Hi,
Is that just for learning or a problem you met in your job? It looks like using fixed effect on longitudinal data (time). Unfortunately, when using ABSORB statement, you are not able to get the predicted values or residuals because the firm was conditioned out in the model, i.e., excluding from the design matrix. It should be similar (or equal) to regression on centered response and predictors on time. However, I do not know how to do it,
JingJu
notes:
一共有两种编程方法
1. absorb firm; class year;
2. class firm year;
I don’t use absorb. As my understanding, it is nothing but a quick computational technique. That is, you are supposed to have same coefficients for fixed effects and model information. However, since absorb factor is not playing a role on design matrix, many results are not given in the model, such as residuals. So the two models above have no difference.
I don’t know if the residual is additive. But I do believe type I ss is additive.