30 Single-Factor Experiments

between a CRD and a RCB design is essentially due to blocking, the value
of the relative efficiency is indicative of the gain in precision due to
blocking.
For our example, the R.E. value is computed as:

R.E. = (3)(648,120) + 4(5)(110,558) = 1.63

(24 - 1)(110,558)

Because the error d.f. is only 15, the adjustment factor is computed as:
k= [(3)(5) + 1][6(3) + 31 =0.982

[(3)(5) + 3][6(3) + 11

and the adjusted R.E. value is computed as:

Adjusted R.E. = (k)(R.E.)
= (0.982)(1.63)

= 1.60
The results indicate that the use of the RCB design instead of a CRD design
increased experimental precision by 60%.

2.3 LATIN SQUARE DESIGN

The major feature of the latin square (LS) design is its capacity to simulta­
neously handle two known sources of variation among experimental units. It
treats the sources as two independent blocking criteria, instead of only one as
in the RCB design. The two-directional blocking in a LS design, commonly
referred to as row-blocking and column-blocking, is accomplished by ensuring
that every treatment occurs only once in each row-block and once in each
column-block. This procedure makes it possible to estimate variation among
row-blocks as well as among column-blocks and to remove them from experi­
mental error.
Some examples of cases where the LS design can be appropriately used are:
" Field trials in which the experimental area has two fertility gradients
running perpeudicular to each other, or has a unidirectional fertility gradi­
ent but also has residual effects from previous trials (see also Chapter 10,
Section 10.1.1.2).
" Insecticide field trials where the insect migration has a predictable direction
that is perpendicular to the dominant fertility gradient of the experimental
field.
• Greenhouse trials in which the experimental pots are arranged in straight
line perpendicular to the glass or screen walls, such that the difference
 Latin Square Design 31
among rows of pots and the distance from the glass wall (or screen wall) are
expected to be the two major sources of variability among the experimental
pots.
Laboratory trials with replication over tifne, such that the difference among
experimental units conducted at the same time and among those conducted
over time constitute the two known sources of variability.

The presence of row-blocking and column-blocking in a LS design, while

useful in taking care of two independent sources of variation, also becomes a
major restriction in the use of the design. This is so because the requirement
tha" all treatments appear in each row-block and in each column-block can be
satisfied only if the number of replications is equal to the number of treat­
ments. As a rest.'t, when the number of treatments is large the design becomes
impiactical because of the large number of replications required. On the other
hand, when the number of treatments is small the degree of freedom associated
with the experimental error becomes too small for the error to be reliably
estimated.
Thus, in practice, the LS design is applicable only for experiments in which
the number of treatments is not less than four and not more than eight.
Because of such limitation, the LS design has not been widely used in
agricultural experiments despite its great potential for controlling experimental
error.

2.3.1 Randomization and Layout

The process of randomization and layout for a LS design is shown below for
an experiment with five treatments A, B, C, D, and E.

0 STEP 1. Select a sample LS plan with five treatments from Appendix K.

For our example, the 5 x 5 latin square plan from Appendix K is:

A B C D E
B A E C D
C D A E B
D E B A C
E C D B A

0 SmP 2. Randomize the row arrangement of the plan selected in step 1,

following one of the randomization schemes described in Section 2.1.1. For
this experiment, the table-of-random-numbers method of Section 2.1.1 is
applied.
* Select five three-digit random numbers from Appendix A; for example,
628, 846, 475, 902, and 452.
32 Single-Factor Experiments
" Rank the selected random numbers from lowest to highest:

Random Number Sequence Rank

628 1 3

846 2 4
475 3 2
902 4 5
452 5 1

Use the rank to represent the existing row number of the selected plan
and the sequence to represent the r,,w number of the new plan. For our
example, the third row of the selected plan (rank = 3) becomes the first
row (sequence = 1) of the new plan; the fourth row of the selected plan
becomes the second row of the new plan; and so on. The new plan, after
the row randomization is:

C D A E B
D E B A C
B A E C D
E C D B A
A B C D E

13 sm'P 3. Randomize the column arrangement, using the same procedure

used for row arrangement in step 2. For our example, the five random
numbers selected and their ranks are:

Random Number Sequence Rank

792 1 4
032 '2 1
947 3 5
293 4 3
196 5 2

The rank will now be used to represent the column number of the plan
obtained in step 2 (i.e., with rearranged rows) and the sequence will be used
to represent the column number of the final plan.
For our example, the fourth column of the plan obtained in step 2
becomes the first column of the final plan, the first column of the plan of
step 2 becomes the second column of the final plan, and so on. The final
 Latin Square Design 33

plan, which becomes the layout of the experiment is:

Row Column Number

Number 1 2 3 4 5
1 E C B A D
2 A D C B E
3 C B D E A
4 B E A D C
5 D A E C B

2.3.2 Analysis of Variance

There are four sources of variation in a LS de-ign, two more than that for the
CRD and one more than that for the RCB design. The sources of variation are
row, column, treatment, and experimental error.
To illustrate the computation procedure for the analysis of variance of a LS
design, we use data on grain yield of three promising maize hybrids (A, B,and
D) and of a check (C) from an advanced yield trial with a 4 X 4 latin square
design (Table 2.7).
The step-by-step procedures in the construction of the analysis of variance are:

o STEP 1. Arrange the raw data according to their row and column designa­
tions, with the corresponding treatment clearly specified for each observa­
tion, as shown in Table 2.7.

o STEP
2. Compute row totals (R), column totals (C), and the grand total
(G) as shown in Table 2.7. Compute treatment totals (T) and treatment

Table 2.7 Grain Yield of Three Promising Maize Hybrids (A,B,and D)

and aCheck Variety (C)from an Experiment with Latin Square Design
Row
Row Grain Yicld, t/ha Total
Number Col. 1 Col. 2 Col. 3 Col. 4 (R)
1 1.640(B) 1.210(D) 1.425(C) 1.345(A) 5.620
2 1.475(C) 1.185(A) 1.400(D) 1.290(B) 5.350
3 1.670(A) 0.710(C) 1.665(B) 1.180(D) 5.225
4 1.565(D) 1.290(B) 1.655(A) 0.660(C) 5.170
Column total (C) 6.350 4.395 6.145 4.475
Grand total (G) 21.365
34 Single'-Factor Experiments
means as follows:

Treatment Total Mean

A 5.855 1.464
B 5.885 1.471
C 4.270 1.068
D 5.355 1.339

03 STEP 3. Outline the analysis of variance as follows:

Source Degree Sum

of of of Mean Computed Tabular F
Variation Freedom Squares Square F 5% 1%
Row
Column
Treatment
Error
Total

13 sTEp 4. Using t to represent the number of treatments, determine the

degree of freedom for each source of variation as:

Total d.f. = 12 - 1 = 16 - 1 = 15

Row d.f. = Column d.f. = Treatment d.f. = t - 1 =4 - 1 = 3

Errord.f.- (t- 1)(t- 2) = (4- 1)(4- 2) = 6

The error d.f can also be obtained by subtraction as:

Error d.f. = Totad d.f. - Row d.f - Column d.f .- Treatment d.f.
= 15 - 3-33-3 = 6

0 smrP 5. Compute the correction factor and the various sums of squares as:

C.F.G

"'(21.365)2 . 28.528952
16
 Latin Square Design 35
Total SS _ZX 2-
C.F.

= [(1.640)2 +(1.210)2 + +(0.660)'] - 28.528952

-1.413923

Row SS t - C.F.

(5.620) 2 +(5.350)2 +(5.225)2 +(5.170)2

4
-28.528952
= 0.030154

Column SS = X - C.F.
I

(6.350) 2 + (4.395)2 + (6.145)2 + (4.475)2

4
-28.528952

= 0.827342

Treatment SS = T - C.F.
I

(5.855)2 +(5.885)2 +(4.270)2 +(5.355)2

4
-28.528952
= 0.426842
Error SS = Total SS - Row SS - Column SS - Treatment SS
= 1.413923 - 0.030154 - 0.827342 - 0.426842

= 0.129585
1sup 6. Compute the mean square for each source of variation by dividing
the sum of squares by its corresponding degree of freedom:

Row MS = Row SS
t-1
0.030154_
0.03015 = 0.010051
3
36 Ningle.Factor Experiments
Column SS
Column MS C- t SS

0.827342 0.275781
___ --
3
Treatment SS
Treatment MS = T t- 1

0.426842
= 3 = 0.142281

Error MS = Error SS
(t- 1)(t- 2)

0.129585
= (3)(2)
= 0.021598
(3)(2)

3 STEP 7. Compute the F value for testing the treatment effect as:

Treatment MS
Error MS

0.142281
0.021598
3 STEP 8. Compare the computed F value with the tabular F value, from
Appendix E, with f, = treatment d.f. = t - 1 and f2 = error d.f. =
(t - 1)(t - 2) and make conclusions following the guidelines in step 9 of
Section 2.1.2.1.
For our example, the tabular F values, from Appendix E, with f, = 3 and
f2 = 6 degrees of freedom, are 4.76 at the 5% level of significance and 9.78
at the 1%level. Because the computed F value is higher than the tabular F
value at the 5% level of significance but lower than the tabular F value at the
1% level, the treatment difference is significant at the 5% level of signifi­
cance.

o Smp 9. Compute the coefficient of variation as:

/Error MS
Grand mean

/0.021598 x 100 11.0%

1.335

o sTrP 10. Enter all values computed in steps 4 to 9 in the analysis of

variance outline of step 3, as shown in Table 2.8.
Note that although the F test in the analysis of variance indicates
significant differences among the mean yields of the four maize varieties
 Latin Square Design 37
Table 2.8 Analysis of Variance (LS Design) of Grain Yield Data In Table 2.7a

Source Degree Sum

of of of Mean Computed Tabular F
Variation Freedom Squares Square Fb 5% 1%
Row 3 0.030154 0.010051
Column 3 0.827342 0.275781
Treatment 3 0.426842 0.142281 6.59* 4.76 9.78
Error 6 0.129585 0.021598
Total 15 1.413923
"cv- 11.0%.
h= significant at 5%level.

tested, it does not identify the specific pairs or groups of varieties that
differed. For example, the F test is not able to answer the question of
whether every one of the three hybrids gave significantly higher yield than
that of the check variety or whether there is any significant difference among
the three hybrids. To answer these questions, the procedures for mean
comparisons discussed in Chapter 5 should be used.

2.3.3 Efficiencies of Row- and Column-Blockings

As in the RCB design, where the efficiency of one-way blocking indicates the
gain in precision relative to the CRD (see Section 2.2.4), the efficiencies of both
row- and column-blockings in a LS design indicate the gain in precision
relative to either the CRD or the RCB design. The procedures are:

C SiuP 1. Test the level of significance of the differences among row- and
column-blocks:
A. Compute the F values for testing the row difference and column
difference as:
Row MS
F(row) = Error MS

0.010051
- <1
0.021598

F(column) = Column MS
Error MS

= 0.275781
0.021598
B. Compare each of the computed F values that is larger than I with
the tabular F values (from Appendix E) with f, = t - 1 and f2 =
It - lXt - 2) degrees of freedom. For our example, the computed
38 Single-Factor Experiments
F(row) value is smaller than 1 and, hence, is not significant. For the
computed F(column) value, the corresponding tabular F values with
f= 3 and 12 = 6 degrees of freedom are 4.76 at the 5% level of
significance and 9.78 at the 1%level. Because the computed F(column)
value is greater than both tabular F values, the difference among
column-blocks is significant at the 1%level. These results indicate the
success of column-blocking, but not that of row-blocking, in reducing
experimental error.

0 STEP 2. Compute the relative efficiency parameter of the LS design relative

to the CRD or RCB design:
• The relative efficiency of a LS design as compared to a CRD:

R.E.(CRD) = E, + E, +(t - 1)E,

(t + 1)Eo

where E, is the row mean square, E, is the column mean square, and E, is
the error mean square in the LS analysis of variance; and t is the number
of treatments.
For our example, the R.E. is computed as:
RE(CRD) = 0.010051 + 0.275781 + (4 - 1)(0.021598)

(4 + 1)(0.021598)

- 3.25

This indicates that the use of a LS design in the present example is

estimated to increase the experimental precision by 225%. This result
implies that, if the CRD had been used, an estimated 2.25 times more
replications would have been required to detect the treatment difference
of the same magnitude as that detected with the LS design.
The relative efficiency of a LS design as compared to a RCB design can
be computed in two ways-when rows are considered as biocks, and
when columns are considered as blocks, of the RCB design. These two
relative efficiencies are computed as:

R.E.(RCB, row)= E, +(1 1)E,

(t)(Er.)

R.E.(RCB, column) = E +(t1)Ee

(th ) (E)

where E,, Eo, Er, and t are as defined in the preceding formula.
 Lattice Design 39
When the error d.f. in the LS analysis of variance is less than 20, the
R. E. value should be multiplied by the adjustment factor k defined as:

3]
[(t- 1)(t- 2) + 1][(t- 1)2 +
[(t- 1)(t- 2) + 31[(- 1)2 + 1]

For our example, the values of the relative efficiency of the LS design
compared to a RCB design with rows as blocks and with columns as
blocks are computed as:
R.E.(RCB, row) = 0.010051 + (4 - 1)(0.021598)

4(0.021598)

= 0.87

R.E.(RCB, column) = 0.275781 +(4 - 1)(0.021598)

4(0.021598)

= 3.94
Because the error d.f. of the LS design is only 6, the adjustment factor k
is computed as:

k = [(4 - 1)(4 - 2) + 11[(4 - 1)2 + 31_ = 0.93

[(4 - 1)(4 - 2) + 3] [(4 - 1)' + 11

And, the adjusted R.E. values are computed as:

R.E.(RCB, row)= (0.87)(0.93) = 0.81

R.E.(RCB, column) = (3.94)(0.93) = 3.66

The results indicate that the additional column-blocking, made possi­

ble by the use of a LS design, is estimated to have increased the
experimental precision over that of the RCB design with rows as blocks
by 266%; whereas the additional row-blocking in the LS design did not
increase precision over the RCB design with columns as blocks. Hence,
for this trial, a RCB design with columns as blocks would have been as
efficient as a LS design.

2.4 LATTICE DESIGN

Theoretically, the complete block designs, such as the RCB and the LS designs
discussed in Sections 2.2 and 2.3, are applicable to experiments with any