if ${yi}^n{i=1}$ is iid find the fisher information of the distribution $f(y; \theta) =…
- User
if is iid find the fisher information of the distribution in terms of the reparametrisation
- Assistant
To find the Fisher information of the distribution ( f(y; \theta) = \theta y^{-\theta - 1} ) with respect to the reparameterization ( \phi = \frac{1}{\theta} ), we'll follow these steps:
-
Understand the Distribution:
The given probability density function (PDF) is: [ f(y; \theta) = \theta y^{-\theta - 1}, \quad y > 1 ] This is similar to a Pareto distribution with scale parameter ( x_m = 1 ) and shape parameter ( \alpha = \theta ).
-
Compute the Fisher Information for ( \theta ):
Fisher information for a single observation is given by: [ I(\theta) = \mathbb{E}\left[ \left( \frac{\partial}{\partial \theta} \log f(Y; \theta) \right)^2 \right] ]
First, compute the log-likelihood: [ \log f(y; \theta) = \log \theta - (\theta + 1) \log y ] Then, take the derivative with respect to ( \theta ): [ \frac{\partial}{\partial \theta} \log f(y; \theta) = \frac{1}{\theta} - \log y ] Squaring and taking the expectation: [ I(\theta) = \mathbb{E}\left[ \left( \frac{1}{\theta} - \log Y \right)^2 \right] ]
For the Pareto distribution: [ \mathbb{E}[\log Y] = \frac{1}{\theta}, \quad \mathbb{E}[(\log Y)^2] = \frac{2}{\theta^2} ] Substituting these: [ I(\theta) = \frac{1}{\theta^2} - 2 \cdot \frac{1}{\theta} \cdot \frac{1}{\theta} + \frac{2}{\theta^2} = \frac{1}{\theta^2} ] For ( n ) independent observations, the Fisher information scales linearly: [ I_n(\theta) = \frac{n}{\theta^2} ]
-
Reparameterize to ( \phi = \frac{1}{\theta} ):
To find the Fisher information in terms of ( \phi ), use the transformation property of Fisher information: [ I(\phi) = I(\theta) \left( \frac{d\theta}{d\phi} \right)^2 ] Since ( \phi = \frac{1}{\theta} ), we have: [ \theta = \frac{1}{\phi} \implies \frac{d\theta}{d\phi} = -\frac{1}{\phi^2} ] Therefore: [ I(\phi) = \frac{n}{\theta^2} \left( \frac{1}{\phi^2} \right)^2 = \frac{n}{\left( \frac{1}{\phi} \right)^2} \cdot \frac{1}{\phi^4} = \frac{n}{\phi^2} ]
-
Conclusion:
The Fisher information for the parameter ( \phi = \frac{1}{\theta} ) based on ( n ) independent and identically distributed observations is: [ I(\phi) = \frac{n}{\phi^2} ]
Final Answer:
After simplifying, the Fisher information is n divided by φ squared. Thus,
I(φ) = n ⁄ φ²
-