LAISHA
Open Laisha

CSIR-UGC NET Bayesian Inference: practice questions

CSIR-UGC NET - Mathematical Sciences (statistics option) · India · National Testing Agency for CSIR

Original practice questions written to the published syllabus · not past papers
10worked here, free
63written to this exam in this subject
11syllabus units

About CSIR-UGC NET

CSIR-UGC NET - Mathematical Sciences (statistics option) is a PhD, fellowship and lectureship exam run by National Testing Agency for CSIR, for candidates in India. Computer-based, 3 hours, Parts A (general aptitude), B and C (subject); statisticians answer from the statistics unit.

Official site: csirnet.nta.ac.in.

We have not been able to confirm these details against the exam body's own specification, so fees, dates and pass marks are left out on purpose. Check the current format, eligibility, fees and dates on the official site.

Bayesian Inference in CSIR-UGC NET

Inference through prior and posterior: conjugate analysis, prior choice, credible intervals, Bayes factors and decision theory, hierarchical models, and computation by Markov-chain Monte Carlo, with agricultural uses in variety trials, genomic prediction and small-area estimation.

Our 63 CSIR-UGC NET questions in this subject: 7 easy, 29 medium, 27 hard, across 11 units.

10 practice questions, worked

Choose an answer, then open the working. Each explanation says why the right answer is right and why the others are not.

These 10 were written to CSIR-UGC NET and are not shown on any other exam's page.

Question 1 · hard

What does the Bernstein-von Mises theorem state, under regularity conditions?

  1. The posterior mean equals the MLE exactly for every sample size
  2. The posterior is unaffected by the prior whatever the sample size
  3. It tends to a normal centred at the MLE with variance 1/[nI]
  4. Credible intervals have exact frequentist coverage in small samples
Answer and working

C. It tends to a normal centred at the MLE with variance 1/[nI]

As n grows, a smooth prior with positive density at the true value is swamped by the likelihood, and the posterior approaches N(θ̂, 1/[nI(θ̂)]). In small samples the prior still matters and coverage is not exact.

Question 2 · medium

Through which quantity does the posterior depend on the data?

  1. The sample mean
  2. A sufficient statistic
  3. The p-value of the data
  4. The sample size alone
Answer and working

B. A sufficient statistic

If T is sufficient, f(x | θ) = g(T(x); θ)h(x), and h cancels in the normalisation, so π(θ | x) depends on x only through T.

Question 3 · medium

Counts are Poisson(λ) with a gamma prior on λ. What is the posterior predictive distribution of a new count?

  1. Poisson with the posterior mean
  2. Negative binomial
  3. Beta-binomial
  4. Gamma
Answer and working

B. Negative binomial

Mixing a Poisson over a gamma distribution for its mean gives a negative binomial; the extra variance reflects uncertainty about λ. Beta-binomial is the predictive for binomial data with a beta prior.

Question 4 · hard

What is Jeffreys' prior for the mean λ of a Poisson distribution?

  1. π(λ) ∝ 1/λ
  2. π(λ) ∝ 1
  3. π(λ) ∝ λ
  4. π(λ) ∝ λ^(−1/2)
Answer and working

D. π(λ) ∝ λ^(−1/2)

I(λ) = 1/λ, so √I = λ^(−1/2). 1/λ is the scale-invariant prior used for a standard deviation (and for a Poisson rate by some authors), not Jeffreys' rule here.

Question 5 · hard

For a strongly right-skewed posterior of a variance component, which statement is correct?

  1. Mean, median and mode differ, and the median is a stable summary
  2. The HPD interval and the equal-tailed interval always coincide
  3. Mean, median and mode always coincide for variance parameters
  4. The mode is always largest, so it is the conservative summary to report
Answer and working

A. Mean, median and mode differ, and the median is a stable summary

For skewed posteriors mode < median < mean typically, sometimes by a lot; the mode can sit at zero for a variance. HPD and equal-tailed intervals differ when the posterior is skewed.

Question 6 · easy

For two simple hypotheses θ = θ₀ and θ = θ₁, what is the Bayes factor B₁₀?

  1. The ratio of posterior probabilities
  2. The p-value of the likelihood-ratio test
  3. The ratio of prior probabilities P(H₁)/P(H₀)
  4. The likelihood ratio f(x | θ₁)/f(x | θ₀)
Answer and working

D. The likelihood ratio f(x | θ₁)/f(x | θ₀)

With no free parameters, each marginal likelihood is just the likelihood at its hypothesised value, so B₁₀ is the likelihood ratio. Posterior odds also involve the prior odds.

Question 7 · medium

For random-walk Metropolis in high dimensions, what acceptance rate does theory suggest is efficient?

  1. About 0.234
  2. Exactly 0.5
  3. About 0.95
  4. About 0.01
Answer and working

A. About 0.234

Roberts, Gelman and Gilks (1997) derived 0.234 as optimal in high dimensions; for one-dimensional updates about 0.44 is suggested. Very high rates mean steps too small; very low, steps too large.

Question 8 · medium

MCMC output gives a posterior mean deviance of 250 and a deviance of 242 at the posterior mean. What are p_D and DIC?

  1. p_D = 8, DIC = 242
  2. p_D = 8, DIC = 250
  3. p_D = 8, DIC = 258
  4. p_D = 492, DIC = 742
Answer and working

C. p_D = 8, DIC = 258

p_D = D̄ − D(θ̄) = 8; DIC = D̄ + p_D = 258 (equivalently D(θ̄) + 2p_D). 250 omits the penalty; 242 is the deviance at the posterior mean alone.

Question 9 · medium

How do Bayes factors and predictive criteria such as WAIC differ in what they compare?

  1. Bayes factors compare predictive accuracy; WAIC compares marginal likelihoods
  2. Both compare predictive accuracy, with different priors
  3. Both compare marginal likelihoods, with different penalties
  4. Bayes factors compare marginal likelihoods; WAIC compares expected predictive accuracy
Answer and working

D. Bayes factors compare marginal likelihoods; WAIC compares expected predictive accuracy

Bayes factors depend on the whole prior (and can be sensitive to it); WAIC and LOO measure how well the posterior predicts new data. They can rank models differently.

Question 10 · hard

Five independent normal means with unit variance are estimated, and ‖x‖² = 12. What shrinkage factor does the James-Stein estimator [1 − (p − 2)/‖x‖²]x apply?

  1. 0.25
  2. 0.583
  3. 0.75
  4. 1.00
Answer and working

C. 0.75

1 − (5 − 2)/12 = 1 − 0.25 = 0.75: each coordinate is pulled 25% towards zero. 0.25 is the amount of shrinkage, not the factor.

A study plan for Bayesian Inference in CSIR-UGC NET

This plan follows where our questions actually fall, so the time goes where the marks are likely to be. It is our reading of the syllabus, not the exam body's weighting.

  1. First, the heavy units (43% of the 63 questions): Markov chain Monte Carlo: Gibbs sampling and Metropolis-Hastings (7, 11%); Non-informative and weakly informative priors (7, 11%); Posterior predictive checks and model comparison (DIC, WAIC) (7, 11%); Bayesian hypothesis testing and Bayes factors (6, 10%).

    Read the manual chapters for these first and do every worked question you can find on them.

  2. Then the middle band: Conjugate families (6, 10%); Convergence diagnostics (6, 10%); Empirical Bayes (6, 10%); Bayes' theorem for parameters: prior, likelihood and posterior (5, 8%).

    Aim for one sitting per unit, then mixed questions across them.

  3. Then the lighter units: Decision theory: loss functions and Bayes risk (5, 8%); Hierarchical (multilevel) models (5, 8%); Posterior summaries and credible intervals (3, 5%).

    Learn the definitions, classifications and standard numbers; these units are more often recalled than worked.

  4. Leave time for the hard ones: Non-informative and weakly informative priors (5 hard of 7), Empirical Bayes (4 hard of 6), Markov chain Monte Carlo: Gibbs sampling and Metropolis-Hastings (4 hard of 7).

    Come back to these after a first pass, untimed, and read every explanation, including for the options you rejected.

  5. Bank the easy marks: Bayesian hypothesis testing and Bayes factors (2 easy of 6); Bayes' theorem for parameters: prior, likelihood and posterior (1 easy of 5).

    Do these early in revision so they are fast on the day.

  6. Finish timed. Sit the subject as a timed set without notes, mark it, and send each wrong answer back to its unit above.

Last updated · How Laisha writes and checks its pages · Report a mistake

Ask Laisha about this

Opens Laisha's chat with your question ready to send. Sign in or start free.

Unlock every question — CSIR-UGC NET Bayesian Inference

US$10 a month · one price worldwide · cancel any time

  • Every practice question in this subject, with full working
  • Timed exam papers and progress by unit
  • Ask Laisha to explain any answer — up to 1,000 questions a month
Free or Smarter
FreeSmarter
Grow guides, pest pages, articlesRead allRead all
Ask Laisha5 a dayUp to 1,000 a month
Exam practiceThe sample on each pageEvery question, timed papers
Crop calendar and pest alerts—For your own fields
Calculators and tools—All
Uploads and reports—100 a month, real files
10 days free for every new account, no card.