109
448

 

_THS_

MTOGRAWS T0 EVALUATE Si’iQNGE-S 1% TYPE
OF DAIR‘! CAETLE

masks for the flame of M 8‘
MIGHIGAN STAYE UNEWQSEW
Ciiixaéfifi; 22E {EGEKE

19273

 

L I B R A R Y
Michigan Stat:
University!‘

ABSTRACT

PHOTOGRAPHS T0 EVALUATE CHANGE IN TYPE
OF DAIRY CATTLE

BY
Charles Lee Goeke

Photographs might offer partial relief to changes
in scale, ideals, and appraisers hampering measures of
trends in type. I examined variation of scores of 161
Jersey cows from 344 7 x 7 cm color transparencies taken
3 months after calving, each rated twice by four judges in
five half«day sessions over 2.5 weeks. The Jerseys were
selected for single traits in closed herds from 1956 to
.1967. Correlations between ratings from duplicate photo-
graphs ranged from .72 to .91 for individual judges. Near-
ness in time of second rating to first of the same
photograph influenced correlations (.91 in same session;
.81 in different sessions).

The average score was 9.0 on a scale of 0 to 17.
Cow within sire, judge, year, sire within line, line,
interaction of year and line, parity number, inbreeding,

and errors of measurement were sources of differences.

Charles Lee Goeke

Variances were 2.18 between means of cows, .07 between
scales of judges, .03 between ratings of the same cow in
different years, and 1.45 residual. Product correlations
between judges averaged .68. This agreement of judges is
near the .74 previously experienced with ratings of the
actual cow. Average yearly scores rose to a high in 1961,
then decreased slightly. Trends in type can be evaluated

by photographs to hold ideals at time of rating constant.

PHOTOGRAPHS TO EVALUATE CHANGES IN TYPE
OF DAIRY CATTLE

By

Charles Lee Goeke

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of ‘

MASTER OF SCIENCE
Department of Dairy Science

1973

6-7‘C‘xO

DEDICATION

To '
my w1 fe Dolores

j-Ip
HP

ACKNOWLEDGMENTS

I express my appreciation to Dr. Lon D. McGilliard
for his interest, guidance, and critical review of this
‘manuscript.

Financial assistance offered by the Department of
Dairy Science is gratefully acknowledged.‘

Use of the Michigan State University computing
facilities was possible in part through support from the
National Science Foundation.

I thank my parents, Mr. and Mrs. Arthur Goeke,
Jr., for their encouragement and interest in education.
Most of all, thanks to my wife, Dolores, for her encourage-

ment, interest, and understanding.

H9
H0
Hp

TABLE OF CONTENTS

LIST OF TABLES . . . .
INTRODUCTION . . . . . . . .
REVIEW OF LITERATURE . .
SOURCE OF DATA . . . . . .

Herd Composition
Management . . . . .
Selection Procedures
Type Scores

METHODS AND RESULTS . . .

Introduction

Duplicate Ratings
Judges, Cows, Years
Age, Parity . . . .
Sire, Line . . . . .
Agreement among Judges
Trends . . . . . . .

CONCLUSION . . . . . . . .
REFERENCES . . . . . . . .

iv

Page

000000:

10

15

15
16
21
22
25
33
35

38
40

LIST OF TABLES
Table ‘ Page

1. Correlations of duplicate ratings in session
combinations for each judge . . . . . . . . . . 18

2. Correlations of duplicate ratings by judge
and proximity of sessions . . . . . . . . . . . 19

3. Least squares deviations of type ratings
of photographs for judge, session, and
judge«session subclasses . . . . . . . . . . . 21

4. Analysis of variance in type ratings
0 to 17 from photographs . . . . . . . . . . . 22

5. Yearly and overall type ratings by ages of
photographs of Jersey cows, combined herds . . 23

6. Number of calvings by age and year
for Jersey cows . . . . . . . . . . . . . . . . 24

7. Mean type ratings of photographs of Jersey
cows for each lactation . . . . . . . . . . . . 26

8. Average type scores of photographs of
daughters for bulls coded by line and
sequence of use . . . . . . . . . . . . . . . . 27

9. Mean yearly type of photographs of Jersey
cows for each line of selection . . . . . . . . 28

10. Mean type of photographs of Jersey cows
for lactationvline combinations . . . . . . . . 29

11. Analysis of variance in type scores from
photographs of Jersey cows . . . . . . . . . . 30

12. Analysis of variance and variance components
of type ratings from photographs of
Jersey cows . . . . . . . . ... . . . ..... . . 31

-.. ‘. '57;— .LLI

,__.e
L.

Table Page

a .n'
1 __L

 

. v

13. Correlations of judges' ratings of type

from photographs . . . . . . . . . . . . . . . . 34

14. Overall yearly averages and average yearly
type scores of photographs of Jersey cows
for each judge . . . . . . . . . . . . . . . . . 36

:~ 1HW‘q-ma f1- _
i J u . - ' H -

' :' " ‘4 -zv-r inmtwrwfinmfi.‘ -

 

   

vi

INTRODUCTION

Measurement of type, body conformation, is a com.
ponent of dairy cattle evaluation. Classification and
the showring are two formal current methods to measure
type. Most type measurements are on actual animals; hon
ever, pictures are important in advertising, in some forms
of showring competition as in All American programs of
breed societies, and could be useful for measurement of
type.

Measuring trends in type of dairy cattle to evalu«
ate programs in breeding, nutrition, or management involves
difficulties not usually experienced with characteristics
more objectively measured. Previous attempts to measure
type and its changes have been with ratings of live ani-
mals by one or more judges. The study examines the
possibility of rating pictures (color transparencies) to
measure type and trends in type to relieve partially com-
plications caused by change in judges' scales, change in
ideals, and change of judges. All three hamper evaluating
trends or changes in type. If relationships between
variations are suitable, another purpose is to apply the

method to data of a breeding project.

Questions asked in fulfilling the purpose were:
(1) Can judges find common factors in pictures upon which
to rate animals? (2) What are possible causes of discre-
pancies between ratings of color transparencies? (3) Are

these small enough to permit detection of trends?

REVIEW OF LITERATURE

Dairy cattle conformation or type has been a part
of evaluating cattle for utility, for aesthetics, and in
merchandising. As early as 1814 the Highland Agricultural
Society was awarding prizes to Ayrshires at local shows.
In 1834 a scorecard or scale of points to guide showring
judging was developed on the Island of Jersey (9). The
scorecard has changed several times to its present form,
the Uniform Dairy Scorecard, providing a uniform guide to
type evaluation of five dairy breeds (1, 9).

Breed organizations have developed programs to
encourage improvement of type. True type models of Hol—
stein bull and cow were developed in 1924 (9). Today each
breed has models which set standards to aid type evalua-
tion. Each dairy breed organization of the United States
has type classification, the earliest started in 1928 by
the Holstein organization. These programs were to provide
more widespread and uniform type evaluation than was pr0*
vided by showring competition (9). Selection of All»
American animals is another program of breed organizations
to promote improvement of type by recognition of superior
show animals. Photographs of an animal and information

about its season‘s show reCord are submitted to a panel

3

of judges to choose All-American, Reserve All-American,

and Honorable Mention winners (9). Breed organizations
have other programs to improve type that differ between
organizations and change over time. Artificial insemina-
tion organizations also have programs for type improvement.
These programs vary from one organization to another.

Photographs are used in merchandising animals, in
competition, for identification, and for other purposes.
Research on the effectiveness of measuring type through
photographs has been limited, but sources of variation in
actual ratings of live animals have been investigated
more thoroughly. My review is of these studies of ratings
of live animals.

Potential sources of variation in type scores are
animals, judges, time of rating, animal's parents, ani-
mal's age, and interactions among these. Variation among
animals is expected to be large since cows differ and
type ratings intend to reflect differences in body con»
formation. Different judges should contribute little
variation in ratings where they, with similar ability,
work from the same model and with the same scale. Large
variation from judges would indicate discrepancy from
these conditions.

A change or difference in scale from one rating
to another would be reflected in variation from cows being

rated at different times. Another source of variation is

groups of paternal half sisters; this can be large when
one bull's daughters differ substantially in type from
daughters of another bull. Age could influence variation
also if type changes appreciably with increased age of
animal.

Variation from interaction between cow and date
of measuring is expected to be large where cows change
differently in type with time (8). A small variance from
interaction of cow with judge indicates judges prefer the
same cows, or have similar ideals.(8). Variation of year
x judge is also small if change in judges‘ scales is simi-
lar from one year to another (8).

Examination of variances of ratings on live animals
may indicate sources of variance in type scores from photOv
graphs. Total variance in type ratings of live animals
was 5.45 with a mean score of 6.64 points on a 0 to 17
scale for Holstein cows (8). A comparable total variance
was 5.26 for Holstein and Guernsey cows (12). Variation
in ratings due to cows was 3.02 or 55% of the total (8)
and 1.64 or 31% (12). Variations between cows were .484
and .727 when Ayrshire cows were rated two or more times
by different inspectors, or rated by one judge three times
in a year on scale 1 to 5 (S).

Judges have contributed little to total variation
of type ratings, variation among them being insignificant

in three studies (8, ll, 3) and accounting for 2.5% in

another (10). Johnson and Lush (7) also stated that
judges contributed a small amount of variance; however,
their experiment was not designed to separate this infor-
mation.

Year or time of classification has been a source
of small variance, 3% and 6% of the total (8, 12). Both
rounds of classifiers and time of classification were in-
significant when official classifiers rated New York herds
at least three times between 1950 and 1963 (3).

Contribution from interactions of cow with date,
cow with judge, and year with judge to the total variation
was also examined. Variance of cow with date was largest,
.65 or 12% and 1.45 or 27.4% of the total (8, 12). Vari-
ation from interaction of cow with judge and judge with
year accounted for 1% and 5% of the total (8). Product-
moment correlations confirm the large interaction between
cow and year or COW'and date and the small interaction of
cow with judge. The animal's change in appearance was an
important source of varianCe while the difference or
shift in judge's ideals with time was unimportant (8).

Residual variation, left unexplained, depends on
what sources were examined or the completeness of the
model. Residual variance was 1.27 or 23% of the total
which included interaction of cow x judge x year plus
anything not accounted for by these sources individually

or in any two.way interaction (8). A residual variance

of .735 or 14% of the total included all not accounted for
by cow, judge, or cow x date interaction (12). ‘Variation
within cow including everything but differences among cows
was .403 (5).

Variance attributed to an animal's parents, age,
and stage of lactation has been examined previously.
Animal's sire accounted for 8.5% of the total variation
(10). Changes in type of cows with advancing age were
not large but were significant between ages 4 and 5; vari~
ance due to age was small (5). Differences between scores
at different stages of lactation were small but signifi-
cant when early or late part of the lactation was compared
with the middle segment. Stage of lactation was a small
source of variation (2, 5).

In this thesis I intend to examine variation and
sources of variation of type ratings where ratings are
from pictures of cows. I shall compare these with studies

of direct ratings of live animals.

SOURCE OF DATA

Type ratings were from photographs of animals in
three Michigan State University Jersey herds. Originally
two of the herds were closed in an inbreeding experiment
while the other was a teaching herd. Herd composition,
management information, selection procedures, and method

of obtaining type scores follow.

Herd Composition

 

Foundation animals for the inbreeding study were
brought to Michigan State University in 1951 from three
California dairies which had been using a series of bulls
from inbreeding experiments of the University of Califorv
nia. In 1954 two mating groups were formed according to
closeness of animal's relationship to each of two sets of
full brothers available for service.

Breeding plans changed in 1955 shifting the groups
so animals with the closest genetic covariance with the
dam of one set were in one group and those most closely
related to the dam of the other set of sires were in a
second group. One of the two groups was chosen by chance

to be selected for milk production and was designated the

select herd. The other group with mating choices by chance
was the control herd. These two herds were kept together
and managed as one herd with about 15 cows in milk in

each (4).

In 1955 the teaching Jersey herd which was housed
at a different location was incorporated into the plan.
This group was unrelated to select and control herds and
was selected for type. In 1961 all animals were moved to

the same location and managed as one herd.

‘Management

 

Standard good herd management procedures were
followed as closely as possible with available facilities.
Environmental conditions for the cows were to be as cons
stant as practical with moderations for animals to be in
nutrition and management experiments. During summer,
lactating cows were on pasture whenever it was available
and practical, with grass silage and grain as a supple-
ment. Corn silage, good quality legume hay, and grain
were fed cows during winter months. Grain was fed accord-
ing to milk production with adequate added amount to
younger cows for growth.

Heifers were bred at first regular heat after
reaching either 15 mo or 250 kg body weight, whichever
came first. A heifer not pregnant at 2 yr of age was

discarded as a nonbreeder.

10

Cows were bred at the heat which was nearest 60

days postpartum. A cow still open after 10 mo was dis-

carded.

Palpations of genital organs were postpartum,

before breeding, 60 to 90 days after breeding, and when-

ever there were irregularities (4).

Selection Procedures (4)

 

1. Removal of cows from herd

A.

C.

Control

Retain all females until one record is complete.
Maintain herd size at about 15 cows in milk.
When it is necessary to eliminate cows to
maintain herd size, excess cows shall be re«
moved at random from among cows which have

completed at least one lactation.

Select herd

1. Maintain herd size at 15 cows in milk.

2. When it is necessary to eliminate animals to
maintain herd size, animals ranking lowest in
index value based on milk production of the
individual and her relatives shall be removed
from the herd.

Type herd

1. Maintain herd size at 15 cows in milk.

2. When it is necessary to eliminate animals to

maintain herd size, animals with the lowest

D. All

11

average type score from ratings three times a
year shall be removed from the herd.

three herds

Any heifer which has not conceived after eight
services or by 24 mo of age shall be removed
from the herd.

Any cow which has not conceived within 10 mo
from the previous calving date shall be removed
from the herd.

Females which cannot be milked by machine
because of udder structure, injury, etc.,
shall be removed from the herd.

Animals shall be removed from the herd for

any disease or injury for which removal is

the recommended veterinary control.

II. Choice and removal of bulls.

A. Control herd

1.

After the birth of each bull, determine by
chance whether he shall be retained or diSv
carded.

When the number of bulls 6 mo to 12 mo of age
exceeds three, or the number over 12 mo of
age, exclusive of herd sire, exceeds two, dis-
card excess bulls from each age group at
random.

Choose herd sire by chance from bulls of

12

breeding age which have not previously served

as herd sire.

Select herd

1.

Designate periodically the best cows from
which to save bulls on basis of milk produc$
tion of them and their close relatives.

When it is necessary to discard young bulls
to maintain proper numbers for replacement,
discard bulls scoring lowest in index value
of milk production of close relatives.

Use the bull with the highest index value as

herd sire.

Type herd

1.

Designate periodically the best cows from
which to save bulls on their average type
classification from evaluations every 4 mo.
When it is necessary to discard young bulls

to maintain proper numbers for replacement,
discard bulls whose dams have lowest average
type scores.

Use the bull whose dam has the highest average
type score as the herd sire.

When a bull is chosen as a herd sire, continue
to use him as herd sire until either another

bull has a dam with a higher average score, or

13

the herd sires in the other two herds have
been replaced.
D. All three herds

1. When a bull is chosen to be herd sire, use
him as soon as he is able to serve. Retain
the previous herd sire or a substitute until
the fertility of the new herd sire is ascer-
tained.

2. Discard infertile bulls as well as those

unable or unwilling to serve.

'Type‘Scores

 

The leader of the project took color transparen-
cies 7 x 7 cm of the right side of each animal in the
three Jersey milking herds approximately 3 mo after she
calved. Cows were weighed, measured, scored, posed and
photographed usually within 3 days of 3 mo after calving.
Photographs accumulated from December 1956 through 1964
with a total of 344 pictures from 131 cows. Each slide
was randomly assigned two of 688 possible numbers from 1
to 689 excluding 234 for sequential presentation for
scoring.

During January, 1965 a panel of four judges simul-
taneouslyand independently evaluated the type conforma«
tion of the Jersey cows from 2 x 2 m projection of each

transparency. The scale was 0 (low poor) to 17 (high

14

excellent) for overall score and 0 (poor) to 5 (excellent)
for each subclass of type. Subclasses included mammary
system, body capacity, dairy character, and general
appearance along with teat placement, rear udder, fore
udder, barrel, chest, hind feet and legs, fore feet and
legs, rump, and breed character.

The presentation included each transparency twice
in the order arranged by procedure to provide randomness.
Three morning sessions and two afternoon sessions with
about 2.5 wk between the first and last sessions were
required to evaluate the slides, The panel has previous
judging experience but had not evaluated type from slides.
Bach judge recorded his own scOres and was uninformed of

duplication of photographs.

METHODS AND RESULTS

Introduction

 

Analysis of variation in scores was separated
into several sections because of the dependency of later
portions on earlier results. Relationships of first and
second ratings of the same photograph were first examined
by two methods. The absolute difference between the two
ratings was regressed on the difference between sequence
numbers to learn if discrepancies were larger between
duplicate ratings separated by more time than between
pictures rated closer in time. Correlations of first and
second ratings of the same slide were also examined for
consistency of ratings, especially those widely separated
by time. These are sources of potential error in scoring.

Another section contains examination of whether
judges' scales of scoring changed from one session to
another. A substantial change would require adjusting
scores to a common session for the remainder of the study.
Scores of same pictures rated in each of two sessions
were compared for change in scales.

Other sources of variation in type scores were

examined after the relationship of duplicate ratings and

15

16

change in judges' scales. Variances due to judges, cows,
year of picture, and all two«way interactions were inves-
tigated for each judge and for judges combined. Average
ratings by each judge for each year of photograph were
also compared. Age at photographing, parity number of
animal, sire, line of selection, and inbreeding were
examined for variation by comparing average scores and
analyzing variance. Variances were partitioned into com-
ponents for cow, year, and judge. Agreement among judges
was examined by comparing productvmoment correlations of

simultaneous ratings of a slide by different judges to

base correlations of two ratings of a slide by one judge.

‘Duplicate Ratings

 

Time between ratings of a slide as a source of
discrepancies between duplicate ratings was examined by
regression of differences in scores on differences in
sequence number and by correlations of duplicate ratings.
Large regressions or small correlations indicate time is
a large source of variation, and causes would need inves-
tigating. 0

Absolute differences between scores of duplicate
ratings by the same judge were regressed on positive
differences in sequence numbers, a nonlinear measure of
interval between repeated ratings by same judge. Slopes
of line for each judge were .0005, .0007, .0002, and
.0003 with F values for the null hypothesis b-O of 3.3,

17

12.6, .5, and 1.9. Only for judge 2 was the slope signifi—
cantly different from zero at P < .01. With a maximum
difference between sequence numbers of two ratings of 688
and a slope of .0007, scores of a photograph rated first
and subsequently last of all by judge 2 would be expected
to differ by less than .5 point, an unimportant difference.
As time between duplicate ratings of a slide increased,
differences between ratings remained almost nil.
Correlation of ratings separated by approximately
equal time appear in columns in Table 1. Groups of seSv'
sion combination correspond to sessions in which first and
second ratings occurred. A picture rated both times in
session 1 was in group (l/l), but a slide appearing first
in session 2 and second in session 5 was in group (2/5).
Approximately half the correlations of duplicate
ratings in the same session were larger than other corv
relations for each judge. The high correlations were in
session combination (1/1) for judges 2 and 4; (2/2) for
judges l, 2, and 3; (4/4) for judges l and 2; and (5/5)
for judges 1, 2, and 3. There were also three correlas
tions of duplicate ratings in the same session which were
lower than other correlations. These were in session com~
binations (1/1) and (4/4) for judge 3, and (1/1) for judge
1. Four of the five correlations of duplicate ratings in
the same session were larger than remaining correlations

for judge 2; however, only one was higher for judge 4.

18

 

muwum

 

 

 

 

 

 

 

 

ma ON a mm wH mN mm on am mH om 0H «N 0N m we .oz
mm. ow. mm. mm. um. um. mm. am. am. mm. ms. nu. em. up. am. e
om. am. an. ow. mm. Nm. mo. mN. mm. em. mm. mm. ow. Ho. 00. n
mm. mm. cm. mm. mm. mm. mm. mm. om. mm. mm. No. am. mm. mm. N
mm. ow. um. ow. ow. ow. mu. em. mm. mm. on. em. mm. mm. mo. H
m\m m\e ¢\e m\& «\m m\m m\N v\N m\~ «\N m\H «\H m\H N\H H\H owvsh
mcoflumafinEou :oflmmom
.omwsn some new m¢0mumcwnaoo :memom :fi mmcfiumh ouwowamsw mo m:0wuwfiopaou .H manna

19

Duplicate ratings can be regrouped into "differ-
ence groups" further to investigate consistency of ratings.
Difference groups are composed of one or more session com-
binations, and correlations in Table 2 are averages of
appropriate session combination correlations. Averages
of correlations were by Zatransformations for normality
of distribution. The 0 group contains ratings of all
slides rated twice in one session while the 1 Difference
group contains ratings of slides evaluated in adjacent
sessions and 2 Difference has one session between sessions
of paired ratings.

Table 2. Correlations of duplicate ratings by judge and
proximity of sessions.

 

 

 

:—

 

 

 

 

 

 

‘Dggfigr‘ Judge Aver... 3:12:£.9f
l 2 3 4 tions
0 .90 .96 .85 .88 .91 70
1 .86 .90 .72 .77 .83 107
2 .84 .90 .62 .81 .81 87
3 .81 .88 .70 .77 .80 . 57
4 .79 .89 .73 .73 .80 26

 

The average correlation of duplicate ratings in
the same session was .91, slightly larger than correlations

of duplicate ratings in different sessions whose averages

20

ranged from .80 to .83 and were not significantly differ-
ent. Ratings of a slide appearing twice in the same
session were more consistent than those in different ses-
sions, but once in different sessions, consistency of
ratings was similar. This with variations of degree was
typical for each judge.

Negligible regression of differences in scores on
difference in sequence number indicated time between rat-
ings was not important. Correlations of duplicate ratings
indicated time introduced error but even when duplicate.
ratings were in different sessions, error was small. I
concluded that time is a minor source of discrepancies
between ratings.

Large average differences between duplicate'ratv
ings in different sessions could indicate change in judge's
scale over time, requiring adjustment of ratings to a
common session for analysis of differences between cows
or trends in type. Lack of apparent reason for time
between ratings being a source of discrepancies in ratings
would necessitate considering it in subsequent analysis.

I compared average scores of identical pictures
in different sessions. These duplicate pictures were in
judgewsessions between which deviations in Table 3 difv
fered substantially. Large changes of scale seemed more
likely between these cells. Differences between dupli-

cates in sessions 1 and 2 for judge 1, sessions 2 and 5

21

Table 3. Least squares deviations of type ratings of
photographs for judge, session, and judge-
. session subclasses.

 

 

- Session
Judge Mean
1 2 3 ' 4 5

l .37 ..29 «.04 .05 -.08 -.17
2 «.03 *.14 .,03 «.12 .32 -.39
3 .11 .32 «.24 «.09 «.10 ’ .42-
4 “.45 .ll .31 .17 «.14 .13

Mean «.16_ . v.06 . .ll .28 «.17

 

Scale of type ratings 0 to 17, low poor to high excellent;
mean = 9.01, 16w good plus.

for judge 2, sessions 2 and 3 for judge 3, and sessions 1
and 3, 1 and 4 for judge 4 were small enough to deny ’
changes in scale important enough to need adjustment.

Thus, I did not adjust scores for subsequent analyses.

Judges, Cows, Years

 

I partitioned variation in type scores of color
transparencies into differences attributed to cow, year,
and interaction between cow and year for each judge. Means
from all three sources differed significantly at P < .01
for each of the four judges. The analysis of variance

with judges combined is in Table 4. Judge, year, cow, and

22

Table 4. Analysis of variance in type ratings 0 to 17
from photographs.

 

 

 

Source df MS F**
Judge 3 87.92 124
Year 8 49.63 70
Cow 160 38.77 55
Year X Judge 24 7.75 11
Cow X Year 175 4.85 ,7
Cow X Judge 480 2.82 4

Error 1901 .71

 

**P < .01 for all classifications.

,all two«way interactions are significant at P < .01. The

three«way interaction is included in the error term.

Age, Parity

 

Differences in ratings because of age might be
reflected by different average scores for groups by age
of cow. Average scores in years 1957 through 1964 for
each age and an overall average are in Table 5. Age
seemed to be an important source of variation with aver-
age scores generally rising with each age group. A

regression equation limited to linear and quadratic terms

23

.msam poem 30H .m u

.o oases :a cam Haou non mzoo mo uoaasz

:moe mpcoHHooxo swan ou noon 30H .ua ou o mwcwpmp max“ mo oawom

 

 

 

 

 

3 2d 3.” £5 35 £5 2M
em oa.m um.m m~.w nw.m HH
0H va.m MH.NH mn.n OH
Nu Hm.w m~.m hm.HH ma.“ mH.m mH.m um.m m
eoa mm.m om.oH um.a Hw.m oo.m o~.m um.m w
NHH No.m om.o we.HH mN.oH nm.oa mn.w mn.w we.» a
NmH ~m.m mm.oa ve.m oa.m um.“ ev.oH wm.oH om.HH oo.» a
cam Hm.m mH.m Hm.m ew.w mm.m ~m.m mH.oH no.» mN.HH m
can mm.w um.u mH.m mv.m ov.m nm.m me.m wo.m oo.m e
owe m~.m Ne.m 5H.w e~.m nw.m no.m oo.m mo.m wm.n n
«em we.» mm.m 55.x om.“ mo.m ~m.w Nw.w ov.w om.w N
coma mama NomH HomH coma mmma mmmH nmma
honssz :50: om<
Mao»
.mwao: womanEou
.mzou sompoh mo mammuwouogm mo mowm an mmcwuma omxu HHmho>o paw >~amo> .m oHnma

24

of mean scores on age shows how average type scores changed
with age.
Average score = 8.20 + .34 (age) . .02 (age)2
Age is age at calving to nearest year. Average type score
rose to a peak around age 8 or 9 yr, then slowly decreased.
The increase in average score with age and the seem-
ing importance of age to type can in part be explained by
selection and consequential reduction in numbers. The
number of animals in each age category decreased rapidly
(Table 6), and since one line was selected on type, some '
animals left the herd for inferior type, increasing the

average score with age.

Table 6. Number of calvings by age and year for Jersey cows.

 

 

1 I _—_— _ L
fi— r1 t —_

 

 

Age at Year
Calving
yr. 1957 1958 1959 1960 1961 1962 1963 1964
2 8 16 28 23 14 12 7 10
3 l 16 ll 16 9 13 6 8
4 4 6 7 5 7 5 3 8
5 l 4 2 3 5 8 4 8
6 2 2 2 3 5 2. 2
7 4 l 1 2 l 2 3
8 l 5 2 2 l 2
9 l 2 2 2 l l
10 l l
11 1 l l

_>_12 1 1 1 2

 

25

Aging can also be measured by parity number since
one picture was taken per lactation. Table 7 contains
means for parities and numbers of observations per lac-
tation. Scores of lactations 1 through 4 compare favorably
with ages 2 through 5, Table 5. Discrepancies are probably
due to calving intervals more than 12 mo.

Differences in ratings because of interaction
between age and year are reflected by differences between
average ratings by age not being the same from one year
to another. Differences between the average scores by
age in Table 5 change from year to year; hence, interac-
tion of age and year seemed another important source of
variation in type scores. However, age can be ignored as
a source of variation in yearly herd averages if distriv

bution of cows by age is similar for each year.

' Sire, Line

 

Significant differences in scores between cows
suggested animal's sire and sequence of sire usage were
likely sources of variation. Too, sire accounted for
8.5% of the total variation of type scores in one study
(10). I coded bulls by line or herd and sequence of use.
Line 1 was selected for milk production, line 2 control,
and line 3 was selected for type. All bulls used before
the breeding project began were pooled to make up sequence

1. Bulls were used from September, 1956 through 1966 with

26

Table 7. Mean type ratings of photographs of Jersey cows
for each lactation.

Lactation ' Number of

 

Number Observations Mean
1 984 8.64
2 656 9.24
3 408 9.06
4 240 9.58
S 136 , 8.89
6 120 9.44
7 104 9.21
8 48 9.19
9 8 9.37

10 8 9.87
11 8 9.25
12 8 9.50
13 8 9.25
14 _8 9.00

 

Scale of type ratings 0 to 17, low poor to high excellent;
mean = 9, low good plus. ,

changes approximately yearly, but as specified in the breed-
ing plan, time was not primary in deciding when to change.
Occasionally bulls in line 3 were switched to keep the

number of bulls used per line about equal.

27

Average type scores for daughters in each sire
sequence in Table 8 differ, sometimes substantially, from
one bull to another in the same line. Differences in
average scores indicate animal's sire is a source of vari-
ation in type ratings and differences between sires can be
detected in ratings of type from photographs. Three of
four bull's daughters in the type line have higher average
ratings than daughters of bulls in the milk line.

Table 8. Average type scores of photographs of daughters
for bulls coded by line and sequence of use.

 

 

 

Sfiggggie ' 1(Milk) 2(0332301) 3(Type)
1 9.01 8.05 10.04
2 . 8.91 9.26 9.97
3 9.01 8.32 8.29
4 8.38 8.75 11.33
5 7.59
6 8.13

 

Scale of type ratings 0 to 17, low poor to high excellent;
mean = 9, low good plus.

Differences in daughter averages between lines,
and selection of one line for type suggests line as a
source of variation. Means of years and lactations for

each line in Tables 9 and 10 show scores for animals

28

Table 9. Mean yearly type scores of photographs of Jersey
cows for each line of selection.

 

 

Line

 

Year 1(Milk) 2(Control) 3(Type)
1957 8.31 7.61 9.65
1958 8.82 8.10 9.80
1959 9.07 8.32 9.95
1960 8.99 8.77 9.94
1961 8.96 8.46 10.32.
1962 8.73 8.46 9.20
1963 9.58 8.26 9.56
1964 8.47 9.73

9.28

 

Scale of type ratings 0 to 17, low poor to high excellent;
mean = 9, low good plus.

selected for type are higher than both other lines except
in 1963. Average scores of animals selected for milk pro-
duction are above control animals for all years and for
all lactations except lactation 6. These averages provide
further evidence that line or herd is a source of differv
ences in type ratings. Interaction of year and line is
suggested by differences between average scores of lines
differing from year to year.

Age, sire, and line as possible sources of varia'~
tion as well as inbreeding were more completely analyzed,

Table 11. This analysis is on data from 1957 through 1964

29

Table 10. Mean type ratings of photographs of Jersey cows
for lactation«line combinations.

 

 

 

Lfififiggion chilk) 2(Cgigiol) 3(Type)
1 8.57 8.09 9.20
2 9.40 8.53 10.01
3 9.11 8.27 10.12
4 9.43 9.07 10.41
5 8.95 6.88 10.16
6 8.83 9.00 10.27
7 8.96 8.33 10.25
8 8.44 10.69
9 9.37

:10 9.50

 

Scale of type ratings 0 to 17, low poor to high escellent;
mean = 9, low good plus.

and uses lactation numbers to measure age. Judge, line,
sire within line, cow within sire, year, year by line
interaction, lactation number, and inbreeding were all
significant at P < .01. This analysis of variance con-
firms that variations from different ages, sires, and
lines in type scores are reflected through photographs.
The previous sections of the analysis answered

one of the study's major questions: what were possible

30

Table 11. Analysis of variance in type scores from photo-
graphs of Jersey cows.

 

 

Source df MS P**
'Judge 3 87.5 171
Line 2 487.9 953
Sire/line ll - 55.6 109
Cow/sire 147 29.8 58
Year 7 23.6 46
Year x line 14 5.6 11
Lactation Number 13 22.5 44
Inbreeding 42 ‘ 54.9 107

Residual 2504 .5

 

**Significant at .01.

causes of differences between ratings of color transparen-
.cies. Judges, years, cows, their twosway interactions,
sires, lines, ages, inbreeding, year x line interaction
and errors of measurement were sources of differences.

Variance components in Table 12 show the contribu«
tion of judges, years, and cows to the total variation in
ratings of color transparencies. The mean of the ratings
was 9.0 with a variance 3.74. Percentage of the total

each component accounts for is also given.

31

Table 12. Analysis of variance and variance components of
type ratings from photographs of Jersey cows.

L
_v._-:—

 

 

 

% of
Source df MS EMS Component total
. 2 2 2
Judge 3 87.92 OR + K20C + KloJ .07 1.9
2 2 2
Year 8 49.63 OR + K20C + KSOY .04 1.1
Cow 160 38.77 03 + K206 2.18 58.3
Rep. 2580 1.45 ofi 1.45 38.7
Total 2751 3.74 100.0
K1 = Number of observations per judge = 688.
K2 = Average number of observations per cow = 17.1.
K3 = Weighted number of observations per year = 298.3.

Constants to estimate components are K1, the num-
ber of observations per judge, K2, the arithmetic mean
number of observations per cow, and K3, a weighted mean
number of observations per year. The equation to calculate

K3 was: ,

number of

years = 9

K1 = number of ob-
servations
in year i

K. - total number
of observa»
tions.

K3 = [l/(nsl)][K.n(XK§/K.)] where n

32

These K terms are approximations because of unequal sub-
class numbers and expected values of means square were
used for simplicity rather than least squares methods.

Variance due to cows, 2.18 or-58.3% of the total
is largest. Standard error of this component was .06.
Studies of ratings on live animals found variance due to
cows of 3.0 or 55% of the total variation (8) and 1.64 or
31% (12). The .07 for variation in judges' scales also
agrees with 0 in other studies (8, 12). Variance due to
year of picture was .04 or 1.1% of the total. This coma
pares with .2 (8) and .3 (12). The residual variance was
1.45 or 38.7% of the total.

Components of variance for type ratings from photo-
graphs are similar in magnitude and percentage of the
total to ratings on live animals. Cow is the largest
source of differences in scores as was hoped because rat-
ings are to detect differences between cows. The magni-
tude of variation of cow compared to other components and
the residual indicates real differences between cows are
detectable in photographs. Variation of judges was small
indicating all judges had similar scales and ideals. This
was also expected since judges should be working from the
same model. Variation in scores from years was also small.
Reasonably large variation from years would be expected if

selection on all animals increased type scores, but only

33

one-third of the cows were being selected for type, and

management was reasonably constant.

 

‘Agreement'amongJudges

Correlations to investigate further sources of
variation to confirm earlier results, and to help answer
the question, "Could judges find differences independently
observed and agreed upon in pictures upon which to rate
animals?" was the final portion of the analysis. Correla-
tions between two ratings of a slide by each judge, along
the diagonal in Table 13, shows agreement of judges with
themselves. Judge 2 was more consistent in his ratings
with a correlation of .91 than judge 3 with .72. All
judges were reasonably consistent in scoring as they could
recognize similarities in the photographs from one rating
to another. I

Offvdiagonal elements in Table 13 are correlations
between simultaneous ratings by different judges. These
range from .59 to .77 averaging .68. Similar types of
correlations between ratings of live animals rated almost
simultaneously by different judges were .74 (8), .61 to
.76 (2) and .73 and .76 (12). Agreement of judges rating
type from photographs is similar to agreement of judges
rating live animals.

Correlations between ratings from photographs by

the same judge at different times are only slightly, if

34

Table 13. Correlations of judges' ratings of type from

 

 

 

 

 

 

photographs.
Judge 1 Judge 2 Judge 3 Judge 4
Judge 1 .85 .77 .59 .72
Judge 2 .91 .61 .71
Judge 3 .72 .69
Judge 4 .81

 

Average correlation of different judges = .68.

any, higher than correlations of simultaneous ratings by
different judges. This similarity of moderately large
correlations indicates lack of influence from the signifiv
cant but small interaction between cow and judge found
earlier in this study. Judges seem to find common factors
in photographs of animals to rate body conformation.

The correlations analysis confirmed earlier
results and indicated cows differ in ways that judges are
able to observe these differences from photographs and
agree relatively well on how to score these differences.
The conSistent correlations wouldn't have existed others
wise. Agreement of correlations of ratings on photographs
with correlations of live animal ratings also suggest

judges find common factors.

35
Trends

The study's second purpose was to apply the method
of using ratings of photographs in measuring type and
trends in type to a set of data if relationships between
variations were suitable. Since variance components of
ratings from photographs resembled those when actual
animals were rated,, the second purpose was attempted.
Data collected for the study were reexamined because they
were the only data available.

Average yearly type scores for the complete herd,
Table 14, rise until 1961. Average scores for each judge
follow a similar pattern. Scores for each line, Table 9,
show average ratings of animals selected for milk produc-
tion, line 1, increase until 1959 then are slightly lower
until a peak in 1963. Average scores of control animals,
line 2, are fairly consistent from 1958 through 1964 with
the highest average in 1960. Type selected animals, line
3, average scores are reasonably consistent also, peaking
in 1961. Average yearly type scores in Table 5 for ages
2, 3, and 4 peak in 1961, with yearly changes similar to
those for overall average. Scores for older animals do
not follow this pattern.

Ratings of photographs show average ratings of
the lines followed similar patterns over the years and
rank as expected. Averages of animals selected for type

are generally above other lines while the herd with chance

36

 

 

 

 

Table 14. Overall yearly averages and average yearly type
scores of photographs of Jersey cows for each
judge.

Judge Ratings

Year’ Ave. .per

1 2 3 4 Judge

1956 3.50 2.50 3.00 5.50 3.63 2

1957 7.87 7.65 9.41 8.26 8.30 46

1958 8.64 8.03 9.90 9.18 8.94 104

1959 8.90 8.56 9.66 ‘ 9.17 9.07 108

1960 9.36 8.84 9.43 9.27 9.23 104

1961 9.06 9.22 9.53 9.22 9.26 86

1962 8.61 8.62 8.96 8.99 8.79 100

1963 8.90 9.13 9.32 9.31 9.17 52

1964 8.77 8.92_ 9.17 9.62 9.12 86

 

Scale of type rating 0 to 17, low poor to high excellent;

me an

9, low good plus.

matings generally has the lowest average scores.

of ages 2 through 4 also followed similar patterns.

Averages

These

averages indicate trends in type can be evaluated by

photographs to hold ideals constant during rating. This

does not prevent changes in ideals with time altering

measurement of trends as the measurement is in accord with

the ideal at time of rating.

At another time measurement

:might be by a different ideal and produce a different

trend even though.ideals were fixed during either measure.

ment .

37

Comparing ratings of animals from photographs
with ratings of live animals at the time of photographing
was not attempted in this study. Analyses of variance
and correlations indicate the two ratings would be com-
parable, but further work is needed. Comparison of rat-
ings from photographs taken at the time of classification'
with classification scores may indicate how alike ratings
from live animals and photographs are. Precautions to
limit variation in scores due to raters, time between
ratings, and changes in scale are needed. If ratings
from photographs are comparable with ratings of live

animals, photographs might be useful in type classifying.

CONCLUSION

Separation of variance into its components resemf
)led reasonably well those when actual animals were rated.
Sow accounted for 57% of the variance in this study while
accounting for 55% and 31% in studies with live animals
(8, 12). In all three studies judge and year acCounted
for little of the total variation.

Agreement of judges ratings in this study also
resembled reasonably well analysis with actual animals.
Sorrelations of simultaneous ratings by different judges
ranged from .59 to .77 averaging .68. Correlations of
ratings on live animals rated almost simultaneously by
different judges were .74 at Iowa, .61 to .76 at West
Virginia, and .73 and .76 at Illinois.

Color transparencies to evaluate type, or trends
in type, may be useful. Subsequent rerating of these
same photographs may in time hint of changes in ideals.
If these do not exist, photographs are probably not
1eeded to assure a single ideal. .

I recommend taking color transparencies of anie
nals in breeding projects which seek to monitor changes

in type. Photographs should be taken at least yearly at

38

 

 

 

 

39

the same stage of lactation. Management of animals should
be as constant as practical throughout the study. Ratings
over a short interval reduce changes in scales and ideals
from first to last rating. Ratings, however, are on the
scales and ideals accepted at the time of rating. If
ideals change over time, rating the photographs some time'
after project completion may help eliminate influence from
ideals at time of rating being more like ideals existing
at the project's completion than like ideals existing at

the project's beginning.

 

10.

REFERENCES

Atkeson, G. 1967.‘ Weighting components of type in
classifying Holsteins. M. S. Thesis, Michigan
State University, East Lansing. rm-j

Benson, R. H., W. J. Tyler, and G. Hyatt, Jr. 1951.
Some causes of variation in type ratings of Ayrshire
cows. J. Dairy Sci. 34:502.

Carter, H. W., J. C. Rennie, and E. B. Burnside. 1965.
Causes of variation in type classification data.
J. Dairy Sci. 48:790 (Abstr.).

 

Dayton, A. D. 1969. The effects of inbreeding on
heritable traits in a herd of Jersey cattle.
Ph.D. Thesis, Michigan State University, East
Lansing.

Hyatt, 6., Jr., and W. J. Tyler. 1948. Variation in
type ratings of individual Ayrshire cows. J.
Dairy Sci. 31:71.

Hyatt,G., Jr., W. J. Tyler, and C. T. Conklin. 1949.
The relationship between type ratings of Ayrshire
females as young heifers and as cows. J. Dairy
Sci. 32:375.

Johnson, L. E., and J. L. Lush. 1942. Repeatability
of type ratings in dairy cattle. J. Dairy Sci.
25:45.

Mc Gilliard, L. D., and J. L. Lush. 1956. Changes
in type classifications of dairy cattle. J.
Dairy Sci. 39:1015.

Porter, A. R., J. A. Sims, and C. F. Foreman. 1965.
Dairy cattle in American agriculture. -Iowa State
University Press, Ames.

Specht, L. W., H. W. Carter, and L. D. Van Vleck.

1967. Type classification scores of New York
Holstein Cattle. J. Dairy Sci. 50:1690.

40

41

ll. Touchberry, R. W., and K. R. Tabler. 1951. The.
changes in type ratings of Holstein and Guernsey
cows when rated by the same three judges at two
consecutive times. J. Anim. Sci. 10:1029.

12. Touchberry, R. W., and K. R. Tabler. 1951. The
changes in the type ratings of Holstein and
Guernsey cows when rated by the same three
judges at two consecutive times. Mimeo.

 

 

mII"("1"lelultlmmluujwuunmnnI

93 915