6

In my textbook, Identifiablity is defined as so:

For any $\theta_1, \theta_2 \in \Theta$ , if $\theta_1 \neq \theta_2 \Rightarrow \Bbb P_{\theta_1} \neq \Bbb P_{\theta_2}$ , where $\Bbb P_{\theta}$ is the probability distribution function.

Then it says that $X \sim N(\frac{\theta_1}{\theta_2}, \theta_3)$ where $\theta_1 \in \Bbb R$ , $\theta_2, \theta_3 \gt 0$ does not define an identifiable statistical model. Why?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
  • 4
    Consider, for example, $(\theta_1,\theta_2,\theta_3)=(0,1,1)$ and $(\theta_1,\theta_2,\theta_3)=(0,2,1)$. Both triples correspond to the same $\text{N}(0,1)$ distribution. – Zen Jan 11 '20 at 16:58
  • How am I supposed to use the definition to prove this example? Since in the definition there are only 2 $\theta$'s. This is my main concern –  Jan 11 '20 at 17:04
  • Do we suppose that $\theta_1' = (\theta_1, \theta_2, \theta_3)$ and then if $\theta_1' = (0,1,1) \neq \theta_1''=(0,2,1) $ does not imply inequality of $\Bbb P_{\theta_1'}$ and $\Bbb P_{\theta_1"}$ ? Hence model is not identifiable? –  Jan 11 '20 at 17:15
  • 1
    Yes, also related is this question on [identification](https://stats.stackexchange.com/questions/20608/what-is-model-identifiability?noredirect=1&lq=1) – Jesper for President Jan 11 '20 at 17:59
  • Please add the [tag:self-study] tag and read its wiki. – kjetil b halvorsen Jan 12 '20 at 14:43

0 Answers0