1

I'm having some problems understanding when multicollinearity is acceptable and when it becomes a problem.

Can someone give me some insight on what should I focus on?

Here's an example of what I'm struggling with.enter image description here

kaba100
  • 11

2 Answers2

4

This is more of a math than a statistics exercise, but here is a huge hint:

$$ \log(y^2) = 2 \log(y) . $$

dsaxton
  • 12,138
  • 1
  • 26
  • 48
1

Multicollinearity is a problem when it is perfect as in the question. This means that there are supernumerary parameters that add nothing to the regression. More precisely, they are worse than nothing as they make regression unstable. Consider for example $y=a+b$, since $a+b=k$, there are no unique solutions for $a$ or $b$ just for $k$, and the equation $y=a+b$ has no unique parameter values. Now when multicollinearity is high but not perfect, what one generally does is search for why that is and search for other $f(x_i)$ to approximate $y_i$.

Carl
  • 13,084