Mains PYQs
UPSC IFoS Mains Statistics paper 1 2017 questions
- The random variable X has the exponential probability density function (pdf) given by
$f(x) = \lambda \exp(-\lambda x),\ x \ge 0,\ \lambda > 0.$
Show that, for any $c > 0$, $P(X > c) = \exp(-\lambda c)$.
Hence show that, for any $x > c$, $P(X > x \mid X > c) = \exp(-\lambda(x - c))$.
Deduce the conditional pdf of X given that $X > c$, and comment briefly.
- 12.5% of the candidates in a specific examination of a certain year are known to have a score of at least 70% in Statistics Paper I, while another 18.1% have a score of at most 38%. Assuming the underlying distribution to be normal, estimate the probability that in a random sample of 5 such candidates, 2 will have a score of 60% or more. [You may use the following information : For a standard normal variate X, $P\{X \le K\} = 0.637, 0.875$ and $0.919$ for $K = 0.35, 1.15$ and $1.40$ respectively]
- The random variable Y has geometric distribution with parameter $p\ (0 < p < 1)$, i.e.,
$P(Y = y) = (1 - p)^y \cdot p$ for $y = 0, 1, 2, \ldots$,
(i) Find the probability generating function of Y, and hence find the mean and variance of this distribution. [8]
(ii) The random variables $Y_1, Y_2, \ldots, Y_n$ constitute a random sample from this distribution. Define $\overline{Y} = \frac{1}{n}\sum_{i=1}^{n} Y_i$. Find an unbiased estimator of $\frac{1}{p}$ (to be shown), and check for its consistency. [8]
- Let $X_1, X_2, \ldots, X_n$ be a random sample from a distribution with probability density function
$f(x) = \beta(1 - x)^{\beta - 1},\ 0 < x < 1,$
where $\beta\ (> 0)$ is an unknown parameter.
(i) Find the maximum likelihood estimator, $\hat{\beta}$, of $\beta$. [5]
(ii) Suppose that the values of $X_1, X_2, \ldots, X_n$ are not known, but you do know Y, the number of $X_i$ less than 0.5. State the distribution of Y. [3]
- The random variables X and Y are jointly distributed with probability density function (pdf)
$f(x, y) = \begin{cases} \dfrac{1}{3\log 2}\left(\dfrac{x}{y} + \dfrac{y}{x}\right), & 1 \le x \le 2,\ 1 \le y \le 2 \\ 0, & \text{otherwise.} \end{cases}$
(i) Find the marginal pdf of X. [4]
(ii) Find the conditional pdf $f(y|x)$, for $1 \le x \le 2,\ 1 \le y \le 2$, and hence evaluate $P[Y < 1.5 \mid X = 1]$. [6]
- Suppose X and Y are independent random variables having Poisson distributions with respective means $\lambda\ (> 0)$ and $\mu\ (> 0)$.
(i) Show that $X + Y$ also follows a Poisson distribution. [5]
(ii) Find $P(X = k \mid X + Y = n)$, where k and n are integers with $0 \le k \le n$. For given $n > 0$, name the distribution you have obtained. [5]
- If a certain team loses one of its matches, then it has probability 0.5 of losing the next match and probability 0.4 of drawing it. If the team draws a match, then it has probability 0.3 of losing the next match and probability 0.4 of drawing it. If the team wins a match, then it has probability 0.2 of losing the next match and probability 0.4 of drawing it.
(i) Model this as a Markov chain, and write down its transition matrix. [5]
(ii) If the team loses its first game of the season, find the probability that it wins its third game. [5]
- Suppose $(X, Y)$ follows bivariate normal distribution $N_2(0, 0, 1, 1, \rho)$. Then show that
$\rho = \cos q\pi$, where $q = P\{XY < 0\}$.
- (i) State the Central Limit Theorem (CLT) for sums of independently and identically distributed random variables. [4]
(ii) Examine if CLT holds for the sequence of random variables $\{X_K\}$ with $P\{X_K = \mp\sqrt{2K - 1}\} = \frac{1}{2},\ K = 1, 2, \ldots$ [6]
- The probability that a certain tomato seed germinates is $\theta$. A gardener sows a set of n such seeds and finds that x of them have germinated. It can be assumed that seeds germinate independently of one another. Find the posterior distribution of $\theta$, assuming that its prior distribution is beta with parameters $\alpha$ and $\beta$.
- (i) Define a family of UMAU confidence sets with a given confidence coefficient $1 - \alpha$ for a parameter $\theta$. [3]
(ii) Derive a UMAU confidence set for $\sigma^2$ with confidence coefficient $1 - \alpha$ in sampling from $N(\mu, \sigma^2)$ with $\mu$ unknown. Show that this set is actually an interval. [7]
- Let $X_1, X_2, \ldots, X_n$ be a random sample from a uniform distribution
$P(X_i = K) = \begin{cases} \dfrac{1}{N}, & K = 1, 2, \ldots, N \\ 0, & \text{elsewhere.} \end{cases}$
Find a Uniformly Minimum Variance Unbiased Estimator (UMVUE) (to be shown) of N.
- (i) Define a maximum likelihood estimator for a parameter $\theta$, and state the large sample properties of this type of estimator under regularity conditions, to be stated clearly. [8]
(ii) Suppose that the number of a particular plant species in sampling quadrats follows a Poisson distribution with mean $\lambda$ and it is required to estimate $\theta = \lambda^2$. A random sample of n such quadrats yields the numbers $X_1, X_2, \ldots, X_n$. Find the unbiased estimator of $\theta$ (to be shown), and also find the Cramer-Rao lower bound for the variance of this unbiased estimator of $\theta$. [12]
- Discuss how you would estimate the unknown number of fishes of a given size (say, $\ge 1$ kg) in a large pond, based on hypergeometric probability model, using catch-recatch method. Indicate how this technique can be used for estimating the unknown number of tigers in a large forest, using pug-mark method.
[Hint : Pug-mark can be inflicted on a tiger from a distance when it comes for drinking in a pond. A tiger thus marked can be identified at a later time, again from a distance.]
- Discuss Wald-Wolfowitz Runs test to examine if two samples of sizes $n_1$ and $n_2$ come from an identical population, against the alternative that the two populations from which the two samples have been taken, differ in any respect whatsoever.
- Discuss the randomised response technique for estimating a sensitive parameter, such as the proportion of tax-evaders in a community.
- Explain the terms 'Estimable Parametric Function', 'Error Function' and 'Best Linear Unbiased Estimator (BLUE)' in connection with linear estimation. Show that the sample mean in sampling from $N(\mu, \sigma^2)$ is BLUE for $\mu$.
- The variance of a stratified random sample mean can be written as
$\mathrm{Var}(\bar{y}_{st}) = \sum_{i=1}^{k} \frac{N_i^2}{N^2}\left(\frac{1}{n_i} - \frac{1}{N_i}\right) S_i^2.$
Explain the (conventional) notation used here. Suppose the total cost of sampling is
$C = C_0 + \sum_{i=1}^{k} C_i n_i,$
where $C_0, C_1, \ldots, C_k$ are positive constants.
If $\mathrm{Var}(\bar{y}_{st})$ is fixed and the stratum sample sizes are chosen to minimise the total cost of sampling, show that the $i^{th}$ stratum sample size $n_i$ is proportional to
$\frac{N_i S_i / \sqrt{C_i}}{\sum_{i=1}^{k} N_i S_i / \sqrt{C_i}},\ i = 1, 2, \ldots, k.$
- (i) State briefly three reasons why an analyst may wish to perform a principal component analysis. [5]
(ii) Under what circumstances would it be sensible to use the variance-covariance matrix instead of the correlation matrix in principal component analysis ? [5]
- Suppose $Y_1, Y_2, Y_3, Y_4$ are independent with
$E(Y_1) = E(Y_2) = \theta_1 + \theta_2$
$E(Y_3) = E(Y_4) = \theta_1 + \theta_3$
$\mathrm{Var}(Y_i) = \sigma^2,\ i = 1, 2, 3, 4.$
Determine the condition of estimability of the parametric function $l'\theta = l_1\theta_1 + l_2\theta_2 + l_3\theta_3$. Obtain a solution of the normal equations and the sum of squares due to error.
- Write in detail about the use of orthogonal polynomials in regression analysis, and how we choose its appropriate degree for a given data set.
- Show that, with usual notation,
$1 - r_{1.2.3\ldots p}^2 = (1 - r_{12}^2)(1 - r_{13.2}^2)\ldots(1 - r_{1p.23\ldots p-1}^2).$
Discuss the significance of this result.
- Describe Lahiri's method, with justification, of drawing a sample of size n from a population of size N with probabilities proportional to the sizes of the respective units.
- Discuss the advantages of factorial designs. Give the complete analysis, including the layout, of a $3^2$-factorial design laid out in r replicates of 3 incomplete blocks each, totally confounding the factorial effect $AB^2$.
- Explain the concept of missing-plot technique and the analysis of an $r \times r$ Latin Square design with one missing value, clearly stating the assumptions needed.
- Define a symmetrical balanced incomplete block design and show that the number of treatments common between any two blocks of a symmetrical balanced incomplete block design is a constant, say $\lambda$.
- Show that a randomised block design is an orthogonal design.
- Define one-stage and two-stage cluster sampling. How do cluster sampling and stratified sampling differ, both in construction and use ? Give an example of a survey that uses both stratification and clustering in the sample design.
- Define Hotelling's $T^2$ statistic, and indicate its applications. Show that
$T^2(p, m) = \frac{mp}{m - p + 1} F_{p,\, m-p+1},$
symbols having their usual significance.
- (i) Discuss the need of ratio estimation and find the bias and the variance of the sample estimator $\hat{R}$, of population ratio R. [7]
(ii) Explain briefly the circumstances under which the ratio estimator of a population mean will be less precise than the sample mean of a simple random sample of the same total size. [3]
- Show that an incomplete block design is connected if and only if the rank of the C-matrix is $v - 1$, where v denotes the number of treatments.