Welcome to CSIR NET Unit 4 — Statistics, Probability & Operations Research
Probability, statistical inference, linear models, multivariate analysis, sampling and design, reliability, linear programming and queues at the depth CSIR NET JRF demands: definitions with every hypothesis, theorems with proofs, live labs, and Part B and Part C problems with every option decided. This page explains how the course works and how to use it on your own or in a classroom.
1. Who this course is for
This course covers Unit 4 of the CSIR NET JRF Mathematical Sciences syllabus: Statistics, Probability and Operations Research. It is written for two kinds of reader at once.
🎓 Students
MSc students (mathematics or statistics) and final-year BSc students preparing for CSIR NET JRF, GATE ST or a university MSc examination and studying on their own. You should know BSc-level calculus (including double integrals and series) and matrices, and you should have taken an introductory probability course. Everything else is built from the ground up, with proofs.
🧑🏫 Teachers
College and university teachers who want a rigorous classroom text. Every section has a lesson plan, discussion questions, the misconceptions to watch for, a suggested homework set, and two interactive labs to project in class.
This is not a summary or a crash course. Unit 4 questions in Part C reward exactly what a summary leaves out: the exact conditions under which a law of large numbers holds, the reason a uniformly most powerful test exists for a one-sided alternative but not for a two-sided one, the difference between pairwise and mutual independence, and the counterexample showing that zero correlation does not imply independence. So every definition here is stated precisely, every central theorem is proved, and the standard counterexamples of each topic are collected in a ⚠️ box. Unit 4 is also the most computational unit of the paper. Many questions need a number (a posterior probability, a variance, an MLE, an expected queue length), so every section works its computations out in full, and every number on these pages has been checked by computer.
2. Two ways to use the course
- Read with a pencil. Copy each definition and theorem into a notebook. Then write down one example and one non-example of your own before you read on.
- Derive before you read the derivation. Give yourself five minutes with each theorem or formula (a mean, a variance, a posterior, a test statistic). Then compare your attempt with the proof given and note the idea you missed.
- Cover the solutions. Attempt every worked example before opening "💡 Show solution". If you are stuck, open one step at a time.
- Play with the labs. Predict what will happen before you press Play. Every simulation uses a fixed random seed, so you and your classmates see the same runs.
- Do the exercises, then the Exam corner. Do at least the homework set suggested in the 🧑🏫 box. For every Part C problem, decide every option and write a one-line reason or counterexample for each. That is the skill the exam tests.
- Revisit after a week. Take the five-question quiz again and redo two Exam-corner problems without notes.
Each section is planned for two or three lectures of about an hour. The 🧑🏫 For the teacher box at the end of each section says which results to prove on the board, which examples to work through, when to project which lab, and which Exam-corner problems to do live. The labs work on a projector or a phone. They need no installation, and once the page has loaded they need no internet. Worked solutions are hidden by default, so you can project a page while students attempt a problem. The exercises are solved in full: use them for tutorials, and use the Exam corner for class tests. The Research–Teaching Nexus at the end of each section is a small project with real data or a simulation, suitable for a weekly assignment.
3. What every section contains
- Motivation: why the idea matters, where it is used, and how it is examined.
- Teaching in coloured boxes: 📘 definitions; 📐 theorems stated with all their hypotheses, each followed by a Proof; 🧭 problem-solving strategies; and a ⚠️ Counterexamples and pitfalls box listing the standard traps.
- ✏️ Worked examples (five to seven per section). They mix computations with "prove or disprove" problems, and each has a full step-by-step solution.
- Two interactive labs that show the idea in motion: a histogram reacting to an outlier, a Bayes tree, sample means turning bell-shaped, a Markov chain settling down, a likelihood peaking at its maximum, a queue growing and draining.
- Exercises: 25 to 40 parts, from routine to research-flavoured, every one solved.
- 🎯 Exam corner: six original problems. E1–E3 are in the style of CSIR Part B (one correct option) and E4–E6 in the style of Part C (one or more correct options), with every option decided. Problems that also suit GATE are tagged GATE ST.
- 🧑🏫 For the teacher, a 🔬 Research–Teaching Nexus mini-project, and a 🧪 five-question quiz.
Words with a dotted underline are in the Glossary: hover over them or tap them to see the meaning. Every key formula is collected in the searchable Formula Sheet, which is useful for revision the week before the exam.
4. The CSIR NET examination in brief
The Joint CSIR–UGC NET in Mathematical Sciences is a single computer-based paper of multiple-choice questions in three parts:
- Part A: general aptitude (logical reasoning, graphical analysis, numerical ability and puzzles). It is common to all subjects and is not covered by this course, although its data-interpretation questions use the descriptive statistics of §1.1.
- Part B: mathematics, with exactly one correct option per question. Each question tests one idea quickly: a computation, a definition or a standard result.
- Part C: mathematics at greater depth. One or more of the options may be correct, and you must choose all the correct ones. A single overlooked counterexample costs the whole question, so Part C rewards precise knowledge of hypotheses.
In Parts B and C the questions are spread over the four units of the syllabus, and candidates answer a prescribed number of the questions offered. Unit 4 questions are usually offered as a block that a candidate may choose to answer. The number of questions, the marks per question, the duration and the negative-marking rules have changed over the years and differ between the parts, so always check the latest official notification for your exam session. Many of the problems in this course also suit GATE Statistics (GATE ST) and are tagged accordingly.
Treat each option as a separate true/false question. For an option you believe is true, know the theorem that proves it, together with its hypotheses. For an option you believe is false, have a specific counterexample ready. For example: "two fair coins: the events first is heads, second is heads and both agree are pairwise but not mutually independent"; "$X\sim N(0,1)$ and $Y=X^2$ are uncorrelated but not independent"; "the Cauchy distribution has no mean, so the sample mean does not settle down"; "for $U(0,\theta)$ the MLE $\max X_i$ is biased". The ⚠️ box in each section is designed to be learned as exactly such a list.
5. A first taste: one statistic, nine chapters
Before any theory, here is how the chapters of this course connect. Take a population, draw a random sample of size $n$, and compute the sample mean $\bar X$. Repeat many times. The lab below collects the sample means. However skewed or discrete the population is, their histogram becomes bell-shaped, centred at the population mean $\mu$, with spread $\sigma/\sqrt n$. Almost every chapter of the course explains a different side of this one picture.
Chapter 1: data and probability
A histogram summarises the data (§1.1). $E[\bar X]=\mu$ and $\operatorname{Var}(\bar X)=\sigma^2/n$ follow from linearity of expectation and independence (§§1.3–1.4).
Chapter 2: limit theorems
The laws of large numbers say that $\bar X\to\mu$. The central limit theorem says that $\sqrt n(\bar X-\mu)/\sigma$ is approximately $N(0,1)$. Characteristic functions prove the weak law and the central limit theorem.
Chapter 3: Markov chains
For a Markov chain, the long-run fraction of time spent in a state is also a sample mean. The ergodic theorem replaces independence by mixing.
Chapter 4: estimation
$\bar X$ is the MLE of an exponential mean and the UMVUE of a normal mean, and $\bar X\pm1.96\,\sigma/\sqrt n$ is a 95% confidence interval. The coverage readout above shows what "95%" means.
Chapter 5: testing
The $z$-test rejects $H_0:\mu=\mu_0$ when $|\bar X-\mu_0|$ exceeds $1.96\,\sigma/\sqrt n$. Its one-sided version, which rejects when $\bar X-\mu_0$ exceeds $1.645\,\sigma/\sqrt n$, is most powerful against every $\mu\gt\mu_0$ by the Neyman–Pearson lemma.
Chapters 6–7: linear models
$\bar X$ is the least-squares estimate in the model $X_i=\mu+\varepsilon_i$ and the BLUE of $\mu$ (Gauss–Markov). In $p$ dimensions, the sample mean vector is $N_p(\mu,\Sigma/n)$.
Chapter 8: surveys
Without replacement from a finite population, $\operatorname{Var}(\bar y)=(1-f)S^2/n$. Stratification can shrink it further.
Chapter 9: queues
Exponential inter-arrival and service times make the M/M/1 queue Markovian. Little's law $L=\lambda W$ links two long-run averages.
6. Map of the course
The nine chapters follow the official Unit 4 syllabus. Chapters 1–3 cover probability and stochastic processes, Chapters 4–7 statistical inference, linear models and multivariate analysis, Chapter 8 sample surveys and the design of experiments, and Chapter 9 reliability and operations research. Each chapter page opens with a short overview, the learning outcomes, a teaser lab and the list of its sections.
Probability
Descriptive statistics and EDA; probability spaces, conditioning, Bayes; random variables, moments; joint and conditional distributions.
Distributions and Limit Theorems
The standard distributions; characteristic functions; Markov, Chebyshev and Jensen; modes of convergence, laws of large numbers, CLT.
Markov Chains and Stochastic Processes
Classification of states, stationary distributions, limiting behaviour; Poisson and birth–death processes.
Sampling Distributions and Estimation
$\chi^2$, $t$, $F$, order statistics; MLE, sufficiency, completeness, UMVUE, Cramér–Rao; confidence intervals and Bayes estimators.
Testing of Hypotheses
Neyman–Pearson, UMP and likelihood ratio tests; large-sample and chi-square tests; nonparametric tests and rank correlation.
Linear Models and Regression
Gauss–Markov theorem and estimability; ANOVA and ANCOVA, fixed, random and mixed effects; regression diagnostics, logistic regression.
Multivariate Analysis
Multivariate normal and Wishart; partial and multiple correlation, Hotelling's $T^2$; PCA, discriminant, cluster and canonical correlation analysis.
Sample Surveys and Design of Experiments
SRS, stratified, systematic and PPS sampling, ratio and regression estimators; CRD, RBD, LSD, BIBD, $2^k$ factorials and confounding.
Reliability, Linear Programming and Queues
Hazard rates, censoring, series and parallel systems; simplex method and duality; queueing and inventory models.
The other three units of the syllabus are covered by the sister courses CSIR NET Unit 1: Analysis and Linear Algebra (which supplies the prerequisites: series, integrals, Lebesgue measure, matrices, eigenvalues and quadratic forms), CSIR NET Unit 2: Complex Analysis, Algebra and Topology, and CSIR NET Unit 3: Differential Equations, Numerical Analysis, Calculus of Variations, Integral Equations and Classical Mechanics.
7. Conventions
- $\mathbb{N}=\{1,2,3,\dots\}$ does not contain $0$. $\log x$ is the natural logarithm.
- $P(\cdot)$, $E[\cdot]$, $\operatorname{Var}(\cdot)$ and $\operatorname{Cov}(\cdot,\cdot)$ denote probability, expectation, variance and covariance. Random variables are written in capitals ($X,Y,\bar X,S^2$) and observed values in lower case ($x_i,\bar x$). "iid" means independent and identically distributed.
- $\Phi$ is the standard normal distribution function and $z_\alpha$ is its upper $\alpha$ point, so that $1-\Phi(z_\alpha)=\alpha$ (for example $z_{0.025}=1.96$). Every quantile quoted in the course has been checked by computer.
- For a data set, $s^2=\frac1n\sum(x_i-\bar x)^2$ uses the divisor $n$ (§1.1). The sample variance $S^2=\frac1{n-1}\sum(X_i-\bar X)^2$, which uses the divisor $n-1$, appears from Chapter 4 on. Always check which divisor a question intends.
- Estimators carry hats: $\hat\theta$, $\hat\beta$. Parametrisations are stated every time: for example, the exponential distribution with rate $\lambda$ has mean $1/\lambda$.
- "Iff" means "if and only if". A theorem stated with "if" claims one direction only. The converse is often false, and Part C questions exploit this.
- Definitions and theorems are numbered by chapter (Definition 1.4, Theorem 1.12). Results in other sections are referred to by section number and name, e.g. "Bayes' theorem (§1.2)". Proofs end with ∎.
Where to begin. Start with Chapter 1: Probability. Its first section, on descriptive statistics, needs nothing beyond school algebra. Its second section builds the axioms of probability on which the whole course rests.