Hi,
I have a model to predict cost using the GenMOD procedures. The distribution used is Inverse Gaussian with a log link. The predicted values in the output dataset and the graph produced by effectplot make sense. The predicted costs are in the order of tens of thousands of dollars. The coefficient parameters instead are decimals. For example how do i interpret that the coefficient estimate of a covariate such as diabetes is 0.1? How can I convert it into dollars? I would really appreciate your feedback. Thank you
This is a case where the SOLUTION vector will be useful. The categorical variable will change only the baseline, unless you have fit the continuous by categorical interaction. In the latter case, there will be a separate slope for each level of the categorical variable. To get the estimates as slope coefficients rather than as deviations, where you will have to do some additive preprocessing to get the slope estimate, use the NOINT option in the MODEL statement.
So to get what your are requesting, do two things: Include the continuous by categorical interaction in the model and include the NOINT option in the MODEL statement. You can then exponentiate the estimates and the confidence bounds to get what you want.
SteveDemja,
Let's suppose that the coefficient is 0.1. Then for every increase of 1 for diabetes (and I don't know what units apply here), you should expect an increase of exp(0.1) =1.105 in the cost. Does that make sense?
SteveDenham
Thank you! So I need to exponentiate the coefficients first to understand how the dependent variable cos increase, because of my log link. Correct?
Is it possible then to construct a Table with $ amounts instead of coefficient parameters? For example if I have a continuous predictor (mg/dl of Glucose) interacting with a categorical predictor (diabetes), and I am interested in the cost increase per mg increase of glucose at different levels of the categorical variable (no diabetes, with diabetes treated with insulin, etc. ), with confidence intervals, can I obtain this from proc genmod?
This is a case where the SOLUTION vector will be useful. The categorical variable will change only the baseline, unless you have fit the continuous by categorical interaction. In the latter case, there will be a separate slope for each level of the categorical variable. To get the estimates as slope coefficients rather than as deviations, where you will have to do some additive preprocessing to get the slope estimate, use the NOINT option in the MODEL statement.
So to get what your are requesting, do two things: Include the continuous by categorical interaction in the model and include the NOINT option in the MODEL statement. You can then exponentiate the estimates and the confidence bounds to get what you want.
SteveDemja,
Hi Steve,
I have not had yet the possibility to work on my dataset to test your suggestion but if I manage to apply it without mistakes it would be exactly what I need.
Thank you!
Registration is now open for SAS Innovate 2025 , our biggest and most exciting global event of the year! Join us in Orlando, FL, May 6-9.
Sign up by Dec. 31 to get the 2024 rate of just $495.
Register now!
ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.
Find more tutorials on the SAS Users YouTube channel.