have on the statistical tests? Detrimental?

What test(s) do you believe would be the best to prove (or disprove) my data as statistically significant? Could you please briefly explain why these tests are more appropriate than others - someone suggested I use Fichers exact however I've also read t-test, Mann Whitney, Chi-squared and Dunnet's. In another thread you suggested using different tests depending on the mean/median or each question. It's all a bit confusing, the last time I did this (Mann Whitney) was back in A-Level psychology!

The other questions are less important, so I'm not sure what to do with them. Just report the percentage responses, the means/medians or perform statistical analysis on all? I'm aiming for top-marks, and am prepared to put the work in to obtain them.

JohnM said:
Another interesting approach with Likert scale analysis is to do a simple linear correlation between each question and a question which is designed to measure "overall" attitude. The questions that correlate highly with the "overall" question are what we call "drivers" - they have the most to do with the respondents' overall attitudes.
I'd like to investigate this further, but not just yet. I need to get the above done first.

Many thanks,
edit: Removed a question (I found the answer) - 03/04/2006
Last edited: