Suppose that $M_1$ is the residual maker for a unity vector (i.e. a vector made of $n$ 1's).

I am told that this matrix, when premultiplying a variable, transforms the variable "into deviations from its sample mean". What does this mean?


1 Answer 1


Let's consider the context of linear regression, where the least square estimator in matrix form is:

$$\hat{\mathbf{\beta}}= \begin{bmatrix} \hat{\beta_{0}} \\ \hat{\beta_{1}} \\ \vdots \\ \hat{\beta_{m}} \end{bmatrix} = (X'X)^{-1}X'\mathbf{y}$$

Let's call $P=X(X'X)^{-1}X'$ the prediction maker matrix.

For the residuals you simply do $\mathbf{y}-\hat{\mathbf{y}}$, which is equal to $(I-P)\mathbf{y}$. Hence, we call $M=(I-P)$ the residual maker matrix.

In your specific example, instead of a predictor matrix, you have a matrix that computes the mean of a variable, let's call it $A$. In this case, for a vector variable $Z$ you'd have $(I-A)Z=Z-\overline{Z}$, which is indeed $Z$ expressed as deviations from its mean value.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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