So for my research I asked people their preferred path adjustment alternative. They could choose from 4 pre made alternatives (A= no adjustment, B=small adjustment, C= some what bigger adjustment, D=biggest adjustment). They do have an order but no values are given to them. This question has been asked for multiple scenario's for example if they are in a rush vs when they are not. 3 parameters defined these scenarios and were tested in each possible combination. So a total of 8 questions.
Each respondent answered each question and the data is thus dependent.
My supervisor advised me to use the Wilcoxon signed rank test. From data that was found online I took the amount of respondes that answer A in situation X and compared it with the amount in situation Y. I did this for A,B,C and D. The problem with this is that i get a really high p-value, since I only have 4 values to compare. Is there a way to address a value to the answer options so i can compare them for every person and thus increase the amount of compared data.
To answer the question what information do I need from my calculation: I want to see whether we have a statistical difference when we change a parameter from A to B so i would like to compare 2 list of answers (out of 8) with each other and see if the difference is significant.
It seems to me you could consider the preferred path adjustment alternatives as responses on a Likert scale from 0 to 3. If so, maybe you could run a one-way repeated measures ANOVA, with these Likert type answers as dependent variable and the scenario's as independent variable, with 8 levels.
Hello Vitali,
The answer depends on what specific questions you'd like answered from your data set.
If it's a matter of whether "Factor X" (your "parameter", which sounds like a dichotomous variable in your query) makes a difference in options chosen, then:
1. For a given factor, sum the scores (using 1-4 as rank values for options A-D, or 0-3, as suggested by Jos Feys ) across the four items that represent one level of the factor and do likewise across the four items that represent the other level of the factor.
2. Run a Wilcoxon signed-ranks sum test on the paired scores and determine whether that factor does or doesn't explain differences in participant choices.
Repeat the process for each factor. Note that this approach overlooks possible interactions. So, if you have different questions about the data, you'd need to look at an alternate method.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis