linear regression/PROC SURVEYREG goodness of fit

[Emma_at_SAS]

I want to check the goodness of fit for a PROC SURVEYREG regression model. I checked R-square for my model which is a small number but I do not know how valid that is to be reported. I appreciate your help. Thanks

[ballardw]

How small is "small"? Subject? Social science type data (most surveys) often reports R-square values much smaller than physical science.

Did you check for anything besides linear fit? Sometimes a "poor" R-square for linear models comes from data that fits some sort of curve.

More information that a simple R-square is likely needed to make a good suggestion.

[Emma_at_SAS]

Thank you @ballardw for your help.

My R-square values are below 0.35 and my survey is in public health. Good point about checking a curve. I will try it. Thanks

[sbxkoenk]

Hello @Emma_at_SAS ,

R-squared is definitely a popular and good measure, but it also has disadvantages.

R-Square is the proportion of variance in the dependent variable which can be predicted | explained from the independent variables. 
The main disadvantage of R-Squared is that it will always increase if you add an additional input to your model (even if that input is not significantly contributing to the power of the model, but is only explaining a bit of noise).
See here :
What does R square mean in variable selection?
https://communities.sas.com/t5/SAS-Data-Mining-and-Machine/What-does-R-square-mean-in-variable-selec...

I also like to look at the RMSE (which is an absolute measure) when predicting | explaining interval targets.

Root Mean Square Error (RMSE) :
Root MSE is the standard deviation of the error term, and is the square root of the Mean Square Residual (or Error).
RMSE can be interpreted as the standard deviation of the unexplained variance, and has the useful property of being in the same units as the response variable.
Lower values of RMSE indicate better fit.

By the way, you can often make better models with PROC REG and PROC SURVEYREG if you use the EFFECT statement  (to make spline effects for example).

Good luck,

Koen