[Limdep Nlogit List] F-test question

[William Spitz]

I am doing an F-test for equality of coefficients in a simple linear model.

Base model is y = a + b X
I am testing whether the observations should be split into Groups 1 and 2:
y = a1 + b1 X + a2 + b2 X

But when I run the restricted model (using CLS: b(1)=b(3), b(2)=b(4)), I 
don't get the same restricted coefficient values as I do when running 
the Base model.
They're pretty close, but not the same -- and it seems to be off by more 
than just rounding error.

Is this a "feature" of imposing restrictions via CLS?

Thanks in advance.
/Bill Spitz