Skip to the content

Topics Covered

LSD Concept Layout Advantages Statistical Analysis Critical Difference
On this page
  1. 1. Concept of LSD
  2. 2. Layout of LSD
  3. 3. Advantages and Disadvantages
  4. 4. Statistical Analysis of LSD
  5. Key Take-aways

1. Concept of LSD

DEFINITION

The Latin Square Design is an arrangement of \(t\) treatments in a \(t \times t\) grid such that each treatment occurs exactly once in each row and exactly once in each column. The two-way blocking (rows and columns) controls two independent sources of variation simultaneously.

LSD is more efficient than RBD when the experimental area shows variation in two perpendicular directions — e.g., a field with both a fertility gradient (rows) and an irrigation gradient (columns).

Constraint: number of treatments \(t\) = number of rows = number of columns = number of replications. So LSD requires \(t^2\) plots.

2. Layout of LSD

A standard 4 × 4 Latin square (each letter is a treatment):

4 × 4 Latin Square Layout Row 1 Row 2 Row 3 Row 4 Col 1 Col 2 Col 3 Col 4 A B C D B A D C C D A B D C B A
Fig 4.1 — Each treatment (A, B, C, D) appears once in every row and every column

Construction

Start with a standard Latin square (first row and first column in alphabetical order). For randomization:

  1. Choose a standard \(t \times t\) square at random from a published catalogue.
  2. Randomly permute the rows.
  3. Randomly permute the columns.
  4. Randomly assign actual treatments to the letters.

3. Advantages and Disadvantages

Advantages

  1. Controls two sources of heterogeneity simultaneously.
  2. Substantially more precise than RBD when both row and column effects are real.
  3. Statistical analysis is straightforward (three-way classification without interaction).

Disadvantages

  1. Number of treatments must equal number of rows and columns — restrictive.
  2. Useful only for moderate \(t\) (typically 5 ≤ t ≤ 8); too few df for error if \(t\) is small.
  3. If row or column effect is absent, LSD is less efficient than RBD or even CRD (waste of df).
  4. Missing values complicate the analysis (Unit 5).

Applications

4. Statistical Analysis of LSD

Mathematical Model

\[ y_{ijk} = \mu + \alpha_i + \beta_j + \gamma_k + \epsilon_{ijk}, \]

where \(\alpha_i\) = row effect, \(\beta_j\) = column effect, \(\gamma_k\) = treatment effect, \(\epsilon_{ijk} \sim N(0, \sigma^2)\). Subscript \(k\) is uniquely determined by the row–column position.

Hypotheses

Sums of Squares

Notation: \(R_i\) = total of row \(i\); \(C_j\) = total of column \(j\); \(T_k\) = total of treatment \(k\); \(G\) = grand total; \(N = t^2\); \(C = G^2/N\).

\[ \text{SS}_T = \sum y^2 - C, \quad \text{SS}_R = \dfrac{\sum R_i^2}{t} - C, \quad \text{SS}_C = \dfrac{\sum C_j^2}{t} - C, \] \[ \text{SS}_{Tr} = \dfrac{\sum T_k^2}{t} - C, \quad \text{SS}_E = \text{SS}_T - \text{SS}_R - \text{SS}_C - \text{SS}_{Tr}. \]

ANOVA Table

SourcedfSSMSF
Rows\(t-1\)SSRMSRMSR/MSE
Columns\(t-1\)SSCMSCMSC/MSE
Treatments\(t-1\)SSTrMSTrMSTr/MSE
Error\((t-1)(t-2)\)SSEMSE—
Total\(t^2-1\)SST

Critical Difference

\[ \text{CD} \;=\; t_{\alpha/2,\, (t-1)(t-2)} \cdot \sqrt{\dfrac{2\, \text{MS}_E}{t}}. \]
EXAMPLE 1 (Full LSD analysis)

4 × 4 Latin square. Cell entries are yields and treatment labels:

C1C2C3C4Ri
R1A=20B=24C=22D=1884
R2B=22A=26D=20C=2492
R3C=18D=22A=24B=2084
R4D=24C=20B=18A=2284
Cj84928484344

Treatment totals: \(T_A = 20+26+24+22 = 92,\; T_B = 24+22+20+18 = 84,\; T_C = 22+24+18+20 = 84,\; T_D = 18+20+22+24 = 84\).

\(N = 16, C = 344^2/16 = 7396\).

\(\sum y^2 = 400+576+484+324+484+676+400+576+324+484+576+400+576+400+324+484 = 7488\).

\(\text{SS}_T = 7488 - 7396 = 92\).

\(\text{SS}_R = (84^2 + 92^2 + 84^2 + 84^2)/4 - 7396 = (7056+8464+7056+7056)/4 - 7396 = 29632/4 - 7396 = 7408 - 7396 = 12\).

\(\text{SS}_C = 12\) (same calculation).

\(\text{SS}_{Tr} = (92^2 + 84^2 + 84^2 + 84^2)/4 - 7396 = 12\) (one treatment differs).

\(\text{SS}_E = 92 - 12 - 12 - 12 = 56\). df: 3, 3, 3, 6.

\(F_{Tr} = (12/3)/(56/6) = 4/9.33 = 0.43\); \(F_{0.05, 3, 6} = 4.76\) ⇒ accept \(H_0^T\).

\(F_R = 0.43\), \(F_C = 0.43\) ⇒ no significant row, column or treatment effects.

EXAMPLE 2 (Significant treatment effect)

5 × 5 LSD with \(\text{SS}_R = 30, \text{SS}_C = 25, \text{SS}_{Tr} = 200, \text{SS}_E = 60\). df: 4, 4, 4, 12.

\(F_{Tr} = (200/4)/(60/12) = 50/5 = 10\). \(F_{0.05, 4, 12} = 3.26\) ⇒ reject \(H_0^T\); treatments differ.

CD: \(t_{0.025, 12} = 2.179\); \(\text{CD} = 2.179 \sqrt{2 \cdot 5/5} = 2.179 \sqrt{2} = 3.08\). Compare treatment-mean differences against 3.08.

Key Take-aways

← Previous: Unit 3 Next: Unit 5 — Missing Values & Efficiency →