Why not just use POLYFIT?
It is more numerically stable than using the Vandermonde matrix directly.
I want to implement an a way of calculating an error vector using Error = (X - Dt), where Dt = [1, 1, 0, 0, 0, . . . , 0]T.
I don't really understand where Dt comes from, but why not implement it directly as you've written it?
Dt=[1 1, zeros(1,length(X)-2)].';