$30
1. Practice with MME
(a) The Gamma(x, y) distribution has mean ๐ฅ โ ๐ฆ and variance ๐ฅ โ ๐ฆ . Find MME for ๐ฅ and ๐ฆ.
(b) Find MME ๐ and ๐ for the Uniform(a, b) distribution. Express your final answer in terms of the
sample mean, ๐ โ ๐ /๐, and sample variance, ๐ โ ๐ /๐ ๐ .
Consistency of MLE
Let ๐ , ๐ , โฆ , ๐ be distributed as Exponential(1/ฮฒ), all i.i.d. Show that the MLE(๐ฝ) will converge to the unknown parameter ฮฒ. Prove this by showing that bias(๐ฝ) and se(๐ฝ tends to 0 as n tends to โ. You can use the fact that the mean and variance of Exponential(ฮป) are 1/ฮป and 1/ฮป2, respectively.
Practice with MLE
(a) Let ๐ , ๐ , โฆ , ๐ be distributed i.i.d. as Poisson(ฮป). Find the MLE of ฮป. (3 points)
(b) Let ๐ , ๐ , โฆ , ๐ be distributed i.i.d. as Normal(ยต, ฯ2). Show that the MLE of ยต and ฯ2 is the same as
the sample mean and (uncorrected) sample variance, respectively. (4 points) (c) Let ๐ , ๐ , โฆ , ๐ ~ Normal(ฮธ, 1). Let ฮด =๐ธ ๐ผ . Use the Equivariance property to show that the
MLE of ฮด is ๐ท โ ๐ , where ๐ท is the CDF of the standard Normal. You can use the MLE of the
Normal as provided in 3(b).
Parametric Inference with Data Samples
2 ๐ค๐๐กโ ๐๐๐๐ ๐
Let ๐ , where ๐ is unknown. Let D = {2, 3, 2} be drawn i.i.d. from X.
3 ๐๐กโ๐๐๐ค๐๐ ๐
(a) Derive ๐ using D as the sample data. Clearly show all your steps. (3 points)
(b) Provide a numerical estimate of the 95%ile confidence intervals for ๐ . Start by deriving ๐ ๐ ๐ : first derive ๐ ๐ ๐ in terms of ๐, and then estimate ๐ ๐ ๐ , as in class. Show all
your steps. Your final answer should be a numerical range. (4 points) (c) Derive ๐ using D as the sample data. Clearly show all your steps.
MME versus MLE using real data
For this question, we will use the acceleration, model, and mpg data from the Autoโmpg dataset (https://www.kaggle.com/uciml/autompgโdataset). Please use the data files on the class website. We will assume that acceleration is Normal(ฮผ, ฯ2) distributed, model year is Uniform(a, b) distributed, and mpg is Exponential(ฮป) distributed. You are to find the MME and MLE estimates of the parameters of the distributions for all 3 datasets. For the Normal MME and Uniform MLE, you can directly use the results from class. For the Normal MLE, use the result from Q3(b); for Uniform MME, use the result from Q1(b).
For the Exponential, we will first derive the estimates.
(a) For the Exp(ฮป) distribution, find the ๐ .
(b) For the Exp(ฮป) distribution, find the ๐ .
(c) For the 3 datasets, find the MME estimates. That is, find the MME for ฮผ and ฯ2 for the acceleration dataset, a and b for the model dataset, and ฮป for the mpg dataset. Provide your answer as a number
with 3 significant digits.
(d) Same as part (c), but this time find the MLE estimates.
Clinical Testing
Consider the sick patient example from class. In a clinical trial of a new disease detection test, there were 100 healthy patients and 100 sick patients. The test correctly identified 98 out of the 100 healthy patients as healthy. The test also correctly identified 99 of the 100 sick patients as sick. The remaining patients were incorrectly classified.
(a) What is the precision of the test?
(1 point)
(b) What is the recall of the test?
(1 point)
(c) What is the Type I error of the test?
(1 point)
(d) What is the Type II error of the test?
(
Waldโs test
(a) Suppose the null hypothesis is H0: ฮธ = ฮธ0, but the true value of ฮธ is ฮธ*. Show that, under Waldโs test, the probability of a Type II error is ๐ท โ ๐ง/ ๐ท โ ๐ง/ .
(Hints: (i) might help to draw a figure; (ii) think about the distribution of the estimate.)
(b) You observe 46 successes in 100 trials of a coin. If the null hypothesis is that the coin is unbiased, use the Waldโs test with the MLE or MME with ฮฑ = 0.05 to Reject/Accept the null. What if the null
hypothesis is that the coin has p=0.7?
More on Waldโs test
(a) Use q8_a.csv dataset and assume it is distributed as Normal(ฮธ, ฯ2). Apply the Waldโs test with ฮฑ =
0.02 to check whether the true mean is ๐ 0.5. Use sample mean to obtain ๐ and corrected
sample variance estimator for obtaining ๐ .
(b) Use q8_b_X.csv and q8_b_Y.csv available at the class website for this question. Each contains 750 samples for X and Y drawn from two independent Normal distributions. Without worrying about the applicability of the test, use Waldโs 2โpopulation test with ฮฑ = 0.05 to test whether the population means of X and Y are same (null) or not (alternative). Is this test applicable here?