Hi there,
I have constructed a degree-of-worry scale on which callers to a medical helpline, are asked to rate their worry for the situation they are caling in about, where 1=minimal worry and 5= maximum worry. I analyse the data as an ordinal data. However, for a joint analysis.
Now, I have collected data on 12.000 individuals - it seems likepople have the ability to rate on a 5 point scale (people are able to discriminate between five levels), however for the sake of power and easily interpretable results I would like to collapse into a 3-point scale. I can see that is frequently done in self-rated health research - but what are the good arguments for doing so?
Reducing to 3-point will reduce the extent you are able to analyse the findings. In many respects you are doing a kind of data reduction at the first stage.
Before administration, sure.
After, you can, but you need to recalculate the Stand Error un order to account for non-symmetrical shifts in the distribution.
Hope this helps,
Matt
In addition to the previous comments, keep in mind that you are losing information when collapsing categories.
And with that many respondents, I am not sure that power is an issue.
Why would you want to reduce the information you have at hand from 5-points to 3 points? Loss of information.
Apart from the above comments, I would like to suggest to go through the manuscript titled:
"Quantifying the relationship between latent variables after successful implementation of QM practices in MSMEs"
Where authors have used a scale from -2 to +2. Where - represented "not important and +2 as "critically important".
It useless to collapse categories. Loss of information does not make your interpretation more robust. And you will tend to collapse under one category (for example 5+4=3 and 1+2 =1) unequal cohorts of answers. So there a no good arguments for collapsing data !
I am agree with above comments, but for me if you have completeness with 3 points is enough. Please check completeness.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence