1
$\begingroup$

I have put some data together and modelled the behaviour of the response ($y$) as function of three independent variables $x_1$, $x_2$ and $x_3$. A simple multi-linear regression. The model looks like:

$y = k + a*x_1 + b*x_2 + c*x_3 + e$

Up to this point everything is OK. But now I want to lower the response by a $15%$. Like reducing some commissions or costs. The only idea I came across is to multiply the responses by $0.85$ and readjust the whole model. Recalculate $a$, $b$ and $c$ with the new values. I have been trying to find another way of doing this without touching the data samples. Just changing and adjusting the coefficients $a$, $b$ and $c$. Does anybody know how this should be done? An idea you come across with would be okay.

$\endgroup$
0
$\begingroup$

The regression coefficients are given by:

$\hat{\beta} = (X'X)^{-1}X'Y$

Thus, if you scale $Y$ values by $0.85$ then this is effectively equivalent to the following:

$Y_{\text{new}} = 0.85e' Y $

where,

$e$ is a vector of ones.

Thus, the new estimate is given by

$\hat{\beta}_{\text{new}} = (X'X)^{-1}X' (0.85 e'Y)$

Thus, you get the following relationshiop:

$\hat{\beta}_{\text{new}} = 0.85 \hat{\beta}$

$\endgroup$
0
0
$\begingroup$

Just subtract 15 from all the y-values, then run the regression. You are using linear regression, and therefore applying an additive "translation" (i.e., subtraction) form the y-values before regression will simply result in coefficients that will predict values whose mean is less 15 units.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.