Show that $P_1$ is the most efficient estimator amongst all unbiased estimators of $\theta$.

547 Views Asked by At

Suppose a random variable $X$ has a binomial distribution with parameters $n$ (the number of independent trials) and $\theta $ (the probability of success on any trial).

Define the estimator of $\theta$: $P_1 = \dfrac{X}{n}.$

Question: Show that $P_1$ is the most efficient estimator amongst all unbiased estimators of $\theta$.

I would like to know what the best approach to this question may be as I am really quite unsure.

Would it be reasonable to find the variance of $P_1$ and show that it equals the Cramer-Rao Lower Bound or is there an alternative method?

1

There are 1 best solutions below

1
On

Cramér Rao inequality works, but there are also a couple of other methods

  • Lehmann Scheffé Lemma

  • Build an UMVUE estimator with $\mathbb{E}[T|S]$ (Rao Blackwell)

If you are interested I will briefly summarize you all the 3 methods.

Let's begin with the fastest one:

Lehmann Scheffé Lemma

If S is a complete sufficient statistic and T, function of S, is an unbiased estimator of $\theta$, then T is UMVUE.

If the model is Binomial (or Bernulli that is the same model) it is easy to verify that the model belongs to the Exponential Family and its canonical Statistc $T=\sum_i {X_i}$ (number of successes) is Complete Sufficient Statistic (and also Minimal)

$\mathbb{E}[\frac{X}{n}]$ or $\mathbb{E}[\frac{\sum_i X_i}{n}]$ in case of the Bernulli pmf is unibased for $\theta$ as $\mathbb{E}[\frac{X}{n}]=\frac{1}{n}n\theta=\theta$ so the estimator is UMVUE.

For the Cramér Rao inequality, in this case, you can use the Sufficient & Necessary condition that assure you that Cramér Rao inequality becomes an equality so you can easily verify that

$$\sum_{i=1}^n\frac{\partial}{\partial\theta}log f(x_i;\theta)=K(\theta,n)[t-\theta]$$

The third way is to build by your own the UMVUE estimator of $\theta$ defining this estimator as

$\mathbb{E}[T|S]$ where T is an unbiased estimator (an estimator at your choice, the simplest you can find) and S is the CSS (Complete and Sufficient Statistic).

All the calculation above are very easy and fast