Scribd
Upload
3 views2 pages

Practical 5

The document outlines a practical assignment for a Statistical Modelling course, focusing on inference in various distributions including Bernoulli, exponential, and geometric. It includes questions on likelihood functions, maximum likelihood estimation, hypothesis testing, and model fitting, with specific tasks related to real-world data analysis. Additionally, it discusses linear regression and the properties of flexible models.

Uploaded by

Paria
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
3 views2 pages

Practical 5

The document outlines a practical assignment for a Statistical Modelling course, focusing on inference in various distributions including Bernoulli, exponential, and geometric. It includes questions on likelihood functions, maximum likelihood estimation, hypothesis testing, and model fitting, with specific tasks related to real-world data analysis. Additionally, it discusses linear regression and the properties of flexible models.

Uploaded by

Paria
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Statistical Modelling 2023-2024

(Prof. C. Ley, S. Loizidou)


Practical 5

Question 1 (Inference in the Bernoulli distribution)

Consider a sample (X1 , . . . , Xn ) from the Bernoulli family with unknown parameter 0 < p < 1. The
density function is given by
f (x; p) = px (1 − p)1−x , x = 0, 1.
We know that for a random variable X that is Bernoulli distributed, E(X) = p and Var(X) = p(1 − p).
(a) Find the likelihood and log-likelihood function.

(b) Show that the score function of p is given by S(p) = ni=1 Xi /p − (n − ni=1 Xi )/(1 − p).
P P

(c) Show that the MLE of p equals p̂MLE = n−1 ni=1 Xi = X̄n .
P

(d) Calculate the total Fisher information matrix of p.

(e) Compute the asymptotic distribution of p̂MLE . Check that the asymptotic variance coincides with
the inverse Fisher information.

(f) Now consider testing 


H0 : p = p0 ,
HA : p ̸= p0 .
Derive the Wald, score and likelihood ratio test of H0 versus HA . What is the rejection region of
each test?

(g) Suppose n = 1000 observations are randomly selected for a poll concerning an election with two
candidates. Of the 1000 participants, 478 answered they would vote for candidate A and 522
answered they would vote for candidate B. Perform all above hypothesis tests to test if one of the
candidates has a significant edge over the other and compare the results.

Question 2 (Inference in the exponential distribution)

Do exactly the same questions (a-f) as above, except that you replace the Bernoulli distribution with the
exponential distribution and that this time no score function or MLE are indicated.

Question 3 (Likelihood ratio test for the geometric distribution)

Suppose that we have a set of random observations from the Geometric distribution with probability
mass function P (X = x) = (1 − p)x−1 p for x = 0, 1, 2, · · · and 0 < p < 1. We are interested in testing
H0 : p = 0.5 versus Ha : p ̸= 0.5. Derive the likelihood ratio test for H0 against Ha . What is the rejection
region of the test?

Question 4 (Which distribution fits the data best?)

The article “A tractable, parsimonious and highly flexible model for cylindrical data, with applications”
introduces a new probability model, the WeiSSVM, and compares its fitting abilities to other models.
The authors have obtained the following table when fitting various models to a wind-temperature data
set:

1
(a) Which model provides the best fit for the data at hand, and why?

(b) It happens that the JW1/ExpVM is a submodel of the WeiSSVM, corresponding to fixing the param-
eters α and λ to the values 1 and 0 (the parameter γ is absent from both models). Explain how to
build a statistical test allowing to test for the JW1/ExpVM submodel within the WeiSSVM model.
The authors could calculate the corresponding test statistic with the values in their table. You can
do that, too, with your acquired knowledge! Explain how.

Question 5 (Linear regression)

Consider a classical linear regression model of the form Y = a0 + a1 X1 + a2 X2 + ϵ with some random error
term ϵ ∼ N (0, σ 2 ) and parameters a0 , a1 , a2 ∈ R and σ 2 > 0. Which properties of a “good flexible model”
does such a linear regression possess, and which not? Motivate your response. Would the presence of an
interaction parameter alter your opinion?

You might also like