linear regression vs multilevel model

Hi people.
I'm currently trying to conduct a simple study to compare
- the conventional way of measuring proptosis on CT using the interzygomatic line
- a new method using a new landmark
Obviously the measurements would be different. My study seeks to the assess if the new method will correlate with the old method.
I've been in discussion with a statistician who suggested a a multilevel model because my study involves paired eyes from each patient.
However, after scouring the website, I see papers using linear regression models for similar studies.
From my understanding, multilevel models assesses how clustered data affects the outcome.
Is this actually necessary if I'm only needing to assess the correlation between two sets of measurement - despite it being paired eyes?


Not a robit
An assumption of linear regression is independence between observations. So having two eyes contributed by a single person would brake the assumption. Think of it like this, if you let me vote twice in an election my two votes would be correlated. Further more I would contribute multiple obs for voter characteristics (e.g., two males, two white persons, etc.), it would up-weight my characteristics as being associated with the outcome. Ophthalmology is notorious for requiring multilevel models to control for data observation dependence.


Fortran must die
It commonly depends on if your clusters or groups actually matter I believe. If they explain little variance I suspect that the linear and multilevel results will be very similar even if formally the linear regression violates the assumption of the method in the case you describe. Formally count data also violates linear regression assumptions and you should run Poisson regression. But in fact it commonly makes no difference.