Understanding output of PROC GAM

[Krishnan]

Dear friends,

I am using Proc GAM on a panel dataset (with 3,00,000 obs) for exploratory data analysis. I wanted to understand what is effect of an independent variable on the dependent variable. After running the Proc GAM code, I got a graphical output with the independent variable on x-axis and the effect on y-axis. How do I interpret this output?
Code used:
ODS HTML;
ODS GRAPHICS ON;

PROC GAM DATA = TMP1.DATASET PLOTS(CLM);
MODEL Y = SPLINE (X);
OUTPUT OUT=ESTIMATE P RESIDUAL UCLM LCLM;
RUN;

ODS GRAPHICS OFF;
ODS HTML CLOSE;

U can e-mail me,if more information is required. This is an urgent request.

Thanks,
Krishnan (krishnan.s@mu-sigma.com)
Business Analyst
Mu Sigma Inc., Bangalore

[statsplank]

Hi Krishnan,

Did you get any p-values in your output?

[Krishnan]

Yes..I got p-values and all that stats. The only issue is to understand the graphical output of PROC GAM. What the graphical output conveys.

[statsplank]

I suspect your graphical output looks similar to those in

Example 36.1 Generalized Additive Model with Binary Data

There, I saw a paragraph that begins with "The plots show that the partial predictions corresponding to both Age and StartVert have a quadratic pattern, while NumVert has a more complicated but weaker pattern." You may consider describing the pattern you see in your plot in similar terms.

[deleted_user]

Thank you for the timely response!

I have understood the interpretation of Proc GAM graphical output.

1) Generalized Additive Model (GAM) : Y = B0 + S(X) ; (This says the relationship between Y and X is unknown and S(X) is a function with unknown relationship)

2) GAM separates out the linear and non-linear trend of the predictor variable to Y = B0 + B1(X) + F(X); (where B1(X) is the linear part and F(X) is the non-linear part)

3) Estimates are computed separately for the linear part and the non-linear part

The predictions from non-linear part (called as partial predictions) are plotted against the predictor variable in the figure. Thus the figure depicts the non-linearity part of the relationship between Y and X (if at all there is any). If there is no non-linearity no curve would be generated.

Krishnan