one way would be to do wilcoxon (2 related samples) seven times for each question but adjust a not 0.05 but 0.05/7 for each test. Or you can do a multivariate t-test.
Hi!
I am a doc student and have a few questions about a pre/post intervention likert scale I am trying to analyze. It is a 9 point sclae (1=strongly disagree to 9=strongly agree). There are 7 questions. The participants completed the survey pre and post an intervention. Here are my questions:
1. What stat test is appropriate to compare pre and post? Would the Mann-Whitney be right?
2. Is it better to look at each question individually, or should i look at the overall total when comparing? For example, should I look at all the participants combined ratings for question 1 pre/post, then question 2 pre/post, etc. Or should I total the overall score (eg, question 1 plus question 2 plus 3...) for all the participants pre and post, and compare those two numbers?
Thanks in advance for any help or advice. I appreciate it!
one way would be to do wilcoxon (2 related samples) seven times for each question but adjust a not 0.05 but 0.05/7 for each test. Or you can do a multivariate t-test.
1) You should try test for dependent sample like.. The Wilcoxon Matched-Pairs Signed-Ranks Test or paired t test.
2) it is depending on the questions. you can always look at each question individually.
Or if you can come up with (meaningful) single weighted average, you can try testing on this variable. one advantage of this is , the weighted score will have more continuous nature..
In the long run, we're all dead.
Thanks for the help! If possible, could you provide some clarification on the following?
1. Should I average the pre and post score for each question (say pre the group average was 5 and then post it was 8.5) when I do the wilcoxon..or would it be better to use the overall total for a particiular question.
2. I only have 7 participants, would these tests still be appropriate?
3. How would I go about finding a meaningful single weighted average?
Sorry if the answers to these seem obvious...this is not my greatest area of strength.
THANK YOU!!
1) & 3)
It is depending on the questions.. suppose q1 and q2 are kind of opposite questions ( eg: .. q1. X performance q2.X not efficient ). In that case you can't take average.
One possible way is: you can club similar questions and come up with 2 or 3 derived variables ( you can come up with one score also, if your questions are featuring same attribute. And if you think all questions are equally important, then you can take simple average )
2) is it 7 questions and 7 participant..?
If yes, how is the average of 7 participant of each questions behave pre/post scenario. If the post performance is higher in all questions then you no need to do any statistical test.. you can conclude based on mathematical logic.
In the long run, we're all dead.
Tweet |