06-01-2018 11:42 AM
So I have these two examples of code to keep the example simple. And it doesn't really matter which data you use it on, the difference will be there anyhow.
proc reg data = A_A_Data(where=(region="0114")) ;
model inrinflytt = utflyttare / noint;
proc ucm data = A_A_Data(where=(region="0114"));
model inrinflytt = utflyttare ;
They both generate a very simple model without intercept. I however can not understand why the coefficient of determination differs between these two when the paramter estimates are exactly the same. Also the MSE differs between the two models. Which tells me that the UCM model is not simply y = bx.
Is it y = bx + e?
My guess is that it has something to do with the error/distrubance term but I can't figure out what it is. Is it that in the UCM it is estimated so the model is y = bx + e but in the proc reg, when estimating y, the model is y = bx. Then e is used for diagnostics, but not for predicting new y values.
Anyone who understand my problem and can help me out?
06-01-2018 12:12 PM
The formula is in the docs