UPSC Statistics 2024 — Paper II

The directive 'describe' demands systematic exposition of procedures and concepts across all five sub-parts. Allocate approximately 15-18 minutes to part (a) requiring calculation, 12-15 minutes each to descriptive parts (b), (c), and (e), and 10-12 minutes to part (d) on dual simplex. Structure as: direct calculation for (a); stepwise procedural description for (b) with industrial applications; precise definitions with formulas for (c); algorithmic presentation for (d); and conceptual definition followed by domain-specific applications for (e).

• Part (a): Correct application of parallel system reliability formula R_system = 1 - (1 - 0.90)³ = 1 - (0.10)³ = 0.999 or 99.9%
• Part (b): CUSUM chart construction using V-mask or decision interval, cumulative sum calculation S_i = Σ(x_j - μ₀), and applications in pharmaceutical quality control or ISRO component manufacturing
• Part (c)(i): Producer's risk (α) as probability of rejecting good lot (AQL quality), typically set at 5% with Type I error interpretation
• Part (c)(ii): AOQL as maximum average outgoing quality after rectifying inspection, formula AOQ = p·P_a·(N-n)/N for sampling with replacement
• Part (d): Dual simplex form with primal maximization converting to dual minimization, conditions for optimality (all c_j - z_j ≤ 0), and iterative steps for pivot selection when b_i < 0
• Part (e): Monte Carlo as stochastic simulation using random number generation, with applications in nuclear shielding design, financial risk modeling, or Indian monsoon prediction

Begin by deriving control limits for X̄-chart and R-chart using standard formulae with A₂, D₃, D₄ constants, explaining why joint monitoring prevents Type I/II errors in SPC (40% time). For Weibull, derive R(t) = exp[-(t/θ)^β] and h(t) = (β/θ)(t/θ)^(β-1), interpreting bathtub curve relevance for Indian manufacturing/equipment reliability (20% time). For the LPP, convert ≥ constraints to standard form using surplus variables and artificial variables (Big-M method), then execute simplex iterations to reach optimality (40% time).

• Part (a): Correct formulae for X̄-chart limits (UCL/LCL = X̄̄ ± A₂R̄) and R-chart limits (UCL = D₄R̄, LCL = D₃R̄) with proper identification of constants
• Part (a): Explanation that X̄-chart monitors process mean while R-chart monitors process variability; joint study detects both shifts and dispersion changes, preventing misinterpretation from exclusive use of either chart
• Part (b): Derivation of reliability function R(t) = exp[-(t/θ)^β] and hazard function h(t) = (β/θ)(t/θ)^(β-1) from Weibull PDF
• Part (b): Interpretation of β < 1 (decreasing hazard, infant mortality), β = 1 (constant hazard, exponential), β > 1 (increasing hazard, wear-out) with industrial examples
• Part (c): Conversion of ≥ constraints to standard form: subtract surplus variables s₁, s₂, s₃ and add artificial variables A₁, A₂, A₃ with Big-M penalty in objective
• Part (c): Complete simplex tableau iterations showing entering/leaving variables, pivot operations, and final optimal solution with Z_max = 12 at (x₁=3, x₂=0) or verified corner point
• Part (c): Verification of solution by checking constraint satisfaction and comparing objective values at all extreme points

Begin with clear definitions for (a)(i)-(iii) on LPP concepts, allocating ~30% time; for (b) define TPM, establish regularity/ergodicity criteria, then compute steady-state probabilities using eigenvalue or algebraic methods (~35% time); for (c) derive failure rate from R(t), analyze its monotonicity, and integrate for MTTF (~35% time). Structure: definitions → theorems → computational steps → physical interpretation.

• (a)(i) Extreme Point Solutions: Definition as feasible region vertices, convex combination property, and Fundamental Theorem of LPP optimality at extreme points
• (a)(ii) Duality Theorem: Statement of weak and strong duality, primal-dual relationship, and economic interpretation of shadow prices
• (a)(iii) Complementary Slackness: Conditions relating primal slack and dual variables, optimality verification tool
• (b) TPM definition with row-stochastic property; regularity (some P^n has all positive entries) vs ergodicity (irreducible + aperiodic); classification of given P as regular and ergodic; computation of limiting distribution π = (5/9, 4/9)
• (c)(i) Failure rate λ(t) = 2/(t₀-t) for 0 ≤ t < t₀ using λ(t) = -R'(t)/R(t)
• (c)(ii) Increasing failure rate (IFR) demonstration as λ'(t) > 0, indicating wear-out phase
• (c)(iii) MTTF = t₀/3 via integration ∫₀^t₀ R(t)dt, with proper handling of improper integral at t₀

Solve this multi-part numerical and theoretical question by allocating approximately 35% time to part (a) on M/M/2 queueing, 30% to part (b) on production inventory EOQ model, and 35% to part (c) on SQC theory. Begin with clear identification of model parameters for each part, show all formulas with standard notation, perform step-by-step calculations with proper unit conversions, and conclude with precise numerical answers and brief interpretations where asked.

• For (a): Correct identification of M/M/2 queue parameters (λ=100/hr, μ=60/hr, c=2), calculation of traffic intensity ρ=λ/(cμ)=0.833, and use of multi-server queue formulas for P₀, Pₙ, and waiting time distribution
• For (a)(i): Computation of P(n>3) = 1 - P₀ - P₁ - P₂ - P₃ using the formula Pₙ = (1/n!)(λ/μ)ⁿP₀ for n≤c and Pₙ = (1/c!cⁿ⁻ᶜ)(λ/μ)ⁿP₀ for n>c
• For (a)(ii): Probability a given teller is idle = P₀ + ½P₁ (or equivalently 1 - ρ), recognizing that idle probability per server differs from system idle probability
• For (b): Identification of production model parameters (p=300/day, d=150/day, D=37,500/yr, C₁=$0.25/yr, C₃=$200), and application of EPQ formulas rather than simple EOQ
• For (b)(i)-(iv): Correct formulas for EOQ/EPQ = √[2DC₃p/(C₁(p-d))], production run length = Q/p, number of runs = D/Q, and maximum inventory = Q(p-d)/p with proper unit consistency
• For (c)(i): Clear distinction between chance causes (random, inherent, unavoidable) and assignable causes (special, identifiable, removable), with 4-5 specific advantages of SQC such as early detection, reduced inspection costs, and customer satisfaction
• For (c)(ii): Systematic description of OC curve construction: define p (lot fraction defective), calculate Pₐ using binomial/Poisson approximation, plot Pₐ vs p, and identify key points (AQL, LTPD, α, β)

This multi-part question requires balanced coverage across five thematic areas. Allocate approximately 20% time each to parts (a), (c), and (e) which demand descriptive-discursive treatment, and 20% combined to parts (b) and (d) which require precise calculations. Begin with a brief roadmap indicating coverage of all sub-parts, then proceed sequentially: Indian statistical infrastructure → AR(1) forecasting with proper formula application → SLSE theoretical exposition → life table computations with force of mortality derivation → psychometric concepts with Binet-Simon/Wechsler references. Conclude each calculation part with interpreted results in context.

• (a) Indian Statistical System: Evolution from PC Mahalanobis era; distinction between Central and State statistical machinery; NSO's role under Ministry of Statistics and Programme Implementation (MoSPI); key organizations—CSO, NSSO, Registrar General of India, NITI Aayog; NSO's functioning through data collection, coordination, and dissemination under National Statistical Commission oversight
• (b) AR(1) forecasting: Correct model specification X_t = μ + φ(X_{t-1} - μ) + ε_t; one-step ahead forecast = μ + φ(x_n - μ) = 74.3293 + 0.5705(67 - 74.3293); five-step ahead forecast converges to mean μ as φ^5 → 0; explicit numerical computation
• (c) SLSE model: Structural form vs reduced form distinction; endogeneity problem causing correlation between regressors and error terms; simultaneous equation bias; why OLS is inconsistent (covariance between Y and u non-zero); need for IV/2SLS methods
• (d)(i) Average age at death: n_a_x = (∫_0^n (x+t)μ(x+t)l(x+t)dt)/(l(x)-l(x+n)) or equivalent life table expression; (d)(ii) Force of mortality μ(x) = -d[ln l(x)]/dx; exact derivation for l(x)=100√(100-x) yielding μ(84)=1/32
• (e) Intelligence tests: Definition as standardized measures of cognitive ability; types—verbal, performance, group vs individual; uses in education, clinical diagnosis, occupational selection; mental age (MA) as performance level relative to age norms; IQ formulas—Stern's ratio IQ (MA/CA×100) and deviation IQ; reference to Indian adaptations

Begin by deriving OLS and GLS estimators using matrix notation for part (a), allocating approximately 40% of time/space given its 20 marks. For part (b), explain the three index number tests conceptually before applying the Time Reversal Test to Laspeyres' formula with the given data (~30%). For part (c), clearly distinguish estimates from projections and outline the component method with fertility, mortality, and migration components (~30%). Conclude with a brief synthesis of how these statistical tools inform policy-making in India.

• For (a): Derivation of β̂_OLS = (X'X)^(-1)X'Y and β̂_GLS = (X'Ω^(-1)X)^(-1)X'Ω^(-1)Y with proper assumptions
• For (a): Comparison of BLUE properties under homoscedasticity vs. heteroscedasticity/autocorrelation; efficiency of GLS over OLS when Ω ≠ σ²I
• For (b): Clear statement of Time Reversal (P01 × P10 = 1), Factor Reversal (P01 × Q01 = Σp1q1/Σp0q0), and Circular (P01 × P12 × P20 = 1) tests
• For (b): Numerical verification showing Laspeyres' index fails Time Reversal Test with explicit calculation
• For (c): Distinction between estimates (current/retrospective) and projections (future-oriented, assumption-dependent)
• For (c): Component method: projection of births by age-specific fertility, deaths by life tables, migration by net migration rates

Explain the identification problem with a concrete supply-demand example, then apply order and rank conditions to the given 3-equation system. For (b), explain why CDR fails using Indian state examples (e.g., Kerala vs. Uttar Pradesh age structures), then detail direct and indirect standardization methods. For (c), define reliability via Spearman-Brown prophecy, discuss test length effects, and enumerate split-half, KR-20, KR-21, and Cronbach's alpha methods. Allocate approximately 30% time to (a), 40% to (b), 20% to (c)(i), and 10% to (c)(ii) based on marks distribution.

• (a) Clear explanation of identification problem using supply-demand simultaneous equations example; distinction between under-identified, just-identified, and over-identified equations
• (a) Correct application of order condition (K - k ≥ m - 1) and rank condition to all three equations; accurate classification of each equation's identifiability status
• (b) Explanation of CDR limitations due to varying age-sex compositions; use of Indian demographic examples (e.g., aging Kerala vs. younger Bihar populations)
• (b) Construction of standardized death rates: direct method (standard population weights) and indirect method (standard mortality rates); formula for Comparative Mortality Index (CMI) and its interpretation
• (c)(i) Formal definition of reliability as ratio of true score variance to observed score variance; statement and explanation of Spearman-Brown prophecy formula for test length effects
• (c)(ii) Comprehensive coverage of reliability estimation methods: test-retest, parallel forms, split-half (including Spearman-Brown correction), Kuder-Richardson formulas (KR-20, KR-21), and Cronbach's coefficient alpha with appropriate formulas

Begin with part (a) by converting percentages to z-scores using standard normal tables, then apply linear interpolation for equispaced difficulty; allocate ~30% time here. For part (b), define each fertility measure sequentially showing progressive refinement from crude to age-specific rates, then explain TFR's utility for population projection—spend ~30% time. For part (c), discuss autocorrelation consequences on OLS properties (BLUE violation), then detail Durbin-Watson test procedure with critical values—allocate ~40% time as this carries highest marks. Conclude with integrated insights on statistical applications in demographic and econometric analysis.

• Part (a): Convert 85% and 25% to z-scores (-1.036 and 0.674), establish equidistant points on difficulty scale, calculate intermediate z-values, convert back to percentages (~58% and ~42%)
• Part (b): CBR definition and limitation (ignores age-sex structure); GFR improvement (restricts to women 15-49); ASFR refinement (age-specific exposure); TFR as sum of ASFRs and utility for replacement-level fertility analysis (India's TFR ~2.0)
• Part (c)(i): Autocorrelation causes (inertia, specification error, cobweb, data manipulation); consequences: OLS estimators remain unbiased but inefficient, standard errors biased, t/F tests invalid, R² misleading
• Part (c)(ii): DW test assumptions (no lagged dependent variable, intercept, non-stochastic regressors); test statistic formula; decision zones (0 to 4 scale); inconclusive region problem; alternative tests (Durbin's h, Breusch-Godfrey)