Mains PYQs
UPSC IFoS Mains Statistics paper 1 2018 questions
- (i) For n events $A_1, A_2, \ldots, A_n$, show that
$$P\left(\bigcap_{i=1}^{n} A_i\right) \geq \sum_{i=1}^{n} P(A_i) - (n-1).$$ [3]
(ii) Let $\{A_n\}$ be an increasing sequence of sets (events), then show that
$$\lim_{n\to\infty} P(A_n) = P\left(\lim_{n\to\infty} A_n\right) = P\left(\bigcup_{n=1}^{\infty} A_n\right).$$ [5]
- Let $\Omega = \{1, 2, 3, 4\}$ and $p_i = P\{i\}$, i = 1, 2, 3, 4.
Assume that
$p_1 = \frac{\sqrt{2}}{2} - \frac{1}{4}$,
$p_2 = \frac{1}{4}$,
$p_3 = \frac{3}{4} - \frac{\sqrt{2}}{2}$, and
$p_4 = \frac{1}{4}$.
Define the events
$E_1 = \{1, 3\}$, $E_2 = \{2, 3\}$ and $E_3 = \{3, 4\}$.
Check whether $E_1$, $E_2$ and $E_3$ are mutually independent.
- Suppose that $X_1, \ldots, X_n$ form a random sample from a Uniform distribution on the interval $[\theta_1, \theta_2]$ where both $\theta_1$ and $\theta_2$ are unknown ($-\infty < \theta_1 < \theta_2 < \infty$). Find the maximum likelihood estimators of $\theta_1$ and $\theta_2$.
- Suppose that $X_1, \ldots, X_n$ form a random sample from a Gamma distribution for which the value of parameter $\alpha$ is unknown ($\alpha > 0$) and value of parameter $\beta$ is known. Show that the joint probability density function of $X_1, \ldots, X_n$ has a Monotone Likelihood Ratio (MLR).
- (i) Let $\{X_n\}$ be a sequence of events such that
$P[X_n = n^2 - 1] = \frac{1}{n^2}$,
$P[X_n = -1] = 1 - \frac{1}{n^2}$.
Check whether Strong Law of Large Numbers holds. Also comment about Weak Law of Large Numbers.
(ii) Let
$X_n \xrightarrow{prob.} X$ and $Y_n \xrightarrow{prob.} Y$, then show that
$X_nY_n \longrightarrow XY$ in probability.
- Suppose that a diagnostic test for HIV(+) status has both sensitivity (P(Test positive | Disease)) and specificity (P(Test negative | No disease)) equal to 0.95 and the real possibility (P(Disease)) is 0.005. Find the probability that a subject is truely HIV(+) given that the diagnostic test is positive.
- Let a continuous random variable X have probability density function (pdf) given by
$$f(x) = \begin{cases} \frac{2x}{\pi^2}, & 0 < x < \pi \\ 0, & \text{otherwise} \end{cases}$$
Find the probability density function of $Y = \sin X$.
- Let (X, Y) be uniformly distributed over
$R = \{(x, y) | x^2 + y^2 \leq 1,\ y \geq 0\}$
Find
(i) the distributions of (X | Y = y) and (Y | X = x),
(ii) E(Y | X = x) and E(X | Y = y), and
(iii) Correlation coefficient $\rho_{XY}$.
- Let X follow Pareto distribution with parameters $\alpha$ and $\theta$ with probability density function (pdf)
$$f(x) = \frac{\alpha\,\theta^{\alpha}}{(x+\theta)^{\alpha+1}},\ x > 0,$$
then show that
$$Y = \ln\left(\frac{X}{\theta}\right)$$
follows Logistic distribution.
- Suppose that $x_1, \ldots, x_n$ form a random sample from a Beta distribution with parameters $\alpha$ and $\beta$ where the value of $\alpha$ is known and the value of $\beta$ is unknown ($\beta > 0$). Obtain a sufficient statistic for $\beta$.
- Let $x_1, \ldots, x_n$ be a random sample from $P(\theta)$. Find Uniformly Minimum Variance Unbiased Estimator (UMVUE) of $e^{-\theta}$. Compute the estimator based on the following sample observations :
1, 3, 0, 8, 5, 6, 9, 2, 7, 5
- The lifetimes of fluorescent lamps are independent exponential random variables with parameter $\beta$. Suppose that $\beta$ has a prior distribution Gamma with parameters 4 and 20,000. After we observe 5 lamps with lifetimes 2911, 3403, 3237, 3509 and 3118 (in hours), we want to predict the lifetime $X_6$ of the next lamp. Obtain the predictive distribution of $X_6$.
- Suppose $X_1, X_2, \ldots, X_n \sim N(\mu, \sigma^2)$ where $\sigma^2$ is known.
(i) Find the Likelihood Ratio Test (LRT) for $H_0 : \mu \leq \mu_0$ vs $H_1 : \mu > \mu_0$.
(ii) Show that the test in (i) is a UMP test.
- For any sequence $\{A_n, n \geq 1\}$ of events with $\sum_{k=1}^{\infty} P(A_k) < \infty$, comment on the following with justification :
(i) $P\left(\bigcup_{k=n}^{\infty} A_k\right) \leq \sum_{k=n}^{\infty} P(A_k)$ [4]
(ii) $P(\limsup_n A_n) = 0$ [6]
- Suppose that an examination contains 99 questions arranged in a sequence from the easiest to the most difficult. Suppose that the probability that a particular student will answer the first question correctly is 0.99, the probability that he will answer the second question correctly is 0.98 and in general, the probability that he will answer the $i^{th}$ question correctly is $1 - \frac{i}{100}$, for i = 1, 2, ..., 99. It is assumed that all questions will be answered independently and that the student must answer at least 60 questions correctly to pass the examination. What is the probability that the student will pass ?
(Tables 1(a) and 1(b) are provided at the end.)
- Let $\{X_0, X_1, \ldots\}$ be a Markov Chain (MC) with transition probability matrix
$$P = \begin{bmatrix} 0 & 1 & 0 \\ 0 & 0 & 1 \\ p & 1-p & 0 \end{bmatrix},\ 0 < p < 1$$
(rows and columns indexed by states 1, 2, 3)
Let $g(x) = \begin{cases} 0, & x = 1 \\ 1, & x = 2, 3. \end{cases}$
If $Y_n = g(X_n)$, $n \geq 0$, show that $\{Y_0, Y_1, \ldots\}$ is not a Markov Chain (MC).
- Derive Kolmogorov – Smirnov test for two samples and illustrate.
- Let N be the incidence matrix of a Balanced Incomplete Block Design (BIBD) of order b × v. Show that $b \geq v$.
- (i) A plane is fitted to n = 33 observations on $(X_1, X_2, Y)$ and it is found that overall regression is just significant at $\alpha = 0.05$ level. Find out $R^2$ based on the available information. [3]
(Tables 2(a) and 2(b) are provided at the end.)
(ii) In a one-way layout, show that for all values of i, i' and j,
j = 1, 2, ..., n,
i, i' = 1, 2, ..., p,
$w_1 = y_{ij} - \bar{y}_{i.}$,
$w_2 = \bar{y}_{i'.} - \bar{y}_{i..}$,
$w_3 = \bar{y}_{i.}$.
are uncorrelated with each other (under usual assumptions). [5]
- Let $\mathbf{Y}$ follow $\mathbf{N}_2\,(\mu_1, \mu_2, \sigma_1^2, \sigma_2^2, \rho)$. Find distribution of $\mathbf{Z} = C\mathbf{Y}$ where C is a $2 \times 2$ non-singular matrix. Also give an explicit form of matrix C such that $C\Sigma C' = I$, where $\Sigma$ is the dispersion matrix of $\mathbf{Y}$.
- Construct a $2^3$ design in two blocks where ABC is confounded.
- Find the condition under which systematic sample mean is more efficient than a simple random sample mean.
- Let $X_1, X_2, \ldots, X_n$ be a random sample from $\mathbf{N}_p(\mu, \Sigma)$. Consider the hypothesis $H_0 : \mu = k\mu_0$, where $\Sigma$ and $\mu_0$ are known. Derive the MLE for k. Show that $-2$ log likelihood ratio is
$$n\bar{X}'\Sigma^{-1}\left(\Sigma - (\mu_0'\Sigma^{-1}\mu_0)^{-1}\mu_0\mu_0'\right)\Sigma^{-1}\bar{X}.$$
Deduce the distribution of the statistic.
- Suppose that a chemical engineer considers the time of reaction for a chemical process as a function of the type of catalyst used. Four catalysts are being investigated and the procedure consists of selecting a batch of raw materials. The observations recorded are as shown below :
| Catalyst | Batch 1 | Batch 2 | Batch 3 | Batch 4 |
|---|---|---|---|---|
| 1 | 73 | 75 | 68 | - |
| 2 | 75 | - | 72 | 75 |
| 3 | 73 | 74 | - | 71 |
| 4 | - | 75 | 67 | 72 |
(columns are Batch of Raw Materials)
Identify the design and analyse it.
(Tables 2(a) and 2(b) are provided at the end.)
- Let $X_1$ and $X_2$ be independent random vectors of order $(n_1 \times p)$ and $(n_2 \times p)$ respectively and let $n_i$ rows of $X_i$ (i = 1, 2) be independently and identically distributed as $\mathbf{N}_p(\mu_i, \Sigma_i)$.
Show that for $\mu_1 = \mu_2$ and $\Sigma_1 = \Sigma_2$,
$$\frac{n_1 n_2}{n}D^2 \simeq T^2(p, n-2),$$
where $n = n_1 + n_2$, $D^2$ denotes sample Mahalanobis distance statistic and $T^2$ denotes the Hotelling's $T^2$.
- Find an unbiased estimator of the population mean under probability proportional to size (PPS) sampling with replacement. Find the variance of this estimator and also give an estimator of this variance.
- In a simple linear regression problem $Y = \beta_0 + \beta_1 X + \varepsilon$, $\varepsilon \sim N(0, \sigma^2)$, in which a patient's response Y to a new drug B is to be related to his response X to a standard drug A. Suppose 10 pairs of observations $(y_i, x_i)$, i = 1... 10 are obtained.
(i) Determine the MLEs of $\hat{\beta}_0$, $\hat{\beta}_1$ and $\hat{\sigma}^2$ of their corresponding parameters,
(ii) Obtain variance ($\hat{\beta}_0$), variance ($\hat{\beta}_1$) and corr ($\hat{\beta}_0$, $\hat{\beta}_1$).
- In a two-way layout with k observations in each cell ($k \geq 2$), construct a test of the null hypothesis that all the interactions are zero. (Model and all assumptions are required to be specified in detail).
- For the covariance matrix given by
$$\Sigma = \begin{pmatrix} 1 & 4 \\ 4 & 100 \end{pmatrix},$$
obtain the proportion of the total population variance explained by the first principal component.
- In simple random sampling where n paired observations $(y_i, x_i)$, i = 1 ... n are drawn, obtain the regression estimators and derive its large sample variance.
- Use the method of Lagrange's multipliers to show that for a least squares problem,
$T = (\mathbf{Y} - X\beta)'(\mathbf{Y} - X\beta) + \lambda'(\mathbf{d} - C\beta)$
is minimized with respect to $\beta$ and $\lambda$ where
$\hat{\beta} = \mathbf{b} + (X'X)^{-1}C'[C(X'X)^{-1}C']^{-1}(\mathbf{d} - C\mathbf{b})$
where $\mathbf{b}$ is unrestricted least squares estimator of $\beta$.
- Let there be two populations $\Pi_1$ and $\Pi_2$. It is known that about 30% of all objects belong to $\Pi_2$ and
C(2 | 1) : cost incurred when a $\Pi_1$ observation is incorrectly classified as $\Pi_2$ observation = 15;
C(1 | 2) : cost incurred when a $\Pi_2$ observation is incorrectly classified as $\Pi_1$ observation = 10.
Suppose the two density functions $f_1(x)$ and $f_2(x)$ (corresponding to $\Pi_1$ and $\Pi_2$) are evaluated at a new observation $x_0$ and $f_1(x_0) = 0.32$, $f_2(x_0) = 0.56$.
Can the new observation be classified from $\Pi_1$ or $\Pi_2$ ?
- A chemical experiment was performed to investigate the effect of extrusion temperature $X_1$ and cooling temperature $X_2$ on the compressibility of a finished product. Knowledge of the process suggested that a model of the form $Y = \beta_0 + \beta_1X_1 + \beta_2X_2 + B_{12}X_1X_2 + \varepsilon$ would satisfactorily explain the variation observed. Two levels of extrusion temperature and two levels of cooling temperature were chosen and all four of the combinations were performed. Each of the four experiments was carried out four times and the data yielded the following information :
ANOVA
| S.V | d.f | S.S | MSS |
|---|---|---|---|
| Due to reg. | - | 881.2500 | |
| $\beta_0$ | 1 | 798.0625 | |
| $\beta_1$ | 1 | 18.0625 | |
| $\beta_2$ | - | - | |
| $\beta_{12}$ | - | 5.0625 | |
| Residual | - | - | |
| Total | 16 | 921.000 | |
Using $\alpha = 0.05$, examine the following questions :
(i) Is the overall regression equation statistically significant ?
(ii) Are all $\beta$ significant ?
(Tables 2(a) and 2(b) are provided at the end.)
- For a $p \times p$ Latin square with rows ($\alpha_i$), columns ($\beta_k$) and treatments ($\tau_j$) fixed, obtain least squares estimators of $\alpha_i$, $\beta_k$ and $\tau_j$, i, j, k = 1, ... p. Derive the missing value formula (when just one observation is missing) for the Latin square design.