Analysis of summated scores

I’m working with data from a medical score system that creates a score for a patient by adding up scores from 11 Likert-scale questions (I’m not a fan of this, but it’s a scale commonly used in the field and it’s the data I have). We are looking at differences in score between two groups. My first instinct was to use a proportional odds model. Another approach which is perhaps more faithful to the true data generating process would involve separate ordinal regressions for each item and then obtaining a posterior distribution for the summated score in both groups. This all said, I have a feeling linear regression might recover approximately the same effect in about 2 minutes.

How are summated scores typically handled?

2 Likes

i’m handling these scores at the moment using CPM as per this Inevitably if there is a difference between groups, people will want to analyse the subscales to see what’s driving it. Regarding linear regression, the scale isn’t meaningful ie presenting marginal means will be difficult to interpret and to communicate the magnitude of the effect. I used CPM to plot the cumulative probs for treatment groups, but maybe im inclined to present marginal means in a forest plot with the composite and susbcales toegther

2 Likes