Mains PYQs
UPSC IFoS Mains Statistics paper 1 2022 questions
- (i) Find $\alpha$ such that $P$ is a finitely additive probability measure, where $\Omega = \{1, 2, 3\}$, $\mathcal{F}$ consists of all subsets of $\Omega$, and $P(\{1\}) = \frac{1}{3}$, $P(\{2\}) = \frac{1}{6}$, $P(\{3\}) = \alpha$. Compute $P(\{1, 2\})$, $P(\{1, 3\})$ and $P(\{2, 3\})$. [4]
(ii) Among $t = 60$ lottery tickets, $w = 20$ win prizes. We buy $b = 6$. What is the probability that $g = 2$ will be winning? Generalize this to arbitrary numbers $t$, $w$, $b$, $g$. [4]
- A fair coin is tossed independently $n$ times. Let $S_n$ be the number of heads obtained. Use Chebyshev's inequality to find a lower bound of the probability that $\frac{S_n}{n}$ differs from $\frac{1}{2}$ by less than 0.1 for $n = 100$.
- If $t$ is a consistent estimator of $\theta$, then prove that $t^2$ is consistent for $\theta^2$.
- Define absorbing, transient, recurrent and periodic states in a Markov chain. Also, test the periodicity of the states of a Markov chain with the following transition probability matrix :
$$P = \begin{bmatrix} 0 & 0.6 & 0.4 \\ 0 & 1 & 0 \\ 0.6 & 0.4 & 0 \end{bmatrix}$$
(rows and columns labelled 1, 2, 3)
- The joint probability density function of two random variables $X$ and $Y$ is
$$f(x, y) = \begin{cases} \frac{1}{4}(1 + xy), & |x| < 1,\ |y| < 1 \\ 0, & \text{otherwise.} \end{cases}$$
Show that $X$ and $Y$ are not independent but $X^2$ and $Y^2$ are independent.
- Suppose the probability generating function of a random variable $X$ is
$$g_X(t) = e^{\lambda(t - 1)}.$$
(i) Find the probability mass function of the random variable $X$. [3]
(ii) Find the probability generating function of $Y = 3X + 2$. [3]
(iii) Obtain variance of $Y$. [4]
- A coin is tossed. If it shows heads, you pay 2 Rupees. If it shows tails, you spin a wheel which gives the amount you win, distributed with uniform probability between 0 and 10 Rupees. Your gain (or loss) is a random variable $X$. Find the distribution function and use it to compute the probability that you will not win at least 5 Rupees.
- (i) If a random sample of size $n$ is taken from $N(\mu, \sigma^2)$ and $\sigma^2$ is known but $\mu$ is not known, then show that
$$s^2 = \frac{1}{n}\sum_{i=1}^{n}(x_i - \bar{x})^2$$
is not a sufficient estimator for $\sigma^2$. Also, suggest a sufficient estimator for $\sigma^2$. [15]
(ii) Describe the role of Cramer-Rao Inequality, Rao-Blackwell and Lehmann-Scheffé theorems in the estimation of unknown parameters of the distributions. [5]
- The observed value of mean of a random sample from $N(\theta, 1)$ distribution is 2.3. If the parameter space is $\theta = \{0, 1, 2, 3\}$, then find the maximum likelihood estimate of $\theta$.
- Let $X$ have a pdf
$$f(x; \theta) = \begin{cases} \theta x^{\theta - 1}, & 0 < x < 1 \\ 0, & \text{elsewhere.} \end{cases}$$
Find the power function of the test to test the simple hypothesis $H_0 : \theta = 1$ against the alternative simple hypothesis $H_1 : \theta = 2$ using a random sample of $X_1$ and $X_2$ of size $n = 2$ and defining the critical region to be
$$W = \left\{(x_1, x_2) : \frac{3}{4x_1} \le x_2\right\}.$$
- (i) Suppose that $X$ is uniformly distributed on the interval $(-2, 3)$. Let $Y = X^2$. Find the density function of $Y$. [12]
(ii) Let the random variables $X$ and $Y$ be jointly distributed. The marginal distribution of $X$ is
$$p_X(x) = \frac{\lambda^x e^{-\lambda}}{x!},\ x = 0, 1, 2, \ldots$$
and the conditional distribution of $Y$ given $X = x$ is
$$p_{Y/X}(y/x) = \binom{x}{y} p^y (1 - p)^{x - y},\ x > y,\ y = 0, 1, 2, \ldots, x.$$
Find the marginal distribution of $Y$. [8]
- Verify that there exists a Minimum Variance Bound Unbiased Estimator (MVBUE) of the parameter $\theta$ of the distribution
$$f(x; \theta) = \frac{e^{-\theta}\cdot\theta^x}{x!};\ x = 0, 1, 2, \ldots$$
Hence, obtain variance of the MVBUE so obtained.
- (i) Let $X_1, X_2, \ldots, X_n$ ($n > 4$) be a random sample from a population $N(\mu, \sigma^2)$. Consider the following estimators of $\mu$ :
$$U = \frac{1}{n}\sum_{i=1}^{n} X_i \quad \text{and} \quad V = \frac{1}{8}X_1 + \frac{3}{4(n - 2)}(X_2 + \ldots + X_{n-1}) + \frac{1}{8}X_n,$$
then examine whether $U$ and $V$ are unbiased estimators $\mu$. Also, find which of the two estimators is more efficient. [12]
(ii) In what situation do we make use of non-parametric tests ? Test the hypothesis of no difference between the ages of male and female employees of a certain company using the Mann-Whitney U-test for the sample data given below : [8]
| Males (Age) | 31 | 25 | 38 | 33 | 42 | 40 | 44 | 26 | 43 | 35 |
|---|---|---|---|---|---|---|---|---|---|---|
| Females (Age) | 44 | 30 | 34 | 47 | 35 | 32 | 35 | 47 | 48 | 34 |
Use 0.10 level of significance with $Z_{(0.10)} = 1.64$.
- The joint probability density function (pdf) of two random variables $X$ and $Y$ is
$$f(x, y) = \frac{9(1 + x + y)}{2(1 + x)^4 (1 + y)^4},\ 0 \le x < \infty,\ 0 \le y < \infty.$$
Find the conditional distribution of $Y$ given $X = x$.
- Explain the problem of multicollinearity. What are its consequences ? State different measures to detect the presence of multicollinearity.
- Given below is an ANOVA table of an RBD with some missing entries denoted by (*). Find out the missing entries.
RBD ANOVA TABLE
| Source | d.f. | S.S. | M.S.S. | $F_{cal}$ |
|---|---|---|---|---|
| Blocks | 3 | * | 2.040 | * |
| Treatments | 5 | 15.440 | * | * |
| Error | * | 7.030 | * | |
| Total | 23 | 28.590 | | |
Stating clearly the hypotheses to be tested, give your conclusions.
Given that $F(3, 15; 5\%) = 3.29$
$F(5, 15; 5\%) = 2.90$
- Define a linear model. Develop $100(1 - \alpha)\%$ confidence interval for an estimable linear parametric function $\lambda'\theta$ in a linear model in a Gauss-Markov set-up ($Y$; $A\theta$, $\sigma^2 I_n$).
- Explain Warner's randomised response technique for sensitive characteristics.
- Let $X_1, X_2, \ldots X_n$ be a random sample from an $N_p(\mu, \Sigma)$. Give the test statistic for testing $H_0 : \mu = \mu_0$ versus $H_1 : \mu \ne \mu_0$; and test the hypothesis $H_0 : \mu = \begin{pmatrix} 7 \\ 11 \end{pmatrix}$ using the data $X = \begin{bmatrix} 2 & 8 & 6 & 8 \\ 12 & 9 & 9 & 10 \end{bmatrix}$.
Given : $F(2, 2; 5\%) = 19$
- The data given below is on yields of a Latin Square Design of order 6.
| Total \ No. | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|
| Rows | 26.35 | 25.85 | 33.45 | 36.75 | 33.25 | 42.85 |
| Columns | 26.75 | 28.90 | 31.45 | 34.40 | 36.40 | 40.60 |
| Treatments | 22.20 | 28.30 | 34.20 | 31.60 | 38.55 | 43.55 |
$\Sigma y_{ij}^2 = 1222.84$, $\Sigma y_{ij} = 198.50$
State all hypotheses and carry out analysis. Write ANOVA, Estimates of test statistics, conclusions and interpretations.
Given that $F(6, 20; 5\%) = 2.60$
$F(5, 20; 5\%) = 2.71$
$F(6, 20; 1\%) = 3.87$
$F(5, 20; 1\%) = 4.10$
- (i) Define a multiple linear regression model. Obtain ordinary least squares (OLS) estimator of regression coefficient $\beta$ and show that it is best linear unbiased estimator (BLUE). [6]
(ii) Give the test statistic for testing linear restrictions of the type $R\beta = r$ in multiple linear regression and obtain its distribution. From this, deduce the test statistic for testing the significance of any regressor, say $X_i$. [4]
(iii) Define the OLS residual. Show that it is heteroscedastic and autocorrelated. How is this residual useful in detecting
(1) heteroscedasticity,
(2) autocorrelation, and
(3) normality ? [5]
- Define ratio estimator and regression estimator. Show that ratio estimator is biased; further obtain the bias. Derive the condition for ratio and regression estimators to be equally efficient.
- What is unequal probability sampling ? When is it to be used ? Under this scheme without replacement,
(i) explain Lahiri's method of sample selection, and
(ii) obtain Horvitz-Thompson estimator of variance of population total and give your comments.
(6+9=15)
- (i) Suppose $X_1, X_2, \ldots X_N$ are independent each distributed according to $N_k(\mu, \Sigma)$ and $C = ((C_{ij}))$ be an orthogonal matrix. Show that $Y_i = \sum_{j=1}^{N} C_{ij} X_j$ is multivariate normal with mean vector $C\mu$ and variance covariance matrix $\Sigma$. [4]
(ii) Let $X_1, X_2, \ldots X_n$ be a random sample from a normal population with mean vector $\mu$ and covariance $\Sigma$. Show that
$$\hat{\mu} = \bar{X} \text{ and } \hat{\Sigma} = \frac{1}{n}\sum_{i=1}^{n}(X_i - \bar{X})(X_i - \bar{X})'$$
are the maximum likelihood estimators of $\mu$ and $\Sigma$. [5]
(iii) Explain canonical correlation analysis. Obtain the first pair of canonical covariates and first canonical correlation given the covariance matrix of two groups of variables
$X^{(1)} = \begin{bmatrix} X_1^{(1)} \\ X_2^{(1)} \end{bmatrix}$ and $X^{(2)} = \begin{bmatrix} X_1^{(2)} \\ X_2^{(2)} \end{bmatrix}$
as $\mathrm{cov}\begin{pmatrix} X^{(1)} \\ X^{(2)} \end{pmatrix} = \begin{pmatrix} \Sigma_{11} & \Sigma_{12} \\ \Sigma_{21} & \Sigma_{22} \end{pmatrix} = \begin{bmatrix} 100 & 0 & 0 & 0 \\ 0 & 1 & 0.95 & 0 \\ 0 & 0.95 & 1 & 0 \\ 0 & 0 & 0 & 100 \end{bmatrix}$ [6]
- Consider the following regression output :
$\hat{y} = 0.2033 + 0.6560X$
$SE = (0.0976)\ (0.1961)$
Residual sum of squares = 0.0544
Regression sum of squares = 0.0358
where
$Y$ = Labour Force Participation Rate (LFPR) of women in 1972
$X$ = LFPR of women in 1968
The regression results were obtained from a sample of 19 cities. The figures in the parentheses give Standard Error (SE) of the estimate. Critical t value at 1%, df = 17 is $t_{17, 1\%} = 2.567$.
(i) How do you interpret the result ? Compute the coefficient of determination $R^2$.
(ii) Test the hypothesis that $H_0 : \beta = 0$ vs $H_1 : \beta > 0$. Which test do you use and why ? What are the underlying assumptions of the test you use ?
(iii) Write 95% confidence interval for $\alpha$ and $\beta$.
(iv) Test that the hypothesis LFPR of women in 1972 is not depending on LFPR of women in 1968.
Given $t(17; 5\%) = 1.740$
$F(1, 17; 5\%) = 4.45$
- (i) What is confounding ? Explain briefly, types of confounding and compare them. [7]
(ii) Obtain single replication of a $2^5$ factorial experiment confounding interactions $X = ABC$ and $Y = ACDE$. [8]
- (i) Let ($Y$; $A\theta$, $\sigma^2 I$) be a Gauss-Markov set-up. Obtain the least squares estimator of $\theta$ and variance of the best estimator of estimable linear parametric function $\lambda'\theta$. [6]
(ii) Define a linear hypothesis. Derive the maximum likelihood ratio test statistic for testing a linear hypothesis $H_0 : \xi_1 = \ldots = \xi_r$ in a Gauss-Markov model ($Y$; $\xi$, $\sigma^2 I_n$), where $\xi$ is a mean vector of $Y$ and $\sigma^2 I_n$ is the dispersion matrix of $Y$. Obtain the distribution of the test statistic. [9]
- Define : Connectedness, Orthogonality and Balancedness. State and prove a necessary and sufficient condition for an incomplete block design to be balanced.