Hallo all you lovely people

I've started getting the data in for my first experiment. It is an online survey, where multiple versions of the same survey were sent to different groups. The survey has 40 questions.

My question is: would it be useful to first know if there is an overall correlation between answers given, and group who answered? In the sense of throwing all the data (nominal) into SPSS without separation by question. My second question (well done for still reading) is, if I do find a positive relationship between ALL questions and group, but the correlations for individual questions are really weak or non-existent for most, is it still worth reporting?

Much ta.

Drew 8-)

Hi, I've got a few more questions for you before I can think about this.....

Does the questionnaire provide you with numerical outcomes for each question?
Are they all asking about roughly the same thing so it makes sense to combine them into one score?
How many different groups did the survey go to?
Can I clarify, did each group get a different version of the survey? if so how different was it? you may get differences in answers from different groups based on the fact the survry was different and not on the fact the groups were different if that makes sense.

But yeah I think you could compare total score accross groups, provided you don't have lots of groups, in which case it would be difficult to see where relationships were.

For example I have a questionnaire measuring quality of life which I gave to two patient groups and one control group. I can sum the answers to all questions to get an overall measure of quality of life and then use a one-way ANOVA to look for group differences (if the survey provides ordinal data you'd need to use Kruskal-Wallis test, and if nominal chi-squared test).

Hope this helps. been sat by at the computer all day so brain a bit frazzled by now!

Hallo - thanks for your advice so far!

The responses are all binary - 0s and 1s. They ask identical questions, but some were presented in a different format to others, and I'm looking to see if the format had an effect on the answers. I'd also like to make sure that the format is what causing differences, and not say sex/ethnicity of participant. How would I account for that?

I was stats frazzled yesterday, so took a day off today and had a walk along the beach. much betteR!

Ok,

If the questions are binary are they all yes/no or positive/negative? if so you could take as the outcome, percentage of questions the participant answered yes. If they all ask about completely different things it's hard to see how you could get an outcome measure without seeing the questionnaire.

For the second part you could do use a mixed ANOVA with questionnaire type and sex or ethnicity as factors.