2

I have a glm() model with a couple of predictors. Two of such predictors are ordered factors with three levels.

One of these two predictors have significant Linear and Quadratic coefficients, but in opposite directions. Assume, for now, that pred_x is the number of HP of a car, and the response variable is speed.

pred_x.L      -0.18224    0.06229   -2.926  0.00344 **   
pred_x.Q       0.20684    0.03658    5.655 1.56e-08 ***

Intuitively, I expected a positive effect only. I'm not sure how to interpret results that are both positive and negative (both significant). Could somebody give me a hand?

double-beep
  • 3,889
  • 12
  • 24
  • 35
Guilherme D. Garcia
  • 149
  • 1
  • 11
  • Hello, I'm having the same problem. Did you end up knowing anything that could help? – Algorithmatic Dec 27 '16 at 21:24
  • Not 100% sure, but at least now I think it's not odd that a Linear and a Quadratic effect, both significant, point towards different directions. Imagine an upward parabola superimposed on a downward line. In other words, there's an effect for the linear trend (which is going down), but there's also a quadratic effect in the shape of an upward parabola. It'd be nice to have a statistician help out here =) – Guilherme D. Garcia Jan 17 '17 at 22:20

0 Answers0