Let's say you were interested in performing a meta-analysis on the predictive validity of various simple linear regression models which differed in the method of extrapolation. One of the main metrics used to evaluate a "model's error" is the SEE
Would it, in theory, be possible to conduct a meta-analysis using the SEE of various models (or the SEE scaled to the mean)? To clarify I think this would involve comparing studies using different approaches to predict an individual's score on an outcome measure and then providing a pooled estimate, for each model type, of the prediction error. I have recently seen this being discussed, but I am unsure how it would work in practice as I am unfamiliar with the process. Would the SEE be used or would they have to perform analyses on the raw prediction residuals with, for example, a hierarchical model with fixed effects for factors thought to moderate the magnitude of the error? Hopefully that makes sense. I only have experience with MA's that use measures such as SMD's or OR's and so have no idea how feasible this would be.