(i) Give an estimate for the population mean under systematic sampling and derive its variance in terms of the intraclass correlation coefficient.
(ii) Discuss the effect of intraclass correlation on the efficiency of this estimator.
5 marks · directive: Discuss
The core of a good answer
Compared with SRSWOR, V(ȳ_ran) = [(N − n)/N](S²/n), systematic sampling is more efficient exactly when ρ_w < −1/(N − 1), equally efficient when ρ_w = −1/(N − 1) (as in a randomly ordered population), and less efficient when ρ_w is larger; its variance is 0 at ρ_w = −1/(n − 1) and largest when ρ_w = 1 (periodic variation with period k).
An outline written by upsc.help, not an official UPSC answer; every fact checked by a separate review. How we work
What the question asks
Use the variance formula from part (i), compare it with the SRSWOR variance for the same n, solve the inequality for ρ_w, and interpret the extreme and typical cases (linear trend, random order, periodicity).
Answer outline
Introduction
From (i): V(ȳ_sy) = [(N − 1)/N](S²/n)[1 + (n − 1)ρ_w]. SRSWOR: V(ȳ_ran) = [(N − n)/N](S²/n).
Step 1: Comparison
- V(ȳ_sy) < V(ȳ_ran) ⇔ (N − 1)[1 + (n − 1)ρ_w] < N − n.
- ⇔ (N − 1)(n − 1)ρ_w < −(n − 1) ⇔ ρ_w < −1/(N − 1).
Step 2: Range of ρ_w
- ρ_w = −1/(n − 1): all sample means equal Ȳ, V(ȳ_sy) = 0 (perfect).
- ρ_w = −1/(N − 1): same precision as SRSWOR.
- ρ_w = 1: units within a sample identical; V(ȳ_sy) = σ², as if only one unit were observed.
Step 3: Population types
- Random order: averaged over all random orderings of the list, V(ȳ_sy) equals the SRSWOR variance (checked by simulation in Python), so systematic ≈ SRSWOR.
- Linear trend: units in a sample spread across the range, ρ_w negative, so systematic sampling gains.
- Periodic variation with period equal to k (or a multiple): every unit in a sample is at the same phase, ρ_w near 1, and systematic sampling is very poor.
Step 4: Efficiency ratio
- E = V(ȳ_ran)/V(ȳ_sy) = (N − n)/{(N − 1)[1 + (n − 1)ρ_w]}.
Conclusion
The efficiency of systematic sampling is governed by ρ_w: it beats simple random sampling only if units within the same systematic sample are less alike than random pairs (ρ_w < −1/(N − 1)). Knowledge of trend or periodicity in the list is therefore essential before using it.
Examples and data to cite
- On a street with ten houses per block, taking every 10th house gives either only corner houses or none: periodicity matching the interval destroys the randomness of the sample.
Facts to remember
- V(ȳ_ran) = (1 − f)S²/n under SRSWOR.
- Systematic better than SRSWOR iff ρ_w < −1/(N − 1).
- Linear trend favours systematic sampling; periodicity with period k harms it.
- Design effect = 1 + (n − 1)ρ_w relative to σ²/n.