Skip to content

Sampling and Design of Experiments Formula Sheet and Short Notes

One-page CSIR NET revision: SRS, stratified, systematic, PPS, ratio and regression estimators, CRD, RBD, Latin square, BIBD and 2^k confounding.

  • 3 min read
  • Updated
On this page · 9 sections
  1. The paper in one table
  2. Simple random sampling
  3. Stratified sampling
  4. Systematic sampling (\( N = nk \))
  5. PPS, ratio and regression
  6. CRD, RBD, Latin square
  7. Block designs and BIBD
  8. \( 2^k \) factorials and confounding
  9. Part C checklist

Every formula from Simple Random, Stratified and Systematic Sampling, PPS Sampling, Ratio and Regression Estimators, CRD, RBD and Latin Square Designs, Block Designs: Connectedness, Orthogonality and BIBD and 2^k Factorial Experiments and Confounding, on one page.

The paper in one table

Per the NTA information bulletin for December 2026 (Mathematical Sciences, code 704):

PartGivenAttemptCorrectWrongMaximum
B4025+3−0.7575
C6020+4.75095

Part C: one or more options may be correct; marks only if you mark all correct options and no wrong one. No partial credit, no negative marking. Details on the Mathematical Sciences overview.

Simple random sampling

SRSWORSRSWR
Samples\( \binom{N}{n} \)\( N^n \) ordered
\( V(\bar y) \)\( (1 - f)\frac{S^2}{n} \)\( \frac{\sigma^2}{n} = \frac{N-1}{N}\frac{S^2}{n} \)
\( E(s^2) \)\( S^2 \)\( \sigma^2 \)
Inclusion\( \pi_i = \frac{n}{N} \), \( \pi_{ij} = \frac{n(n-1)}{N(N-1)} \)\( P(i \text{ appears}) = 1 - (1 - \frac{1}{N})^n \)
  • \( \frac{V_{WR}}{V_{WOR}} = \frac{N - 1}{N - n} \). Proportion: \( V(p) = \frac{N - n}{N - 1}\frac{PQ}{n} \).
  • Sample size: \( n_0 = \frac{z^2S^2}{d^2} \) (or \( \frac{z^2PQ}{d^2} \)), \( n = \frac{n_0}{1 + n_0/N} \), round up.
  • Benchmark: \( \{2, 4, 6, 8, 10\} \), \( n = 2 \): \( S^2 = 10 \), \( V_{WOR} = 3 \), \( V_{WR} = 4 \).

Stratified sampling

$$\bar y_{st} = \sum W_h\bar y_h, \qquad V(\bar y_{st}) = \sum W_h^2(1 - f_h)\frac{S_h^2}{n_h}$$
Allocation\( n_h \propto \)Variance
Proportional\( N_h \)\( \frac{1 - f}{n}\sum W_hS_h^2 \)
Neyman\( N_hS_h \)\( \frac{(\sum W_hS_h)^2}{n} - \frac{\sum W_hS_h^2}{N} \)
Cost-optimum\( \frac{N_hS_h}{\sqrt{c_h}} \)for fixed budget, \( n = (C - c_0)\frac{\sum N_hS_h/\sqrt{c_h}}{\sum N_hS_h\sqrt{c_h}} \)
  • \( V_{ran} - V_{prop} \approx \frac{1 - f}{n}\sum W_h(\bar Y_h - \bar Y)^2 \) (means); \( V_{prop} - V_{opt} = \frac{1}{n}\sum W_h(S_h - \bar S)^2 \) (SDs).
  • Neyman asks \( n_h > N_h \): take that stratum completely, re-allocate the rest.

Systematic sampling (\( N = nk \))

  • \( k \) possible samples; \( \pi_i = \frac{1}{k} \); many \( \pi_{ij} = 0 \), so no unbiased variance estimator from one sample.
  • \( V(\bar y_{sy}) = \frac{N-1}{N}\frac{S^2}{n}[1 + (n - 1)\rho_w] \); \( V_{sy} - V_{ran} = \frac{n-1}{n}(S^2 - S^2_{wsy}) \).
  • Better than SRS iff \( S^2_{wsy} > S^2 \). Linear trend \( Y_i = i \): \( V_{ran} = \frac{(k-1)(nk+1)}{12} \), \( V_{sy} = \frac{k^2-1}{12} \), \( V_{st} = \frac{k^2-1}{12n} \). Period \( k \): worst case.
  • \( N \ne nk \): circular systematic sampling (start in \( 1..N \), wrap round).

PPS, ratio and regression

EstimatorFormulaVariance / MSE
Hansen-Hurwitz (PPSWR)\( \frac{1}{n}\sum\frac{y_r}{p_r} \)\( \frac{1}{n}\sum P_i(\frac{Y_i}{P_i} - Y)^2 \)
Horvitz-Thompson\( \sum_s\frac{y_i}{\pi_i} \)YG: \( \sum_{i
Ratio\( \frac{\bar y}{\bar x}\bar X \)\( \frac{1-f}{n}\bar Y^2(C_y^2 + C_x^2 - 2\rho C_xC_y) \)
Product\( \frac{\bar y\bar x}{\bar X} \)\( \frac{1-f}{n}\bar Y^2(C_y^2 + C_x^2 + 2\rho C_xC_y) \)
Regression\( \bar y + b(\bar X - \bar x) \)min \( \frac{1-f}{n}S_y^2(1 - \rho^2) \) at \( b = \frac{S_{xy}}{S_x^2} \)
  • Selection: cumulative total; Lahiri (accept \( (i, j) \) iff \( j \le X_i \); expected trials \( \frac{M}{\bar X} \)).
  • Fixed size: \( \sum\pi_i = n \), \( \sum_{j \ne i}\pi_{ij} = (n-1)\pi_i \). Two draws WOR: \( \pi_i = p_i[1 + \sum_{j \ne i}\frac{p_j}{1 - p_j}] \).
  • Ratio beats \( \bar y \) iff \( \rho > \frac{C_x}{2C_y} \); product iff \( \rho < -\frac{C_x}{2C_y} \); regression \( \ge \) ratio, tie iff \( B = R \).
  • Ratio bias \( \approx \frac{1-f}{n}\bar Y(C_x^2 - \rho C_xC_y) \); \( \frac{\lvert B \rvert}{\sigma} \le C_{\bar x} \).

CRD, RBD, Latin square

DesignError dfMissing plot
CRD (\( t \) treatments, \( n \) plots)\( n - t \)nothing special
RBD (\( t \) treatments, \( r \) blocks)\( (r - 1)(t - 1) \)\( \frac{rB' + tT' - G'}{(r-1)(t-1)} \)
LSD (\( m \times m \))\( (m - 1)(m - 2) \)\( \frac{m(R' + C' + T') - 2G'}{(m-1)(m-2)} \)
Graeco-Latin (\( m \times m \))\( (m - 1)(m - 3) \)
  • Each missing value: error df \( -1 \). RE of RBD over CRD: \( \frac{(r-1)M_B + r(t-1)M_E}{(rt-1)M_E} \).
  • Latin squares of order \( m \): \( m!(m-1)!L_m \), with \( L_3 = 1 \), \( L_4 = 4 \), \( L_5 = 56 \): 12, 576, 161280.
  • MOLS: at most \( m - 1 \); complete set for prime powers; no orthogonal pair for \( m = 2, 6 \).

Block designs and BIBD

  • \( C = R - NK^{-1}N' \), \( Q = T - NK^{-1}B \), \( C\hat\tau = Q \); \( C\mathbf 1 = 0 \), rank \( \le v - 1 \).
  • Connected iff rank \( C = v - 1 \) iff all contrasts estimable (binary: \( n \ge v + b - 1 \) needed).
  • Orthogonal iff \( n_{ij} = \frac{r_ik_j}{n} \); no incomplete binary design is orthogonal.
  • BIBD \( (v, b, r, k, \lambda) \): \( bk = vr \), \( \lambda(v-1) = r(k-1) \), \( b \ge v \), \( r \ge k \); \( NN' = (r-\lambda)I + \lambda J \).
  • \( C = \frac{\lambda v}{k}(I - \frac{J}{v}) \); \( \hat\tau_i = \frac{kQ_i}{\lambda v} \); \( V(\hat\tau_i - \hat\tau_j) = \frac{2k\sigma^2}{\lambda v} \); \( E = \frac{\lambda v}{rk} \); error df \( bk - b - v + 1 \).
  • Complement \( (v, b, b - r, v - k, b - 2r + \lambda) \). Symmetric with \( v \) even needs \( r - \lambda \) a square (so no \( (22, 22, 7, 7, 2) \)).
  • Benchmarks: \( (7, 7, 3, 3, 1) \): \( V = \frac{6}{7}\sigma^2 \), \( E = \frac{7}{9} \). \( (9, 12, 4, 3, 1) \) resolvable, \( b = v + r - 1 \).

\( 2^k \) factorials and confounding

  • \( [X] = \prod(\ell \pm 1) \), minus for letters in \( X \). Effect \( \frac{[X]}{r2^{k-1}} \); \( SS_X = \frac{[X]^2}{r2^k} \); \( V = \frac{\sigma^2}{r2^{k-2}} \).
  • Yates: \( k \) passes of pair sums (top) and pair differences (bottom).
  • \( 2^p \) blocks: choose \( p \) effects; all \( 2^p - 1 \) products confounded (\( ABC \cdot BCD = AD \)).
  • Principal block: even overlap with each confounded effect; a subgroup; other blocks are cosets.
  • \( 2^4 \) in 4 blocks: some two-factor interaction is always confounded. \( 2^5 \) in 4 blocks: \( ABC, CDE, ABDE \).
  • Partial confounding: information \( \frac{r'}{r} \); SS \( \frac{[X]'^2}{r'2^k} \). \( 2^3 \), 4 replicates, 2 blocks each: error df 17.

Part C checklist

Back to the Sampling Theory and Design of Experiments chapter overview. For the descriptive statistics behind \( S^2 \), \( \rho \) and coefficients of variation, see Descriptive Statistics and Exploratory Data Analysis.

Advertisement
Advertisement