Scribd
Upload
4 views2 pages

Designs

Uploaded by

Cù Băm
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
4 views2 pages

Designs

Uploaded by

Cù Băm
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

1

Designs
Peter J. Cameron
3 5
Block designs were first used in the design of experi-
7
ments in statistics, as a method for coping with sys-
tematic differences in the experimental material. Sup-
pose, for example, that we want to test seven differ-
ent varieties of seed in an agricultural experiment,
and that we have 21 plots of land available for the
experiment. If the plots can be regarded as identical, 2 6 4
then the best strategy is clearly to plant three plots Figure 1 A block design.
with each variety. Suppose, however, that the available
plots are on seven farms in different regions, with three
Fifteen young ladies in a school walk out three abreast
plots on each farm. If we simply plant one variety on for seven days in succession: it is required to arrange
each farm, we lose information, because we cannot dis- them daily so that no two shall walk twice abreast.
tinguish systematic differences between regions from
differences in the seed varieties. It is better to follow The solution requires a (15, 3, 1) Steiner triple system
a scheme like this: plant varieties 1, 2, 3 on the first with the extra property that the 35 blocks can be par-
farm; 1, 4, 5 on the second; and then 1, 6, 7; 2, 4, 6; titioned into seven sets called “replicates,” each repli-
2, 5, 7; 3, 4, 7; and 3, 5, 6. This design is represented cate consisting of five blocks that partition the set of
in Figure 1. points. Kirkman himself gave a solution, but it was not
This arrangement is called a balanced incomplete- until the late 1960s that Ray-Chaudhuri and Wilson
block design, or BIBD for short. The blocks are the showed that (v, 3, 1) designs with this property exist
sets of seed varieties used on the seven farms. The whenever v is congruent to 3 mod 6.
blocks are “incomplete” because not every variety can For which v, k, λ do designs exist? Counting argu-
be planted on every farm; the design is “balanced” ments show that, given k and λ, the values of v
because each pair of varieties occur together in a block for which (v, k, λ) designs exist are restricted to cer-
the same number of times (just once in this case). tain congruence classes. (We noted above that (v, 3, 1)
This is a (7, 3, 1) design: there are seven varieties; each designs exist only if v is congruent to 1 or 3 mod 6.)
block contains three of them; and two varieties occur An asymptotic existence theory developed by Richard
together in a block once. It is also an example of a Wilson shows that this necessary condition is sufficient
finite projective plane. Because of the connection with for the existence of a design, apart from finitely many
geometry, varieties are usually called “points.” exceptions, for each value of k and λ.
Mathematicians have developed an extensive theory The concept of design has been further general-
of BIBDs and related classes of designs. Indeed, the ized: a t–(v, k, λ) design has the property that any
study of such designs predates their use in statistics. t points are contained in exactly λ blocks. Luc Teir-
In 1847, T. P. Kirkman showed that a (v, 3, 1) design linck showed that nontrivial t-designs exist for all t,
exists if and only if v is congruent to 1 or 3 mod 6. but examples for t > 3 are comparatively rare.
(Such designs are now called Steiner triple systems, The statisticians’ concerns are a bit different. In our
although Steiner did not pose the problem of their introductory example, if only six farms were avail-
existence until 1853.) able, we could not use a BIBD for the experiment,
Kirkman also posed a more difficult problem. In his but would have to choose the most “efficient” possible
own words, design (allowing the most information to be obtained

1
2 Princeton Companion to Mathematics Proof

from the experimental results). A BIBD is most effi-


cient if it exists; but not much is known in other cases.
There are other types of design; these can be impor-
tant to statistics and also lead to new mathematics.
Here, for example, is an orthogonal array: in any two
rows of this matrix, each ordered pair of symbols from
{0, 1, 2} occurs just once:

0 0 0 1 1 1 2 2 2
0 1 2 0 1 2 0 1 2
0 1 2 1 2 0 2 0 1
0 2 1 1 0 2 2 1 0
It could be used if we had four different treatments,
each of which could be applied at three different levels,
and if we had nine plots available for testing.
Design theory is closely related to other combina-
torial topics such as error-correcting codes; indeed,
Fisher “discovered” the Hamming codes as designs
five years before R. W. Hamming found them in the
context of error correction. Other related subjects
include packing and covering problems, and especially
finite geometry, where many finite versions of classical
geometries can be regarded as designs.

You might also like