03-24-2014 05:37 PM
I have a two-part question:
What is the easiest way to calculate a partial f-test for two or more variables from a model estimated via proc autoreg? My current idea was to just estimate the model with all variables and then with the variables in question removed, and then divide the marginal mean square model by the full model mean square to get the value for the f-test of the variables in question, and then use sas function "probf(x,df1,df2)" to get the p-value to determine significance of variables in question. - If there is a better/easier way, please let me know (I couldn't find an option in autoreg to do this for me, but perhaps I overlooked something?)
- If this is the way to go about this then I have a further question:
To go about this method, I need the Model Mean Square, or the Model Sum of Squares & Model Degrees of Freedom from autoreg, but it seems that it only output the Mean Square Error / Error Degrees of Freedom. Is there any way that I can get Model Mean Square from autoreg?