Polynomial regression: export estimated value and error

[hhchenfx]

Hi Everyone,

I have a timeseries and would like to run a polynomial regression of value on time and save the estimated value and error to a new file.

Can you please help?

Thank you,

HHC

/*THIS ONE WORK*/
proc datasets lib=work kill noprint;run;DM "log; clear; ";ods html close;ods html;

data test;
infile datalines missover;
input month Orders;
yearMonth = mod(month,12);
actual= not missing(orders);
datalines;

1 26156
2 32396
3 27415
4 24129
5 30350
6 21992
7 24705
8 36434
9 28858
10 37353
11 43137
12 52119
13 42119
14 42119
15 62119
16 72119
17 42119
18 62119
;

proc sgplot data=test;
reg x =month  y=orders /degree=10 clm;
title 'sgplot';

run;


proc glm data=test;
model orders = 
month 
month*month  
month*month*month  
month*month*month*month 
month*month*month*month*month
month*month*month*month*month*month
month*month*month*month*month*month*month
month*month*month*month*month*month*month*month
month*month*month*month*month*month*month*month*month
month*month*month*month*month*month*month*month*month*month
/ PREDICTED E;
output out=glm p=pred r=residual;
;
title 'GML regression';
run;

[PaigeMiller · Accepted solution]

Example: https://documentation.sas.com/doc/en/statug/15.2/statug_reg_gettingstarted02.htm

[hhchenfx]

Thank you,

HHC

[PaigeMiller]

@hhchenfx says

/*THIS ONE WORK*/ proc glm data=test; model orders = month month*month month*month*month month*month*month*month month*month*month*month*month month*month*month*month*month*month month*month*month*month*month*month*month month*month*month*month*month*month*month*month month*month*month*month*month*month*month*month*month month*month*month*month*month*month*month*month*month*month / PREDICTED E;

It may work, meaning that the code doesn't have errors and produced output, but a 10th order polynomial fit to this data (or any other data that I know of) is completely inappropriate in the real world. You have most likely fit the noise of the data rather than finding a real signal, meaning that you have not really found a useful answer (you have found code that doesn't cause errors, that's all).

[WarrenKuhfeld]

As @PaigeMiller says, don't do that! If you want something more flexible than a cubic polynomial, use a cubic B-spline. Many of the linear modeling procedures such as transreg and glmselect can easily do it.