add fixed effect in proc nlin

[Nieves]

Hi SAS experts,

how to add the firm and year fixed effect in the PROC NLIN statement for the piecewise regression ? my current code does not have any fixed effect in the regression Xpart = a1 + b1*X; and Xpart = a1 + c*(b1-b2) + b2*X;. Thanks !

PROC NLIN DATA=local.plot2 MAXITER=1000 METHOD=MARQUARDT;
PARMS a1=1 b1=-2 c=3 b2=4 ;
Xpart = a1 + b1*X;
IF (X > c) THEN DO;
Xpart = a1 + c*(b1-b2) + b2*X;
end;
MODEL Y = Xpart;
OUTPUT OUT=PIECEFIT R=RESID P=PRED;
RUN;
* -- FIT THE POWER MODEL (FOR COMPARISON) USING NONLINEAR PROCEDURE -- *;
PROC NLIN DATA=local.plot2 MAXITER=1000 METHOD=MARQUARDT;
PARMS a=1 b1=2 b2=3;
MODEL Y = a1 + b1*X**b2;
OUTPUT OUT=POWERFIT R=PWR_RESID P=PWR_PRED;
RUN;
* -- COMBINE OUTPUT FROM BOTH MODELS -- *;
DATA ALL;
SET PIECEFIT POWERFIT;
RUN;
* -- PLOT DATA, PIECEWISE REGRESSION FIT, AND POWER MODEL FIT -- *;
SYMBOL1 f=marker v=U i=none c=blue;
SYMBOL2 v=none i=join line=1 w=3 c=blue;
SYMBOL3 v=none i=join line=2 w=3 c=black;
AXIS2 label = (a=90 r=0);
PROC GPLOT DATA=ALL;
PLOT Y*X=1 PRED*X=2 PWR_PRED*X=3 / OVERLAY FRAME VAXIS=AXIS2;
RUN;

[Nieves]

Hi SAS experts,

how to add the fixed effect in the PROC NLIN statement for the piecewise regression ? my current code does not have any fixed effect in the regression Xpart = a1 + b1*X; and Xpart = a1 + c*(b1-b2) + b2*X;. Thanks !

PROC NLIN DATA=local.plot2 MAXITER=1000 METHOD=MARQUARDT;
PARMS a1=1 b1=-2 c=3 b2=4 ;
Xpart = a1 + b1*X;
IF (X > c) THEN DO;
Xpart = a1 + c*(b1-b2) + b2*X;
end;
MODEL Y = Xpart;
OUTPUT OUT=PIECEFIT R=RESID P=PRED;
RUN;
* -- FIT THE POWER MODEL (FOR COMPARISON) USING NONLINEAR PROCEDURE -- *;
PROC NLIN DATA=local.plot2 MAXITER=1000 METHOD=MARQUARDT;
PARMS a=1 b1=2 b2=3;
MODEL Y = a1 + b1*X**b2;
OUTPUT OUT=POWERFIT R=PWR_RESID P=PWR_PRED;
RUN;
* -- COMBINE OUTPUT FROM BOTH MODELS -- *;
DATA ALL;
SET PIECEFIT POWERFIT;
RUN;
* -- PLOT DATA, PIECEWISE REGRESSION FIT, AND POWER MODEL FIT -- *;
SYMBOL1 f=marker v=U i=none c=blue;
SYMBOL2 v=none i=join line=1 w=3 c=blue;
SYMBOL3 v=none i=join line=2 w=3 c=black;
AXIS2 label = (a=90 r=0);
PROC GPLOT DATA=ALL;
PLOT Y*X=1 PRED*X=2 PWR_PRED*X=3 / OVERLAY FRAME VAXIS=AXIS2;
RUN;

[sbxkoenk]

It's not clear to me what you mean with "adding a fixed effect"?

• Is your code giving errors (then provide us with the LOG)?
• Is your code not giving errors but not doing what you want either, then explain further!
• Do you have other effects that you want to account for in Y (besides X) and you do not know how to integrate them in the formulas?

[EDIT] Also better to move this topic thread to the "Statistical Procedures" board!

Analytics (header) > Statistical procedures.

Can someone with superpowers do so?

Thanks,

Koen

[Nieves]

Hi Koen,

Thank you very much for your reply. My code is not giving errors but not doing what I want. I would like to add firm and year fixed effect in the regression. Currently, I only have explanatory variable x in the regression. I'm not sure how to include firm and year fixed effect in the proc nlin process. 

[sbxkoenk]

Hello @Nieves ,

I think you just have to add FIRM and YEAR to Xpart = (both Xpart-s!).

But it seems you are doing econometrics and having panel data (time series cross-sectional) data.

For panel data (and for having correct parameter estimation methods for non-independent observations and clusters in the data), there is PROC MIXED (or PROC GLIMMIX), PROC PANEL, PROC TSCSREG and PROC CPANEL (SAS VIYA Econometrics).

But these 4 procedures do not allow for the segmented regression (broken-stick regression) you are aiming at.

Segmented regression (Segmented Models) is also known as 'broken-stick regression' or 'Change Point Models'.

You can also use PROC MCMC (Markov chain Monte Carlo) for 'Change Point Models'. See the doc for PROC MCMC.

I hope you can sort it out.

Kind regards,

Koen