1
$\begingroup$

I am working through some textbook problems and came across a problem I am having difficulty with. The problem asks to give the bias of a point estimate, namely for a given set of data $X = \{x_1, x_2,\ldots,x_n\}$ drawn from a discrete $\text{Uniform}\{0,1,\ldots, k\}$, what is the bias of the estimator $\hat\theta = \max(X)$?

I'm pretty sure I have a representation for $E[\hat\theta]$, namely:

$$E[\hat\theta] = \sum_{i=0}^{n-1} \left(\frac{k-1}{k}\right)^{ni} \left(1-\left(\frac{k-1}{k}\right)^n\right)(k-i)\\ \hspace{.91cm} = \left(1-\left(\frac{k-1}{k}\right)^n\right)\sum_{i=0}^{n-1} \left(\frac{k-1}{k}\right)^{ni} (k-i)$$

The first term represents the probability that the numbers larger than $k-i$ have not been drawn, while $1 - (\frac{k-1}{k})^n$ represents the probability of any number in the range being drawn. Thus, those terms together represent the probability that $k-i$ is the largest number in the sample. However, to measure the bias, I need to simplify the summation as bias is $E[\hat\theta] - k$, but I am a bit unsure how to simplify the summation.

$\endgroup$
  • $\begingroup$ Could you clarify what you mean by "Uniform$(0,k)$"? Is is a discrete uniform distribution on the integers $\{0,1,\ldots, k\}$ or a continuous uniform distribution on the interval $[0,k]$? $\endgroup$ – whuber Jun 8 at 23:17
  • $\begingroup$ It is discrete, sorry about that. $\endgroup$ – user9933193 Jun 9 at 5:14
  • $\begingroup$ Please edit the information into your question. $\endgroup$ – Glen_b Jun 9 at 6:27
1
$\begingroup$

Let $X_1,\dots,X_n$ be a random sample from a discrete uniform distribution $\text{U}\{0,1,\dots,\kappa\}$. Let $\hat{\kappa}=\max\{X_1,\dots,X_n\}$. Then (Why? Work out the details), $$ \text{E}[\hat{\kappa}] = \sum_{m=0}^{\kappa-1} \Pr\{\hat{\kappa}>m\} = \sum_{m=0}^{\kappa-1} \left( 1 - \left(\frac{m+1}{\kappa+1}\right)^n\right) = \kappa - \frac{1}{(\kappa+1)^n} \sum_{m=1}^\kappa m^n, $$ in which $\sum_{m=1}^\kappa m^n$ can be written as a polynomial in $\kappa$ (if you want an explicit formula which involves Bernoulli numbers). Subtract $\kappa$ from $\text{E}[\hat{\kappa}]$ to get the bias of $\hat{\kappa}$.

kappa <- 10
n <- 15
N <- 10^5
x <- matrix(sample(0:kappa, N*n, replace = TRUE), ncol = n)
kappa_hat <- apply(x, 1, max)
(bias_MC <- mean(kappa_hat) - kappa)
(-1/(kappa+1)^n) * sum((1:kappa)^n)
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.