Hi Everyone,
I want to analyse how well two scales agree for detecting changes in wellbeing. I look at the Bland-Altman Method. One scale is a simple score from 0 - 10 for wellbeing. The other scale is the well established Beck’s Depression Inventory-II (goes from 0 - 63) . I wonder how to analyse the agreement?
Thanks!
Fred
You could just run a correlation for these values. If your data follow an approximately normal distribution, then you can use Pearson's r. If not, then you can use Spearman's rho instead.
Hi Paul! Thanks for your thoughts! The Bland Altman recommends the use of a limits of agreement calculation because using correlation only says there may be a association, whereas agreement says how well two scales agree. Put simply, they compare the average of the two scales's values against the substraction of the valuas. And calculate a 95% limits of agreement interval. So to get back to my original question, how to take an average when the values from the two scales differ so much? Hopes this makes any sense ....
this is their analysis: Article Statistical-Methods For Assessing Agreement Between 2 Method...
What is your n? What other variables do you have iin your research?
Hello! We still have to calculate the required N. What do you mean with other variables? I have 1 VAS (0-10) question and a Beck's questionnaire. My first question is: how do I standardize the answers to be able to compare the two scales?
Dear Levente Kriston , nice to hear from you! Hope you are well. Thanks for your answer. I worked out a strategy such as you also suggested, by using the Z-scores. One thing we still need to figure out is how to calculate the sample size to conduct a meaningful limits of agreement analysis. What ideas do you have? Best, Frederik
Thanks for your stimulating thoughts here. I am also thinking in terms of population. Scales frequently are considered to be valid when they have a Crohnbach of .75 or more. Correct? I may be mixing things up here, but if that means that around 75% (or more) of observations are valid: Let's say we accept that 35% difference is acceptable between the two scales. In z-core terms, this population equals a CI of [ -0.5 to 0.5]. From here we can calculate the sampling error and so on... This is getting interesting :-)
Hello there. I would go for the Pearson r correlation between scores from both scales on the same population. Keep it simple as things would go very complex if you keep using the "what if" in your plan. Data could be collected in a single set, or consider administering the second scale a week later. Then you can run correlation between the two sets of data. Re z score, I'm still not confident it determines much at this stage. Hope this helps.
Cheers
You get a sum or squares from BDI for example and the you correlate the the scales with each others. A Cronbach's alpha >.70 is okay.
Hello,
This link talk about: Agreement Between Two Ratings with Different Ordinal Scale
Chapter Agreement Between Two Ratings with Different Ordinal Scales
- Badges
- Science topic
- Similar topics
- Psychology