1
$\begingroup$

To draw the lines below, generalized linear mixed model regression is used in order to model the effect of time (represented on the x-axis) on percentage of TDR (represented on the y-axis). I would expect to see a linear increase over time, as it is linear regression. Why is it not the case?

enter image description here

Improve this question
$\endgroup$
1
  • 2
    $\begingroup$ They have probably included a quadratic component, i.e. something like $\hat y=\beta_0 + \beta_1 x + \beta_2 x^2$ or perhaps some other non-linear transformation of the independent variable. $\endgroup$
    – matteo
    Commented Jan 22, 2018 at 10:53

1 Answer 1

Reset to default
1
$\begingroup$

A generalized linear [mixed] model models a transformation of the expected value with a linear [mixed] model. The transformation is determined by a (potentially non-linear) link function, $g(x)$: $\mathbb{E}(Y)\sim g^{-1}(X\beta\, [+Zu])$.

Thus the inferred relationship is not necessarily linear.

See this question for an extended Explanation.

Furthermore, a linear model needs to be linear only in the coefficients, $\beta$, not the the independent variable. E.g. $y_i=\beta_0+\beta_1x_i+\beta_2x_i^2+e_i$ is linear in $(\beta_0, \beta_1, \beta_2)^T$.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thanks. So no need to add a quadratic component as Matteo Lisi suggested? $\endgroup$
    – Anthony
    Commented Jan 22, 2018 at 11:03
  • $\begingroup$ I noticed only now that the dependent variable is a probability, so they probably have used a non-linear link function, as indicated in @YAK answer (+1). In this case, you'd expect a non-linear relationship: the outcome is bounded in $(0,1)$ (0%- 100%) while the linear predictor (year) is not. $\endgroup$
    – matteo
    Commented Jan 22, 2018 at 11:44

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.