LAISHA
Open Laisha

ASRB Agricultural Research Service Bayesian Inference: practice questions

ASRB Agricultural Research Service (ARS) - Agricultural Statistics scientist · India · Agricultural Scientists Recruitment Board (ICAR)

Original practice questions written to the published syllabus · not past papers
10worked here, free
31written to this exam in this subject
10syllabus units

About ASRB Agricultural Research Service

ASRB Agricultural Research Service (ARS) - Agricultural Statistics scientist is a government recruitment exam run by Agricultural Scientists Recruitment Board (ICAR), for candidates in India. Combined NET/ARS/SMS/STO recruitment: preliminary (objective), mains (descriptive, discipline paper) and interview for ARS.

Official site: asrb.org.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 ASRB Agricultural Research Service

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 31 ASRB Agricultural Research Service questions in this subject: 4 easy, 25 medium, 2 hard, across 10 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 ASRB Agricultural Research Service and are not shown on any other exam's page.

Question 1 · easy

Weekly trap catches arrive one at a time. How can the posterior for the pest density be updated?

  1. Use each week's posterior as the next week's prior
  2. Wait until all data are in, since updating is invalid
  3. Average the weekly posteriors with equal weights
  4. Refit the prior from scratch each week, ignoring earlier data
Answer and working

A. Use each week's posterior as the next week's prior

With independent data, sequential updating gives the same final posterior as analysing all data at once. Averaging posteriors or discarding earlier data loses or distorts information.

Question 2 · medium

Why is the posterior predictive distribution of next season's yield wider than the plug-in distribution f(y | θ̂)?

  1. It uses a larger prior variance than the data
  2. It also includes uncertainty in θ
  3. It ignores the observed data entirely
  4. It double-counts the sampling variance
Answer and working

B. It also includes uncertainty in θ

p(y | x) = ∫ f(y | θ)π(θ | x)dθ mixes the sampling distribution over the posterior of θ, adding parameter uncertainty to the inherent variability. Plug-in predictions treat θ̂ as exact and are too narrow.

Question 3 · easy

A uniform prior is placed on a germination probability p. Which statement is correct?

  1. It is improper, because p has no upper bound
  2. It is Jeffreys' prior for the binomial model
  3. It is not uniform on the log-odds scale
  4. It is uniform on every transformation of p
Answer and working

C. It is not uniform on the log-odds scale

Flatness depends on the parameterisation; transforming p changes the density by the Jacobian. The uniform on (0, 1) is proper, and Jeffreys' prior is beta(1/2, 1/2), not uniform.

Question 4 · medium

Experts give a median and a 90% range for the yield response to a new practice. What does experience with elicited priors suggest?

  1. Elicited priors need no sensitivity analysis once fitted
  2. Experts tend to be under-confident, so elicited priors are narrowed
  3. Experts tend to be over-confident, so elicited priors are widened
  4. The expert median should replace the data in the analysis
Answer and working

C. Experts tend to be over-confident, so elicited priors are widened

Studies of elicitation show over-confidence: stated ranges are too narrow. Widening the prior and reporting sensitivity to it are standard practice.

Question 5 · medium

A breeder wants to know whether a new variety beats the check by at least 5%. Which Bayesian summary answers this most directly?

  1. The prior probability that the advantage exceeds 5%
  2. A Bayes factor for the point null of zero advantage
  3. The posterior mode of the advantage
  4. P(advantage > 5% | data)
Answer and working

D. P(advantage > 5% | data)

The posterior probability of the interval hypothesis is the direct answer and is robust to the prior choices that make point-null Bayes factors unstable.

Question 6 · medium

In a hierarchical model, what happens to the group estimates as the between-group variance τ² → 0, and as τ² → ∞?

  1. τ² → 0 gives complete pooling; τ² → ∞ gives no pooling
  2. Both limits give the overall mean for every group
  3. τ² → 0 gives no pooling; τ² → ∞ gives complete pooling
  4. Both limits give the same estimates as the group means
Answer and working

A. τ² → 0 gives complete pooling; τ² → ∞ gives no pooling

With B = σ²/(σ² + τ²): τ² → 0 makes B → 1 (all groups at μ); τ² → ∞ makes B → 0 (each group keeps its own estimate).

Question 7 · medium

A random-walk Metropolis sampler accepts 98% of proposals and its trace plot drifts slowly. What is the remedy?

  1. Thin the chain more heavily
  2. Increase the proposal step size
  3. Decrease the proposal step size
  4. Shorten the burn-in period
Answer and working

B. Increase the proposal step size

Near-universal acceptance means tiny steps that explore slowly; larger steps lower acceptance but speed exploration. Thinning does not improve mixing.

Question 8 · medium

Probit models for binary germination outcomes are often fitted by Gibbs sampling after introducing latent normal variables. What is this technique called?

  1. Importance sampling
  2. Data augmentation
  3. Thinning
  4. Rejection of burn-in
Answer and working

B. Data augmentation

Adding latent variables zᵢ with yᵢ = 1{zᵢ > 0} makes all full conditionals standard (truncated normal and normal), the Albert-Chib data-augmentation scheme.

Question 9 · hard

A hierarchical model with a small between-trial SD mixes badly, with a 'funnel' shape between the SD and the trial effects. What is the standard remedy?

  1. Removing the hierarchy and fitting trials separately
  2. Fixing the SD at its prior mean
  3. A non-centred parameterisation of the trial effects
  4. Thinning the chain by a factor of 100
Answer and working

C. A non-centred parameterisation of the trial effects

Writing θᵢ = μ + τηᵢ with ηᵢ ~ N(0, 1) breaks the strong dependence between τ and the θᵢ when τ is small. Fixing τ ignores its uncertainty.

Question 10 · medium

A model will be used to predict variety yields at new locations. How should cross-validation be organised?

  1. Leave out single plots at random
  2. Leave out the varieties with the lowest yields
  3. Use the same data for fitting and testing
  4. Leave out whole locations or trials at a time
Answer and working

D. Leave out whole locations or trials at a time

The validation must mimic the prediction task; leaving out single plots lets the model use information from the same location and overstates accuracy at new sites.

A study plan for Bayesian Inference in ASRB Agricultural Research Service

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 (45% of the 31 questions): Convergence diagnostics (4, 13%); Non-informative and weakly informative priors (4, 13%); Bayes' theorem for parameters: prior, likelihood and posterior (3, 10%); Bayesian hypothesis testing and Bayes factors (3, 10%).

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

  2. Then the middle band: Decision theory: loss functions and Bayes risk (3, 10%); Empirical Bayes (3, 10%); Hierarchical (multilevel) models (3, 10%); Posterior predictive checks and model comparison (DIC, WAIC) (3, 10%).

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

  3. Then the lighter units: Posterior summaries and credible intervals (3, 10%); Markov chain Monte Carlo: Gibbs sampling and Metropolis-Hastings (2, 6%).

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

  4. Leave time for the hard ones: Convergence diagnostics (1 hard of 4), Non-informative and weakly informative priors (1 hard of 4).

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

  5. Bank the easy marks: Non-informative and weakly informative priors (2 easy of 4); Bayes' theorem for parameters: prior, likelihood and posterior (1 easy of 3).

    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 — ASRB Agricultural Research Service 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.