Longitudinal Data Analysis for Ranked/Nominal Data

[SAS_Novice22]

Hello SAS Experts,

I am hoping to evaluate longitudinal changes in both nominal and/or ranked data. I have a survey that has been provided to customers where they can select three of their top observed behavioral improvements (can be nominal -all top 3 or ranked #1, #2, #3) from a list of 45-items at each follow-up. We are collecting this information from each participant (~300) at every 3 month follow-up and there are at least 6 follow-ups planned (hence longitudinal data).

I am wondering if anyone has any opinions or ideas on some standard analyses/methods that exist that could be used to analyze this data and if so any references that I could look at further?

I appreciate any guidance or help you could provide.

Thank you in advance.

[ballardw]

What does the analysis plan say about how the analysis was to be conducted after the survey was completed?

There should be a plan which specifies the intended analysis before collection starts. That way the survey can be designed to support the analysis needed.

The next might be what are you looking for in the analysis? No change of responses? Change to specific responses?

[SAS_Novice22]

Hello @ballardw 

Thank you for your response. All are excellent questions.

We have not yet developed an analysis plan as we are just discussing how we could in fact use this data once it is captured. Initially the intent was to just summarize the most common reported behavioral improvements using descriptive statistics.

We are not planning to use the survey in the way it has been developed. Where the 45 items (across 8 domains) are rated 0-2 where 0 indicates behavior is ‘not true’, 1 ‘somewhat or sometimes true’ and 2 ‘very true’. Global scores range from 0 to 100. 

The PI instead would like customers to indicate the top 3 behavioral improvements they have observed since the last follow-up by selecting from the 45-item RSBQ. I think we would ideally like to identify patterns overtime in these behavioral improvements. 

Because these are nominal or possibly ranked data I am wondering if there is an analysis approach we could employ for these purposes.

[Casamaretam]

I'm not a SAS expert, but I have some experience with longitudinal data analysis. One common approach for analyzing longitudinal data with ranked or nominal outcomes is to use generalized estimating equations (GEEs). GEEs allow you to model the longitudinal changes in your outcomes while accounting for the within-subject correlation over time. Another option could be to use mixed-effects models or random-effects models, which also account for the correlation between repeated measurements. Also, if you're interested in learning more about data analysis in general, there are some great Power BI courses available online. They can help you learn things way faster than you would by yourself.