I am working on research where I have 66 participants who perform an exercise every day, and then they need to respond to 14 questions with a Likert scale (1-10). They performed this exercise for 21 days and every day, submit the answers. One of the questions was how relaxed did you feel? The response is from 1-10. 10 is the highest level of relaxation. I also have independent variables like gender, age, marital status. I have been working with R studio data, but I am not sure how to analyze it. I took biostatistics classes, but I am not sure how to analyze it in the best way. Does anybody can recommend to me what would be the best approach. My question/hypothesis is: After 21 days of practice, did these participants feel more/less/equal, for example, relaxed? If yes, no, is there any gender difference, marital status, age? What regression model should I use?
Hello Karolina,
If you are using individual question/item responses as outcome variables, then you'd likely want to consider a method such as ordinal regression. Since your research question implies that the principal comparison of interest is baseline to end of study, you would only need to use those two observations (among the 21 that you have).
A big part of the answer to your query hinges on what scores convince you that there was a change in the outcome of interest. For example, does the difference between a score of "4" and a score of "3" a genuine decline in perceived relaxation, or is that degree of shift tantamount to an ignorable difference? Remember, there's measurement error associated with these ratings, in addition to extraneous mood confounds that will pop up from day to day. As well, do you want to predict: (a) the actual score at end of study; or (b) the change status (positive, none, negative)? Ordinal regression could handle either (but the interpretation might be simpler with change status).
If you decide that you're interested in the trajectory of self-reported relaxation (or whichever outcome) scores over the duration of the study, you might consider growth curve models. The Mplus software package can easily handle such designs, even with ordinal measures. I'm willing to wager that there's at least one R library as well that could be applied.
As a note for future endeavors, individual item responses aren't as reliable as responses aggregated across a set of related items. In fact, that was the method that Rensis Likert recommended (summated scores across related items) in his monograph from which we all came to know of the Likert-type response scale.
Good luck with your work.
You can use modes, frequency, means, or Chi squared for categorical non-parametric data
I fully agree with David Morse. Maybe it would be interesting to first look at the histograms of the distributions of the responses to see if or to what extend these look 'normal'. Maybe after you ran the item analysis suggested by David, you could also calculate the correlation between the 21 days (1,2,3 .. 21) and the means or medians of the responses. This should be significant if the relation is linear. It might also be interesting to draw a line graph of these means (or medians) as in R script in:
https://r-graphics.org/index.html
for the Figure 7.16
If the distributions are obviously not normal (e.g. very skewed) you could rely on the LD.F1 test from the nonparametric nparLD R package (see attachment).
Thank you for all your thorough answers. Can you recommend a book, a youtube channel, a website where I can study more about the ordinal regression model, Likert scale, and all different approaches that I should use for my data?
David Morse, unfortunately, I do not have the baseline, I only have the answers after day 1 ( including day 1 -but they already performed the practice), I would like to see how the candidates changed their status and if there is some correlation in gender/age/and other independent variables. I also would like to know which the dependant variable changed their status positively/negatively. "As a note for future endeavors, individual item responses aren't as reliable as responses aggregated across a set of related items." - related items, do you refer to independent variables? What kind of software do I need to use and how to prepare the data in excel? Also, if this is a Likert scale - can I do mean, SD, sum?
You could first convert the categorical data into continuous interval-level data using Rasch analysis. Rasch analysis does this with log-transformation. Then calculate average values / plot the person-measures obtained from Rasch analysis etc.
- Badges
- Science topic