91Ó°ÊÓ

The Poisson distribution is a probability distribution used to model the number of events occurring within a fixed interval of time or space. These events happen independently at a constant average rate, known as the parameter \( \lambda \). In our case, \( Y_{1}, Y_{2}, \ldots, Y_{n} \) are random samples from a population with a Poisson distribution.

Key characteristics of the Poisson distribution include:

• Events are independent of each other.
• A fixed period or space is considered for the model.
• The number of occurrences can be any non-negative integer.

For example, if we are analyzing the number of emails a person receives per hour, and that average number is 3 emails (\( \lambda = 3 \)), the Poisson distribution can help us determine the probability of receiving a different number of emails during that hour.

The probability mass function (PMF) for a Poisson distribution is given by:\[ P(Y = k) = \frac{e^{-\lambda} \lambda^k}{k!} \]where:

• \( k \) is the number of occurrences.
• \( e \) is approximately 2.71828, the base of the natural logarithm.
• \( \lambda \) is the average rate of occurrence.

The Likelihood Ratio Test (LRT) is a statistical test used to compare the fit of two competing models based on their likelihoods. It is particularly useful in deciding between a null hypothesis \( H_0 \) and an alternative hypothesis \( H_a \). In our scenario, we aim to test if the parameter \( \lambda \) of a Poisson distribution differs from a specified value.

Here's how the LRT works for the Poisson distribution in this exercise:

• The likelihood of the data under the null hypothesis \( \lambda_0 \) is compared to the likelihood under the alternative hypothesis \( \lambda_a \).
• The ratio of these likelihoods forms the test statistic that determines the rejection region.
• The test is most powerful when the likelihood ratio is maximized.

For a Poisson-distributed random sample, the likelihood function is given by:\[ L(\lambda) = e^{-n\lambda} \lambda^{\sum_{i=1}^n Y_i} \prod_{i=1}^{n} \frac{1}{Y_i!} \]When dealing with likelihood ratios, we analyze the expression:\[ \frac{L(\lambda_a)}{L(\lambda_0)} = \left(\frac{\lambda_a}{\lambda_0}\right)^{T} e^{-n(\lambda_a - \lambda_0)} \]where \( T = \sum_{i=1}^n Y_i \). This helps us decide whether to reject or accept \( H_0 \). We reject \( H_0 \) if this ratio surpasses a certain critical value, which is determined by the significance level \( \alpha \).

The LRT effectively tells us whether the sample data provides enough evidence against the null hypothesis, always considering our defined threshold for making a decision.

A Uniformly Most Powerful (UMP) test is the best possible statistical test that maintains the highest power for a given significance level over all possible values of the parameter under consideration in the alternative hypothesis. However, deriving such a test isn't straightforward, especially for more complex hypotheses.

In the context of our exercise:

• We test \( H_0: \lambda = \lambda_0 \) against \( H_a: \lambda > \lambda_0 \), aiming for a UMP test.
• A UMP test for the Poisson distribution typically focuses on one-sided alternatives, like \( \lambda = \lambda_a > \lambda_0 \).
• The likelihood ratio test can serve as UMP specifically for simple cases where \( \lambda \) in the alternative hypothesis is a fixed value.

In practical terms, a UMP test for the case where \( \lambda \) can be any value greater than \( \lambda_0 \) does not exist. This is because the power of the test, which is its ability to correctly reject a false null hypothesis, cannot be uniformly maintained across a range of \( \lambda \) values. Instead, the likelihood ratio test we're using only provides the most powerful test for targeted values like \( \lambda = \lambda_a \). Therefore, understanding that the search for a UMP test often results in tests that are optimal under particular conditions, but not universally, is crucial for grasping why certain tests are chosen.