I have developed a questionnaire. I want to try and analyse question drop-off. There is a clear negative correlation between question number and percentage of respondents who answered that question. No question has a higher response rate then the proceeding question. I want to see if this negative correlation is significant so I assume I can do a simple Pearson etc?
However, I would also like to see if there is any significant elbow point throughout the questions, where there is a significant difference in the % of responses between a question and the following question. Ie qu10 has 60% response rate and qu11 has a 50% response rate. Is that significant compared to the drop-offs between other questions? Any ideas who test? I am sure I read that there is a chi Square test for trend? And, if so, can I also look at negative correlation in the same test so I don't have to do both?
It sounds like correlation will work fine for your first question. It sounds like it's not really a linear response, so perhaps Spearman correlation would be more appropriate. Although you may want to use something like linear regression to determine the slope.
Considering the break-point question, there are a few approaches that may work, depending on what the situation is.
A potential option is Pettitt’s test for change. I have a note at the bottom of the page here: https://rcompanion.org/handbook/F_13.html .
Another approach might be segmented regression, potentially a linear-plateau model. (Or in your case, a plateau-linear model). About a third of way down the page I have a section on linear-plateau and related models here: https://rcompanion.org/handbook/I_11.html . There's also an example of Cate-Nelson model, which might also work for you.
In any case, start with a plot of response rate vs. question. That will inform what analysis makes sense.
- Badges
- Science topic
- Similar topics
- Journal Impact Factor