A DOUBLE CHANGE-OVER DESIGN FOR DAIRY CATTLE

FEEDING EXPERIMENTS v
W. G. COCtIRAI~, K. M. AUT!~EY .~ND C. ¥. CANIkTON
Iowa Agriczdtural Experiment Station, Ames, Iowa

INTRODUCTION

During the winter of 1939-40, a feeding trial was carried out by the Iowa
Agricultural E x p e r i m e n t Station with eighteen Holstein cows, on three
planes of feeding. The three rations tested, and hereinafter designated as
A, B and C, respectively, were as follows:
A. Roughage alone (alfalfa hay and corn silage).
B. Limited grain (alfalfa hay, corn silage and grain fed at the rate of 1
pound for each 7 pounds of milk produced).
C. Full grain (alfalfa hay, corn silage and grai~l fed at the rate of 1
pound for each 3.5 pounds of milk produced).
The grain mixture consisted of the following constituents in parts b y
weight : corn, 4 ; oats, 4 ; cracked soybeans, 1 ; bone nleal, 0.25 ; and salt, 0.15.
The experiment was confined to a sillgle lactation, and was of the short-
time switch-over type, each cow receiving each ration for a period of six
weeks. The layout of the experiment was designed so as to avoid as f a r as
possible the difficulties that commonly arise in interpreting the results of
switch-over trials, and to permit tests of significance of the differences be-
tween the effects of the rations. It is hoped that a brief description of the
layout and of the statistical analysis may be of interest to those engaged in
dairy work.
DESCUIrTrON OF THE DESIGN
I n the ordinary group trial, where each cow receives only a single ration,
~he average milk-yield obtained for a given ration depends on the yielding
ability of the cows receiving the ration. Since cows a r e highly variable in
this respect, the experimental error from this source is often large, though
it may be reduced by skillful grouping of the cows. The switch-over trial
represents an attempt to eliminate this source of variation entirely from the
comparisons between rations, by feeding every cow in t u r n on all rations.
However, in designing switch-over trials, other sources of experimental error
must be taken into account if tile maximum gain in accuracy is to be realized.
One arises from [Link] drop in milk yields towards the end of the
lactation period. Thus if a cow receives rations A, B and C in succession,
her total milk yields under the three rations are scarcely (~omparable, since A
was tested during the most productive period and C during the least. In
Reeei~-ed for publication 5une 16, 1941.
+ Journal Paper No. J-199 of the Iowa AgriculturM Experiment Station, Ames, Iowa.
Project Numbers 416 a~ld 514.
<)37
938 W.G. COCItRAN~K. IV[. AUTREY~ AND C. Y. CANNOX

most switch-over trials whose results we have examined, this drop in yield is
a preponderating factor, the yields falling even when the poorest ration is
given first and the best ration last.
This difficulty can be largely overcome by designing the experiment so
that in any experimental period one-third of the cows are receiving each
ration. F o r instance, one-third of the cows might receive the sequence ABC,
one-third the sequence BCA, and one-third the sequence CAB. W i t h this
arrangement the three experimental periods are equally represented in the
average milk yield for any ration. However, the effects of the lactation
curve are not eliminated entirely from the experimental error of these aver-
ages unless the n a t u r a l rate of fall in yield from period to period is the same
for the three groups of cows. F o r example, if the sequence A B C is given
to a group of cows whose lactation curves show high yields in the first period,
thereafter dropping rapidly, whereas the other two sequences are given to
cows whose lactation curves have a much smaller rate of fall, ration A is
favored relatively to B and C when the average yields over all three groups
are calculated. Thus the division of the available cows into three groups to
receive the three sequences should not be made simply at r a n d o m ; instead,
the object should be to obtain groups which are as similar as possible in the
rates of fall of their lactation curves.
I n the present experiment, the eighteen cows were first divided into six
groups of three on the basis of expected yielding ability, the three cows
within each group being chosen as similar as possible. The three sequences
were then allotted at random to the three cows of each group. The reasoning
was that cows with the same yielding ability would be likely to show a similar
rate of fall in yields. This assumption appeared to be justified by the
results of the trial in which this grouping was used, the total rate of fall per
group decreasing markedly from the high-yielding to the low-yielding
groups, as shown in table 1. I n this table, as in subsequent tables, all the
milk yields were converted to f o u r per cent, fat-corrected milk (2).
The decreases in yield from period I to period I I and from period I I to
period I I I were almost twice as great for the high-yielding groups as for
the low-yielding groups.
TABLE 1
Tolal yield a~d rat~ of fall in yi('ld o[ fat-corrected q~il~"

Groups~ of three cows

1 2 4 5 6
Total milk yield (]bs) ....... 14,351 14,168 12,735 1],391 10,603
Decrease in yield (lbs.)
P e r i o d I-II .............................. 868 889 385 521 472
Period I I - I I I ........................ 1,019 923 598 506 495
Group 3 omitted because oue cow bee-tree sick and had to be rejected.
 DAIRY CATTLE FEEDING EXPERIMENTS 939

W i t h this design, each group of three cows constitutes an independent

experiment, of a type known as the 3 × 3 Latin square (fig. 1) (1).
(lOW
Period
I 2 3
I A B C
II B C A
IIl C A B
FIG. 1. T h e 3 :< 3 L a t i n Square Design (No. 1).

A valid estimate of the experimental error m a y be obtained by an analysis

of variance, as follows :
Degrees of
freedom
Between cows .................................. 2
Between periods ......................................... 2
Between rations .......................................... 2
Error ........................................................................ 2

Total ........................................................................................ S

Differences between cows and between periods are not included in the
estimate of error, since by the n a t u r e of the design they are excluded f r o m
the t r u e errors of the ration means. E a c h of the six groups provides a
separate analysis of variance of the above type. On combining the six
analyses, the following subdivision of degrees of freedom is obtained for the
whole experiment :
Degrees of
freedom
Between groups .................................. 5
Between cows within groups ................. 12
B e t w e e n p e r i o d s w i t h i n g r o u p s ...................... 12
Between rations .................................................. 2
Interaction of rations and groups ... I0
E r r o r ..................................................................... 12

Total ............................................................................. 53

The 12 degrees of freedom for between cows, between periods and error
are the totals of the corresponding terms in the individual groups. A sinfi-
l a r set of 12 degrees of freedom is obtained for between rations. This m a y
be divided into two p a r t s : 2 degrees of freedom representing the average
differences between rations over the whole experiment, a n d 10 degrees of
freedom representing variations f r o m group to group in the differences
between the rations. I f the differences between the rations are the same in
all groups ( a p a r t f r o m experimental errors), the mean square for this t e r m
should be no larger t h a n the error m e a n square, and m a y legitimately be
included in the error. However, the t e r m should be calculated separately
to test this point, p a r t i c u l a r l y where the groups differ widely in yielding
ability or in other respects, e.g., breed.. The remaining 5 degrees of freedom.
corresponding to differences between the group totals, are not included in
the individual analyses, being given here for completeness.
940 w.G. COCttRANt 1K. M. AUTREYt AND C. Y. CANNON

F o r r e a s o n s to be e x p l a i n e d later, t h r e e of the g r o u p s w e r e a s s i g n e d to
t h e L a t i n s q u a r e g i v e n above, a n d t h e r e m a i n i n g t h r e e to t h e c o m p l e m e n t a r y
s q u a r e a s s h o w n i n f i g u r e 2. This feature introduces no change in the
method of analysis.
A ]~ C
C A ]3
B C A
:FIG. o The 3 x 3 Latin Square (No. 2).

ANALYSIS OF VARIANCE OF THE TOTAL DIGESTIBLE NUTRIENT CONSU3IPTION

T h e c o n s u m p t i o n o f d i g e s t i b l e n u t r i e n t s p e r c o w f o r e a c h p e r i o d is g i v e n
i n t a b l e 2. Each of the groups in this table was analyzed separately and the
d e g r e e s o f f r e e d o m , a s w e l l as t h e s u m s o f s q u a r e s , w e r e c o m b i n e d ( t a b l e 3 ) .
according to the procedure previously described. The data on group 3 were
o m i t t e d f r o m t a b l e 2 d u e t o t h e l o s s o f o n e o f t h e c o w s i n it, t h u s c a u s i n g a
loss o f n i n e d e g r e e s o f f r e e d o m i n t h e c o m b i n e d a n a l y s i s .

TABLE 2
Individt~al total d~gcstible n ~ t r i e n t c o n s u m p t i o n ~nits: lbs. p e r cow per period
( t o t a l o f six w e e k s )

Group 1 Group 2

Cow Cow
Period Period Period
1 2 3 total total
4 5 6

I A 608 B 885 C 940 2433 A 527 B 696 C 989 2212

II B 715 C 1087 A 766 2568 C 883 A 635 B 899 2417
III C 844 A 711 B 832 2387 B 785 C 901 A 657 2343
Total 2167 2683 2538 7388 2195 2232 2545 6972

Group 3 Group 4
-I
Cow Cow Period
7 8 9 ]0 11 12 total

This group omitted because cow No. 7 ",vas A 472 ]3 734 C 897 2103
removed from the experiment in Period I, C 819 A 644 B 766 2229
owing t o kidney cancer. B 778 C 953 A 706 2437
2069 2331 2369 6769

Group 5 Group 6

Cow Cow Period

Period
Period total total
1.3 14 15 ]6 ]7 18

I A 586 B 635 C ~05 2026 A 489 B 593 C 788 ]870

II B 723 C 799 A 542 2064 C 730 A 536 B 695 1961
III C 892 A 595 ]3 681 2168 B 674 C 758 A 609 2041
Total 2201 2029 2028 6258 1893 1887 2092 5872

A = roughage, B =Iimited grain, C = full grain.

 DAIRY CATTLE FEEDING EXPERIMENTS 941

TABLE 3
Analysis of variance (3) of t.d.n, cons~[Link]'a units: total cons~[Link] per
cow per period, in lbs.

D e g r e e s of S u m s o~ Mean
freedom squares squares

1. B e t w e e n g r o u p s .......................................... 4 157,936 39,484f

2. B e t w e e n cows w i t h i n g r o u p s ............ 10 105,336 10,534f
3. Between periods withiu groups 10 40,534 4,053 ~
4. B e t w e e n r a t i o n s .......................................... 2 533,869 266,934t
5. R a t i o n × g r o u p i n t e r a c t i o n s ............ 8 ~ 18 12,021 ~ . . . . . . 1,503
6. E r r o r .................................................................. 10 ~ 14,100 J z % , z ± 1,410 ) 1~451
7. T o t a l .................................................................. 44 863,796

Signlficallt.
f Highly significant.

Since the analysis of variance in table 3 shows the " r a t i o n × group inter-
a c t i o n " to be insignificant, this t e r m was inehlded in the experimental error.
The mean square of the ration effects (266,948) is significantly greater t h a n
the error mean square (1450), signifying t h a t the consumption of n u t r i e n t s
was appreciably increased by the addition of grain to the ration. The F-test
(3) reveals t h a t the differences between groups are also highly significant,
the mean square for that t e r m being 39,490 as compared with only 1450 for
Lhe error. This shows that the a r r a n g e m e n t of cows into outcome groups
was well justified.
TABLE 4
Digestible nutrient consumptions-totals and ~eans ~tnits : lbs. for six wee~'s periods

Ration totals Ration means S t a n d a r d error

Rations (15 cows)

A ............. 9,083 605.5

B .................... 11,091 739.4 _+9.83
C ............................ 13,085 872.3
Grand total = 33,259 G e n e r a l m e a n = 739.1

Table 4 shows the mean digestible n u t r i e n t consumption for each ration,

and the s t a n d a r d error per cow. The s t a n d a r d error, ~/1451'= 38.08 is 5.15
per cent of the general mean, 739.1. The s t a n d a r d e r r o r f o r the mean of 15
cows is 3 8 . 0 8 / ~ / ~ = 9.83 pounds, or only 1.33 per cent of the mean. The
relative p e r f o r m a n c e of the groups was in the order anticipated, all differ-
ences between pairs of rations being highly significant.

ADJUSTMENTS FOR CARRY-OVER EFFECTS

I n the case of total n u t r i e n t consumption the amount eaten in one

period did not a p p e a r to be influenced by the ration given in the previous
period. However, with other factors, p a r t i c u l a r l y total milk yield, a carry-
over effect of the ration given in the previous period m a y be anticipated,
owing to the shortness of the change-over period f r o m one ration to another.
942 w.G. COCItRAN~ K. 1K. AUTREY~ AND C. Y. CANNON

I f such residual effects are present, simple averages do not give unbiased
estimates of the effects of the rations. F o r instance, with the layout shown
in figure 1, ration A is preceded by ration C in both the second and third
periods, and likewise B by A and C by B. I f A is the poorest ration and
C the best, the milk yields for ration A may be increased by the beneficial
carry-over effect of C, while those for ration C may be depressed by the
carry-over effect of A. In these circumstances, the simple averages would
underestimate the differences between the direct effects of the rations.
Thus when carry-over effects are present, the average milk yields under
each ration must be adjusted in some way to avoid bias. These adjustments
can clearly be made only if the design of the experiment enables us to esti-
mate the sizes of the carry-over effects. This cannot be done with the design
shown in figure ], since each ration is always followed by the same ration.
Thus, if this design is used for all six groups, no corrections for carry-over
effects are possible, except perhaps by inspecting the daily yields, omitting
from the results those parts of the experimental periods which appear to
show carry-over effects. This method is at best somewhat arbitrary, and
would be unsatisfactory if carry-over effects persisted through most of the
succeeding periods.
However, by using both the cycles shown in figures 1 and 2, a direct
evaluation of the carry-over effects is possible. This may be seen by examin-
ing the total milk yields for each of the six cycles, shown in table 5.
TABLE 5
Total milk yi¢,lds for each of the six ration ~.ycb~.~
( T o t a l s f o r 3 cows (18 weeks) in lbs.)

Sets ~
Period
1 2 3 4 5 6

I ...... A 4,383 B 5,370 C 5,720 A 4,427 ]~ 4,675 C 5,236

I I ...... B 4,057 C 4,980 A 4,208 C 4,535 A 3,637 B 4,420
I I I ...... C 3,834 A 3,264 B 3,571 I~ 3,833 C 3,787 A 2,956

Set totals ] 2,274 ].3,6] ~ 13,499 ] 2,795 ] 2,099 ]2,6] 2

Grand total 7~,893
A = Roughage B - ]Amited g r a i u C = Full g r a i n
R a t i o n totals :
A = 22,875 B = 25,926 C = 28,092

To complete set 1, in which a cow was rejected, values were iuserted fro' the missing
cow by a method described below.

The results during the first period do not require adjustment, since all
cows received the same preceding ration. I n the second period, two sets of
cows, 3 and 5, received roughage. Of these, set 3, having had full grain
in the previous period, gave a total of 4208 lbs., while set 5, having had
limited grain in the previous period, gave a total of 3637 lbs. The differ-
ence, 571 lbs., is an estimate of the difference between the carry-over effect
 DAIRY CATTLE t~Et~)ING E~XPER,IMENTS 9~

of full grain and that of limited grain. Similarly, a comparison of the

carry-oYer effects of full graln and roughage is obtained from sets 1 and 6
in the second period, and a comparison between limited grain and roughage
f r o m sets 2 and 4. The same comparisons are repeated with the results in
the t h i r d period. In all six comparisons, the carry-over effect was in the
same direction as the direct effect (i.e., full grain > limited grain > rough-
age), indicating strongly that real carry-over effects were present.
We m a y now consider how to obtain estimates of the direct effects of the
rations, free from the disturbance due to carry-over effects. This may be
done in m a n y ways. F o r instance, an unbiased comparison of the direct
effects of roughage and limited grain in each period is given by the mean
milk yields of the following groups of cows:
Period Roughage Limited grMn
I Sets 1 a u d 4 Sets 2 and 5
II Set 3 Set 6
III Set 2 Set 4

Since sets 3 and 6 both had f u l l grain in the first period, their yields in
the second period, u n d e r roughage and limited grain respectively, are sub-
ject to the same carry-over effect, and similarly for sets 2 aud 4 in the t h i r d
period. An estimate of the difference between the two rations for the whole
experiment could be obtained by taking some simple average of the differ-
ences in the individual periods. Such an estimate is, however, open to criti-
cism on the grounds that only half of the available cows are used in the
second and t h i r d periods. A f u r t h e r objection, probably more serious, is
that the estimate would be p a r t l y a group comparison, since only the cows
in sets 2 and 4 are common to both roughage and limited grain. The use
of such an estimate would therefore defeat the purpose of the switch-over
design, which is to avoid group comparisons.
To retain the fnll advantage of the switch*over design, the estimate of
the difference between the effects of two rations must clearly be unaffected
by differences in the yielding ability of the cows or by differences between
the average yields for different periods. The most accurate estimates, sub-
ject to these restrictions, are given by a technique known as the method of
least squares, which is frequently used for dealing with similar problems
arising in field experiments. In this procedure, the observed milk yield for
any cow in any period is expressed as a linear function of the effects of the
cow's yielding ability, the period, the current and previous rations, and the
experimental error. F o r example, consider a cow receiving in succession
limited grain, full grain and roughage. Its milk yields yl, y2, y3 in the three
periods are expressed as follows :
y~=m+p~ +rib+%
y2 =m-~ p 2 + d o + r b + e 2
Y3=m+P3 + d~+rc+e3
944 w.G. COCttRANt K. 3~. AUTREY~ AND C. Y. CANNON

w h e r e m = m e a n y i e l d f o r t h e cow
P~, P2, P~ = effects of t h e t h r e e p e r i o d s
d~, db, d~ = d i r e c t effects of t h e r a t i o n s
rb, r~ = r e s i d u a l effects of t h e r a t i o n s g i v e n i n the p r e v i o u s p e r i o d
e~, e2, e3= e x p e r i m e n t a l e r r o r s .
N o p a r a m e t e r is r e q u i r e d to r e p r e s e n t r e s i d u a l effects in t h e first period.
T h e c o n s t a n t s p~, p2, p:~ r e m a i n u n c h a n g e d f o r all t h r e e cows in the same
g r o u p , since o n l y t h e g r o u p a v e r a g e s of t h e differences b e t w e e n p e r i o d s a r e
e l i m i n a t e d f r o m t h e e x p e r i m e n t a l e r r o r s . A f t e r s e t t i n g u p e q u a t i o n s of this
t y p e f o r e v e r y cow, t h e c o n s t a n t s a r e e s t i m a t e d b y m i n i m i z i n g t h e sum of
s q u a r e s of t h e e x p e r i m e n t a l e r r o r s . T h e d e t a i l s of t h e m a t h e m a t i c a l a n a l y s i s
w i l l n o t be g i v e n here. To c a l c u l a t e t h e a d j u s t e d y i e l d s g i v e n b y t h e least
s q u a r e s solution, t h e f o l l o w i n g q u a n t i t i e s ( T a b l e 6) a r e r e q u i r e d , all b e i n g
e a s i l y o b t a i n a b l e f r o m t a b l e 5.
TABLE 6
Data required for calculating the adjusted yields for each ratio~
A 22,875 a 15,950 s2 + s6 26,226
]3 25,926 b 15,407 s3 + s~ 26,294 T 76,893
C 28,092 c 15,725 s~ + s~ 24,373 (2/3) T 51,262

H e r e A , B, C a r e s i m p l y the t o t a l y i e l d s u n d e r t h e t h r e e r a t i o n s , as given
a t t h e f o o t of t a b l e 5. T h e q u a n t i t i e s a, b, c a r e the t o t a l y i e l d s of a l l cows
r e c e i v i n g these r e s p e c t i v e r a t i o n s in the previous period.
e.g., a = 4057 t 3571 + 4535 + 3787 = 15,950, etc.
A l s o s2 ~ s6 = t o t a l of sets 2 a n d 6 = 13,614 + 12,612 = 26,226, etc. 1 F i n a l l y .
we r e q u i r e d t w o - t h i r d s of t h e t o t a l of all y i e l d s in t a b l e 6 [ (§) T = 51,262].
T h e a d j u s t e d m e a n y i e l d s p e r cow p e r p e r i o d ( t o t a l of six weeks) a r e t h e n
given by the following equations:
Roughage
72:?~ = 5 A + (2a - b - e) -: s~ + s~ - (~) T
= 5 x 22,875 + (2 × 15.950 - 15A07 - 15,725) -i 26,226 - 51,262
= 90,107

I t e n e e 5"~ = 1251.5
Limited grain
72y~ = 5B + (2b - a c) + s~ ~ s4 - ( ~ ) T = 103,801 g i v i n g y~ -- 1441.7
Full grain
72yo = 5C + (2c - a - b) + s~ -t so - (~) T = 113,664 giving' y~ = 1578.7
The c o r r e s p o n d i n g u n a d j u s t e d m e a n y i e l d s p e r cow p e r p e r i o d a r e the
t o t a l s A, B a n d C, d i v i d e d b y 18 ( t h e n u m b e r of cows in each t o t a l ) .
The a d j u s t m e n t s r e d u c e d t h e m e a n y i e l d u n d e r r o u g h a g e b y 19 lbs., a n d
i n c r e a s e d t h e m e a n y i e l d u n d e r f u l l g r a i n b y a b o u t t h e s a m e a m o u n t , these
Sets 2 and 6 received roughagc during period I I I ; Sets 3 and 4 received limited
grain~ and Sets 1 and 5 full grain.
 DAIRY C A T T L E F E E D I N G EXPERIMENTS 945

~TABLE 7
Mea~ yields of f a t corrected millv per cow per period of six wee~s

Unadjusted Adjusted for carry-over effects

Ration Increase over Increase over

Mean Mean
roughage roughuge

lbs. lbs. % lbs. lbs. %

A 1270.8 1251.5
B 1440.3 + ]69.5 + 13.3 1441.7 + 190.2 + 15.2
C 1560.7 + 289.9 + 22.8 1578.7 + 327.2 + 26.1
Standard error± 21.4 + 30.3 ± 2.4

effects being in the direction anticipated by the previous discussion. It is

interesting to observe that the mean yield is practically unaltered for the
cows receiving limited grain, for which the beneficial carry-over effect from
full grain and the detrimental effect from roughage appeared to cancel. By
failure to adjust for carry-over effects, the differences between the rations
would have been underestimated by about 11 per cent.
To verify that the yields have been freed from carry-over effects, we may
create artificially an additional carry-over effect of say, the full grain ration,
by adding 1000 to the yields in table 5, wherever full grain was fed in the
previous period. The yields affected are 3264 (set 2), 4208 (set 3), 3833
(set 4) and 4420 (set 6). With this change, the reader may verify that the
adjusted mean yields become:
l~oughage 1325.6 lbs., Limited grain 1515.8 lbs., Full grain 1652.7 lbs.
The effect of the artificial carry-over effect is simply to increase all three
means by the same amount, 74.1 lbs., the differences between the ration means
remaining exactly the same as before. With unadjusted means, on the other
hand, the yields for roughage and limited grain would both be increased by
111.1 lbs., that for full grain being unchanged. The same property holds for
the adjusted mean yields if a constant amount is added to all yields following
any one of the three rations.
It may be of interest to examine the differences between the carry-over
effects of the three rations. For a single cow (total of six weeks), these are
given by the deviations of the following quantities (Table 8) from their
mean.

TABLE 8
Estimates of the carry-over effects per cow (total for 6 weelcs)

Deviation
(ms.)
R o u g h a g e : ......................... ( 3 a + A + s2 + s ~ ) / 2 4 = 9 6 , 9 5 1 / 2 4 = 4 , 0 4 0 - 58
L i m i t e d g r a i n ' . ............ (3b+B+s~+s~)/24=98,441/24=-4,102 ÷ 4
F u l l g r a i n : .............................. ( 3 c + C +s~+ss)/24=99,640/24=4,152 +54
946 w.G. COCIIRAN~ K . M. AUTREY~ A N D C. Y. C A N N O N

The difference in carry-over effect between roughage and full grain was
112 lbs., about one-third of the corresponding difference (327 lbs.) between
the direct effects. The same ratio holds a p p r o x i m a t e l y for the other two
differences between the rations. The carry-over effects seem strikingly large
in relation to the direct effects; however, the figures given above agree well
with the rough estimates which m a y be made f r o m table 5. The six cows
receiving limited grain during the second and third periods and full grain
in the previous period gave a total yield of 4420 + 3833 = 8253 lbs., as com-
p a r e d with 7628 lbs. for the six cows having limited grain preceded by rough-
age. The difference, 625 lbs., corresponds to 104 lbs. per cow, which checks
closely with the estimate of 112 lbs. given above.
The a d j u s t m e n t s required f o r the direct effects m a y be obtained f r o m the
above estimates for the carry-over effects. F o r example, of the eighteen
yields f r o m cows receiving roughage, six followed a previous ration of lim-
ited grain, six followed full grain, while none followed roughage. Thus the
total yield of the eighteen cows should be reduced by six times the sum of the
carry-over effects for limited grain and roughage, i.e., the m e a n should be
reduced by
(6/18) [54 ~ 4] = 58/J3 = 19.3 lbs.;
in agreement with the reduction already given in table 7.

T t t E A N A L Y S I S OF V A R I A N C E W t I E N CARRY-OVER E F F E C T S A R E P R E S E N T

I n analyzing the results for total n u t r i e n t consumption, we discarded

results for the group containing the cow which was removed f r o m the experi-
ment. While this is the simplest procedure, it results in an appreciable loss
of information. B y a technique described b y Yates (4) it is possible to in-
clude in the analysis the two healthy cows in g r o u p 3. Tile first step in this
technique is to estimate results for the missing cow b y the method of least
squares. The completed data can then be analyzed by the usual procedure.
with certain slight changes mentioned below.
The individual yields of f a t corrected milk are shown in table 9, the
values inserted for the missing cow being denoted by asterisks. I n the
analysis of variance, the total sum of squares and the sums of squares for
groups, cows within groups, periods and the periods by groups interaction
are found in the same way as for t o t a l n u t r i e n t consumption. Only the
sums of squares between rations require special consideration. Of these,
two degrees of freedom measure differences between the direct effects of the
three rations, and two measure differences between the carry-over effects.
However, the two sets of two degrees of freedom do not add to the correct
total sum of squares for direct and carry-over effects, because the two effects
are entangled b y the n a t u r e of the design. This total sum of squares (4
degrees of freedom) m a y be obtained either as direct effects (ignoring carry-
over effects)+ carry-over effects, or as carry-over effects (ignoring direct
effects) + direct effects.
 DAIRY CATTLE FEEDING EXPERIMENTS 947

TABLE 9
I n d i v i d t ~ a l y i e l d s o f f a t c o r r e c t e d . m i l k ~ w i t s : l b s . p c r c o w p c r p e r i o d ( t o t a l o f s i z weeks)

Group 1 Group 2

Cow Cow
Period Totals Totals
1 2 3 4 5 6

I A 1376 B 2088 C 2238 5702 A 1863 B 1748 C 2012 5623

II ~ 1246 C 1864 A 1724 4834 C ].755 A 1353 B 1626 4734
III C 1151 A 1392 B 1272 3815 B 1462 C 1339 A 1010 3811
Totals 3773 5344 5234 14351 5080 4440 4648 14168

Group 3 Group 4

Cow Cow
Period Totals Totals
7 8 9 10 11 12

I A 1665 ~ B 1938 C 1855 5458 A ]384 B 1640 C 1677 4701

II B 1517 ~ C 1804 A 1298 4619 C 1535 A 1284 B 1497 4316
I][ C 1366 ~ A 969 B 1233 3568 B 1289 C 1370 A 1059 3718
Totals 4548* 4711 4386 13645 4208 4294 4233 12735

Group 5 Group 6

Cow Cow
Period Totals Totals
13 14 15 16 17 18

I A 1342 B ]344 C 1627 4313 A 1180 B 1287 C 1547 4014

II B 1294 C 1312 A 1186 3792 C 1245 A 1000 B 1297 3542
III C ]317 A 903 B 1066 3286 B 1082 C 1078 A 887 3047
Totals 3953 3559 3879 11391 3507 3365 3731 10603

A = Roughage, B = Limited grain, C = Fuji grain.

Values inserted for cow which was rejected d u r i n g the first period.

The sum of squares for direct effects (ignoring carry-over effects) is

f o u n d in the same way as for total n u t r i e n t consumption, being :
(1/18) [ (22,875) 2 ~ (25,926) 2 + (28,092) 2 - 1/3 (76,893) 5] = 763,282.
The sum of squares for direct effects (adjusted for carry-over effects) is ob-
rained f r o m the adjusted ration totals given above (p. 944), as follows:
(1/360) [(90,107) 2 + (103,801) 2 + (113,664) 2 - 1/3 (307,572) 2] = 777,534.
To calculate the two sums of squares for carry-over effects, we require the
totals f o r the three rations, with and without a d j u s t m e n t for direct effects.
The f o r m e r have already been obtained (Table 8).
The adjusted sum of squares is:
(1/72) [(96,951) 2 + (98,441) 2 + (99,640) 2 - 1/3 (295,032) 2] = 50,409.
The u n a d j u s t e d totals are given b y subtracting A, B and C, respectively
f r o m the corresponding adjusted totals. The sum of squares is:
(1/90) [(74,076) 2 + (72,515) 2 + (71,548) 2 - 1/3 (218,139) 2) ] = 36,158.
As a check, we m a y note t h a t :
763,282 + 50,409 = 813,691 : 777,534 + 36,158 = 813,692;
 948 W.G. COCI-IRAN, K . 5I. A U T R E Y , AND C. ¥ . CAI~NON

t h i s figure b e i n g t h e iota] s u m of s q u a r e s (4 d e g r e e s of f r e e d o m ) f o r d i r e c t
a n d r e s i d u a l effects. The e r r o r s u m of s q u a r e s is f o u n d b y s u b t r a c t i o n .

TABLE 10
Analysis of variance of f a t corrected m~l} units : total yield per cow per period, in lbs.

Degrees of Sum of Mean

freedom squares squares
Total .......................................................................................... 50 5,124,267
Periods .................................................................................... 2 2,041,769
Interaction of periods with groups ............ 10 193,341 19,334
Groups .................................................................................... 5 1,311,769 262,354
Cows within groups ...................................................... 11 ~ 654,638 59,513
i Residual (ignoring direct) ...... 2 36,158
Direct ............................................................ 2 777~534 ~ 388,767
Rotations Direct (~gnoring residual) ...... 2 763,282
l~esidual ...................................................... 2 50,409 ~ 25,204
Error .......................................................................................... 181 109,058 6,059
One degree of freedom subtracted for missing cow.
t Two degrees of freedom subtracted for missing cow.

T h e c o m p l e t e a n a l y s i s of v a r i a n c e is s h o w n in t a b l e 10, b o t h sets of 2
d e g r e e s of f r e e d o m f o r d i r e c t a n d c a r r y - o v e r effects b e i n g i n c l u d e d . W h i l e
o n l y one set n e e d be c a l c u l a t e d to o b t a i n t h e e r r o r s u m of s q u a r e s b y sub-
t r a c t i o n , b o t h a r e r e q u i r e d if t e s t s of significance of the d i r e c t a n d c a r r y -
o v e r effects a r e d e s i r e d . T h e m e a n s q u a r e s f o r d i r e c t effects (388,767) a n d
c a r r y - o v e r effects (25,204), a r e b o t h significant, t h e 5 p e r c e n t F - v a l u e f o r 2
a n d 18 d e g r e e s of f r e e d o m b e i n g 3.55. T h e e s t i m a t e d s t a n d a r d e r r o r p e r
cow ( t o t a l of six weeks) is 77.84 lbs., o r 5.5 p e r c e n t of the m e a n y i e l d of
1424 lbs. T h e c o r r e s p o n d i n g s t a n d a r d e r r o r f o r the m e a n of 18 cows w o u l d
be 77.84/~v/18. H o w e v e r , t h i s m u s t be i n c r e a s e d to allow f o r t h e m i s s i n g cow,
a n d f o r t h e a d j u s t m e n t s n e c e s s a r y to c o r r e c t f o r c a r r y - o v e r effects. The
effect of t h e m i s s i n g cow is to d e c r e a s e t h e effective r e p l i c a t i o n f r o m 18 to
n e a r l y 16.5, w h i l e t h e a d j u s t m e n t s i n c r e a s e t h e s t a n d a r d e r r o r b y t h e f a c t o r
k/1.25. T h u s t h e s t a n d a r d e r r o r of t h e a d j u s t e d m e a n y i e l d s in t a b l e 7 is
77.84 × ~ / ~ / ~ j ~ 1 6 . 5 = 21.4 lbs., o r 1.5 p e r c e n t of t h e m e a n . T h e d e s i g n
a t t a i n e d a s a t i s f a c t o r y d e g r e e of p r e c i s i o n ; a n o b s e r v e d i n c r e a s e of 5 p e r cent
w o u l d h a v e b e e n d e t e c t e d as s t a t i s t i c a l l y significant, w h i l e t h e a c t u a l differ-
ences b e t w e e n t h e r a t i o n m e a n s w e r e all h i g h l y significant.
I t h a s a l r e a d y been n o t i c e d ( t a b l e 1) t h a t t h e r a t e s of f a l l in y i e l d f r o m
p e r i o d I to p e r i o d I I I d i f f e r e d g r e a t l y f o r the d i f f e r e n t g r o u p s of cows. This
f a c t is reflected i n t h e a n a l y s i s of v a r i a n c e , t h e m e a n s q u a r e f o r t h e i n t e r -
a c t i o n of p e r i o d s w i t h g r o u p s b e i n g 19,334 as a g a i n s t 6,059 f o r t h e e r r o r
m e a n s q u a r e . I f t h e six cycles h a d been a s s i g n e d to t h e e i g h t e e n cows a t
r a n d o m , w i t h t h r e e cows to each cycle, t h e e s t i m a t e d e r r o r m e a n s q u a r e
w o u l d h a v e been a p p r o x i m a t e l y (24 × 6,059 + 10 × 1 9 , 3 3 4 ) / 3 4 = 9,963. The
device of first d i v i d i n g t h e cows i n t o six g r o u p s of t h r e e on t h e basis of ex-
p e c t e d y i e l d i n g a b i l i t y t h e r e f o r e r e s u l t e d in a m a r k e d i n c r e a s e in precision.
 DAIRY CATTLE FEEDING EXPERIMENTS 9~9

Similarly, the mean square between cows in the same group (59,513) is
much larger than the error mean square. This indicates that a n y estimates
based on group comparisons would have been subject to much higher ex-
perimental errors than those avoiding the use of group comparisons.
There is, of course, nothing to be gained by adjusting the means in cases
where there are no carry-over effects. The important feature of the design
is that it enables corrections to be made where these are considered necessary.
A n examination of the records, presented as in table 5 will usually be suffi-
cient to decide the issue. I n more doubtful cases, the direct and residual
effects should be calculated as above, comparing the mean square for residual
effects with the error mean square in the analysis of variance. I f the former
mean square is noticeably higher than the latter, it is safer to make the
adjustments.

OTIIER NU!VfBERS OF REPLICATIONS AND TREATMENTS

I n the above design, the experimental unit consists of six cows, one re-
ceiving each cycle. To retain the full balance of the design, the number of
cows used must be a multiple of six. I n an experiment with 6k cows, the
calculations are exactly as described above, except that all divisors should be
multiplied by k/3. Thus if 24 cows had been used, the expressions 72ya,
72yb, 72y¢ for the adjusted totals would be replaced by 96ya, 96yb, 96yc
respectively, the righthand sides of the equations remaining unchanged.
The same rule applies to the divisors in the analysis of variance. Where
there are no missing cows, the standard error of the adjusted means is
/~5s2
~ - - where s 2 is the error mean square in the analysis of variance.

A d j u s t m e n t s of this type cannot be made when only two rations are being
compared in a switch-over design, since each ration must always precede the
other. To obtain direct information on carry-over effects, say in a single
reversal trial, it would be necessary to include cows receiving the sequences
A A and BB, as well as cows receiving the switch-over sequences AB and BA.
W i t h four treatments there are twenty-four possible cycles. However,
the important features of the design can be obtained with only twelve cows,
by the use of the following trio of 4 × 4 Latin squares :

Cows Cows Cows

Period
1 2 3 4 5 6 7 8 9 10 11 12

I ................... A B C D A B C D A B C D
I I ................... B A D C D C B A C D A B
I I I ................... C D A B B A P C D C B A
IV .................. D C B A C D A B B A D C

Ym. 3. Change-over design for four rations.

950 W.G. COCHRAN', K. M. AUTREY, AND C. Y. CANNON

Of the three A ' s in period I I , one follows B, one follows C and one follows
D. The same is true for a n y ration in a n y period a f t e r the first. The quan-
tities required for estimating the direct and residual effects of ration A (with
and without a d j u s t m e n t ) are shown below.
TABLE 11
Calculation of t],c direct and resid~al effects i~z an e x p e r i m e n t w i t h f o u r rations

Divisors

:Effect Totals For For analy-

sis of
means variance

D i r e c t ( u n a d j u s t e d ) ... A 12 12
D i r e c t ( a d j u s t e d ) ...... llA+ (3a-b-c-d) +s~+s~+slo-½ T 120 1320
Residual (unadjusted) 4a + s~ + s7 + slo 132
R e s i d u a l ( a d j u s t e d ) .. 4a + A + s~+ s7 + sj~ 30 120

The symbols here have the same definitions as on pages 14 and 15. The
sets sd, sT, s~o are those in which A is given during the f o u r t h period, the same
rule a p p l y i n g to rations B, C and D. The first column of divisors gives the
numbers by which the corresponding totals must be divided to obtain esti-
mates per cow per period. The second column shows the divisors required
in the analysis of v a r i a n c e ; in each case these a p p l y to the sums of squares
of deviations of the totals for the four rations f r o m their mean. Since the
divisors are f o r a single unit of 12 cows, they must be multiplied by k where
k units are used in the experiment. The f o r m u l a for the standard error of
/~.ls 2
the adjusted means per cow per period is ~/12-k-' where s ~ is the error mean
square.
Since f o u r is p r o b a b l y the m a x i m u m n u m b e r of rations that could be
given in a single lactation, no designs are presented for higher numbers of
rations.
I t should be m~derstood that the designs described above a p p l y particu-
l a r l y to short-time trials. Since it is difficult to keep a group of cows going
for several lactations without some losses, such designs would require modi-
fication for use in long-time trials (of 3--4 years). A n i m a l losses in experi-
ments of this t y p e destroy the balance of the design, thus making a thorough
and efficient analysis of the results difficult.

SUMMARY

The design and statistical analysis of a short-time switch-over trial com-

p a r i n g three rations are discussed in this paper. The principal objects of
the design are to secure accurate comparisons of the effects of the rations
and unbiased estimates of the experimental errors. The design also makes
it possible to e s t i m a t e and adjust for carry-over effects, which have usually
 DAIRY CATTLE FEEDING EXPERIMENTS 951

been ignored in switch-over trials. The appropriate statistical analyses are

illustrated by numerical examples of the ease in which carry-over effects are
negligible (nutrient consumption) and the ease in which they are not
negligible (milk production). A corresponding design for four rations is
briefly described.
A detailed report of the results of this feeding trial will be given in a
future paper.
REFERENCES
(i) ]FISHEI~, R.A. The Design of Experiments. Chap. 5. Oliver and Boyd, London~
Eng. 1935.
(2) GAINES, W. L., AND I)AVIDSON, 1~. A. Correction of Milk Yield for Fat Content. Ill.
Agr. Exp. Sta. Bul. 265. 1923.
(3) SNEDECOR,G . W . Statistical Methods. 3rd ed. Chap. 10 and 11. Collegiate Press,
Inc., Ames, Iowa. 1940.
(4) YATES, F. The Analysis of Replicated Experhnents when the Field Results are In-
complete. Empire Jour. Expt. Agr.~ 1: 129-142. 1933.