I hope someone can suggest a reference or recommend a methodological approach to address this problem. I want to fit a model with several predictors that describes the relationship between the response Y and time in years (the model form has not yet been determined). I am interested in understanding how each of the predictors contributes to the fitted model. The challenge is that data at the "observation level" are not available on any of the other predictors; only summary data (means, variances N's, etc.) are available for each year. So I can fit a "primary model" using data on the individual observations, that describes the relationship between the response Y and time. I can also independently fit several "secondary models" (one for each predictor) using the mean response for each year, based on that predictor. Is it possible to 'adjust' the trend as described by the "primary model" by the trend(s) described by the "secondary models", given that the models were fit independently? Is there a reference that might detail how one would do this? Must the primary and secondary models have the same form (all polynomial models, for example) to do this?
-------------------------------------------
James Gear
Senior Statistician
Midwest Employers Casualty Company
-------------------------------------------