Timeline for Linear regression vs Fixed Effects model for calculation of slope of clinical parameter vs time across multiple subjects
Current License: CC BY-SA 4.0
4 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 25, 2022 at 15:34 | comment | added | EdM | @psm with a mixed model you force the slopes (and intercepts; both at baseline covariate values) to have a best-fitting Gaussian distribution among subjects. The modeling puts more weight on subjects with more observations, minimizing noise arising from individuals with few observations. If you model effects of clinical covariates via interactions with time at the fixed-effect level, the individual subject slopes are usually of only secondary interest; they account efficiently for within-subject correlations. Do consider other approaches like generalized least squares, however. | |
| Feb 25, 2022 at 14:57 | comment | added | psm | Thank you for that in-depth answer - you've provided a lot of clarity on the topic. Could a downside of using such a model be that it creates some interdependence between subjects, such that non-systematic and non-informative noise in some subjects could influence slope calculations in other? | |
| Feb 25, 2022 at 14:55 | vote | accept | psm | ||
| Feb 25, 2022 at 14:43 | history | answered | EdM | CC BY-SA 4.0 |