with the included randomized design. Various arrangements

are possible; the plots may be placed between blocks or

between treated plots within blocks (Fig. 9).

Excluded controls

: control plots are selected outside the

trial area and not adjacent to it, in an area with conditions

closely similar to those of the trial. Replication is not

essential but may be useful if the area is not homogeneous.

The

observations

made

these

plots

should

not

included in the statistical analysis.

Adjacent controls

: each plot is divided equally into 2

subplots, and one of these (at random) is left untreated.

Observations are made in the same way in both sub-plots.

The

observations

made

these

plots

should

not

included in the statistical analysis unless due allowance is

taken of the fact that the design is a form of split-plot. In a

split-plot design, the variability within plots may differ

from that between plots; consequently the analysis of vari-

ance should include 2 strata of error. Specialist statistical

advice may be necessary to interpret the results.

1.6.3 Choice of the type of arrangement of untreated con-

trol

The choice of the arrangement of the untreated control

depends on its role(s) in the trial. Although the included

control has very often been used in the past in efﬁcacy

evaluation trials, and is still frequently used in practice, it

is not necessarily the most suitable. The following decision

scheme gives guidance.

(a) If the control is used in a statistical test, then the

‘included control’ is essential. If not, another type of

control can be used. In either case the heterogeneity of

the plots should be considered;

(b) If heterogeneity is high, the ‘adjacent control’ is suit-

able. If heterogeneity is low or moderate, the interfer-

ence of the control plots with the adjacent plots should

then be considered;

cent plots, then the ‘imbricated control’ is suitable;

(d) If control plots are liable to interfere with adjacent

plots, then the ‘excluded control’ should be used.

1.7 Selection of the sample size in a plot

The main purpose of taking several samples inside a plot is

to reduce the variability of the estimated plot mean to a

suitable level for the assessed variable. The sample size

should be large enough to achieve this purpose. The sample

size required depends greatly on the nature of the observa-

tion and the variability within the plot. EPPO standards on

the assessment of speciﬁc pests, weeds and diseases give

advice on sample sizes. In practice, sample sizes of 10

–

elements are usually enough to accomplish the goal of cor-

rect estimation of the mean value in a plot, depending on

the inherent variability. Note that, if the treatments are

applied to plots, then increasing the sample size within

plots

only

gives

strictly

limited

return

efﬁciency,

because between-treatment comparisons should be made at

between-plot scale.

Sampling

should

always

random

and

should

ade-

quately cover the area of the plot and the experimental

untreated area;

control plot (area under observation)

Block 1

Block 2

Block 3

Block 1

Block 2

Block 3

Block 4

Block 1

Block 2

Block 3

Block 4

Fig. 9

An example of the use of imbricated control plots for a randomized block trial with 4 blocks and 4 treatments.

Design and analysis of efﬁcacy evaluation trials

373

2012 OEPP/EPPO,

Bulletin OEPP/EPPO Bulletin

, 367–381