topic Re: matched Pairs vs. Longitudinal Analyis vs. ??M in Discussions

matched Pairs vs. Longitudinal Analyis vs. ??M

Clarknj — Mon, 09 Dec 2024 04:52:11 GMT

I have a dataset of 100 students that have undertaken a specialized training program and measure their proficiency (0 to 100). It is measured at a number of points in time in a number of different ways Test (A, B, C, D, E, F). For each student, there is the initial evaluation, 2 months later evaluation and the final evaluation 3 months later. I want to determine if there is significant improvement but am confused on what statistical tests to conduct. I want to compare month 3 to 1 as well as month 1, 2 and 3.

For example. I could use matched pairs and check month 3 vs. month 1 and see if there is a significant difference between the 100 students. I could also calculate month 3 - month 1 and see if that is significantly different than 0? I also have a few categorical factors I would like to see if they impact the proficiency gains.

Currently, the data is one row per student and columns for variables/factors (categories and efficiency test for initial, 2 months later and 3 month and this is repeated for each of the 6 tests (A thru F). Is there an easy way of completing the analysis without running the matched pairs individually? I’ve heard of longitudinal analysis and think that could help in seeing if the performance changes over time from month 1 to 2 to 3.

JMP PRO 18

Re: matched Pairs vs. Longitudinal Analyis vs. ??M

MRB3855 — Mon, 09 Dec 2024 11:41:40 GMT

Hi @Clarknj : The paired t-test is a special case of a mixed model. It is equivalent to a mixed model with student as a random effect and the factor of interest as a fixed effect. i.e., stack your 1, 2, and 3 month data into one column, and have month as a fixed effect (along with your other categorical factors, as needed) and student as a random effect in the mixed model platform. This kind of thing was discussed here:

https://community.jmp.com/t5/Discussions/Paired-t-test-on-long-format-table/m-p/740466#M92189 (worthwhile to read the entire thread).

The mixed model platform will allow you to make all comparisons of interest.

Re: matched Pairs vs. Longitudinal Analyis vs. ??M

Clarknj — Mon, 09 Dec 2024 16:03:26 GMT

Thanx for the link and explanation. Is this the same as Longitudinal Analysis where the focus is on effects over time? The fixed effect would be the time and then would get a response over that time period and be able to determine if segments of the time period are more significant than others?

Re: matched Pairs vs. Longitudinal Analyis vs. ??M

MRB3855 — Tue, 10 Dec 2024 02:06:31 GMT

Hi @Clarknj . You are now at the tip of a very large iceberg!
Short answer? Yes, you can carry out Longitudinal Analysis (repeated measures) this way. One of the key elements will be your choice of covariance structure. See the following to get you started.

https://community.jmp.com/t5/JMP-Knowledge-Base/Analyzing-Repeated-Measures-in-JMP-Software/ta-p/575343

You are entering advanced territory; the following may help.

https://support.sas.com/resources/papers/proceedings/proceedings/sugi30/198-30.pdf

https://faculty.washington.edu/heagerty/Courses/VA-longitudinal/private/Littell-StatMed2000.pdf

if you get stuck, do come back; but please be advised that the potential complexities and nuances of your problem may lead to more than an online community (such as ours) can provide for…