Summation issue in proof expectation Fisher score function for Poisson distribution is zero

356 Views Asked by At

In came across the result that the expectation of the Fisher score function is zero so I wanted to check this for a discrete distribution (in my case Poisson) since my textbook only focusses on continuous distributions (although the discrete case should be similar).

To this end I first computed the log-likelihood function $l(\lambda,y)$ which for the Poisson distribution is given by: $l(\lambda,y) = \sum_{i=1}^{n} [y_i] \cdot \ln(\lambda) - n\lambda - \ln[\prod_{i=1}^n [y_i!]]$ and consequently the score function which is given by: $u(\lambda,y) = \frac{\sum_{i=1}^{n} [y_i] - n\lambda}{\lambda}$.

Now $E[u(\lambda,Y);\lambda] = 0$ should hold. During the process of showing this result however I get stuck in the summations. I have performed the following steps:

$E[u(\lambda,Y);\lambda] = \sum_{i=0}^{\infty} [\frac{\sum_{i=1}^{n} [y_i] - n\lambda}{\lambda} \cdot \frac {\lambda^{y_i} \cdot e^{-\lambda}}{y_i!}] = \frac{e^{-\lambda}}{\lambda} \cdot (\sum_{i=0}^{\infty} [\frac{\sum_{i=1}^{n} [y_i] \cdot \lambda^{y_i} }{y_i!}] - n\lambda \cdot \sum_{i=0}^{\infty} [\frac{\lambda^{y_i}}{y_i!}]) = \ \ \ \ \frac{e^{-\lambda}}{\lambda} \cdot (\sum_{i=0}^{\infty} [\frac{\sum_{i=1}^{n} [y_i] \cdot \lambda^{y_i} }{y_i!}] - n\lambda \cdot e^{\lambda})$

(in the last step I use the power series expansion of $e^{\lambda}$).

Is the above correct? And if so, how should I proceed?