. Hump". y

vu-w..."~‘...

nou‘

 

r‘H E513

Date

0-7639

 

J- IIIIIIIIIIIIIIIIIIII IIIIIIIIIIIII

312_93fl1 106
I 1; 1.5.1:. 49

 

I "P? 0 ..
l". ¥ 3 -.,_ . _
I lsua:£-4:. 3 It. 315.,1‘E'J

Qfliw

LJJUdvg-hrh 3233'

 

 

_"

This is to certify that the

thesis entitled

CONSIDERATION OF MATES' SCORES IN
THE RANKING OF GUERNSEY SIRES FOR GENETIC
ABILITY IN BODY CONFORMATION

presented by
John Paul Walter

has been accepted towards fulfillment
of the requirements for

M.S. , Dairy Science
degree in

 

 

OS’V/(M % Wfl/f

Major professor

7%. 2Q 1982',

 

 

 

MSU

LIBRARIES

 

 

11

RETURNING MATERIALS:
PIace in book drop to
remove this checkout from
your record. FINES wiII
be charged if book is
returned after the date
stamped below.

 

 

 

a9?)

 

CONSIDERATION OF MATES' SCORES IN THE RANKING
OF GUERNSEY SIRES FOR GENETIC ABILITY

IN BODY CONFORMATION

BY

John Paul Walter

A THESIS

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

MASTER OF SCIENCE

Department of Animal Science

1982

ABSTRACT
CONSIDERATION OF MATES' SCORES IN THE RANKING

OF GUERNSEY SIRES FOR GENETIC ABILITY
IN BODY CONFORMATION

BY
John Paul Walter

Body conformation scores of 192,640 Guernsey'cows for
1956-1977 were matched to their dams' scores to investigate
the effect.of'mates' scores on sire rankings based on progeny
records. A chi-square goodness of fit test indicated that
many Guernsey sires had.mates that were non-representative
samples of the pOpulation. Mates' effect on sire rankings,
therefore, was examined for each of twelve type traits.
Sire rankings from two models were compared using two data
sets, one containing all progeny of the sires and the other
containing only each sire's first twenty daughters. Both
models include a random sire factor and fixed age, month,
and herd-year factors. One model contained an additional
covariate of dam's score. Eighty sires having a total of
33,634 daughter first scores were used in the analysis.
Results indicate that the addition of mate's score as a
covariate did not appreciably change sire rankings when all
progeny were included. When only the initial daughters
were used, rankings were changed somewhat more, but with

no improvement in accuracy.

DEDICATION

In gratitude,

to my Dad and Mom

ii

ACKNOWLEDGMENTS

The author wishes to thank Dr. Ivan L. Mao for
his support throughout this graduate program. His
encouragement and advice, as well as the warmth shown by
him and his family, made my effort considerably easier
and more pleasant. My appreciation is also extended to
members of my graduate committee, Dr. J. L. Gill,

Dr. C. R. Anderson, Dr. W. T. Magee, and Dr. R. W.
Erickson for their contributions and advice.

I would like to express thanks to the many friends
within the Department of Dairy Science, faculty and stu—
dents alike, who were always generous with their help
and support.

The contribution of the American Guernsey Cattle
Club, who supplied the data for my study, is gratefully
acknowledged.

And, finally, I would like to thank my parents,
my sisters Mary Pat, Mary Jo, Michelle, Polly, and Carrie,
and my brothers Chuck, Tom, Chris and Matthew, for their

patience and love.

iii

TABLE OF CONTENTS

Page
LIST OF TABLES ..... ...... . ......................... vi
LIST OF FIGURES .................................... vii
1. INTRODUCTION .... ........ . ....... . ......... ..... 1
2. REVIEW OF LITERATURE ........................... 4
2.1 Type Classification of Dairy Cows ......... 4

2.2 Sources of Variation in Body Conformation
Scores: Environmental ............. ..... 7
2.2.1 Herd ..... .................... . ..... 10
2.2.2 Age at Classification .............. 11
2.2.3 Stage of Lactation ..... ............ 12
2.2.4 Season of Classification ........... 13
2.2.5 Year of Classification ...... ....... 14
2.2.6 Classifier ........................ 14

2.3 Sources of Variation in Body Conformation
Scores: Genetic ........... .......... ... 16
2.3.1 Sire .................... ...... ..... 16
2.2.2 Group of Sire ...................... 22
2.2. 3 Relationships Among Sires .......... 22
2.2.4 Cow ........................ . ...... 23

2.4 Relationship of Body Conformation to

Production .............................. 25
2.4.1 Ayrshire ........................... 25
2.4.2 Jersey ...... ..... .................. 27
2.4.3 Brown Swiss ........................ 27
2.4.4 Guernsey ..... ...................... 28

2.4.5 HOlStein ............OOOOOOOOOOOOOOO 28
2.5 Effects of Body Conformation in Herd

Management Traits ....................... 30

2.6 Prior Consideration of Mates in Evaluation
of Bulls .. ..... ...... ........... . ....... 34
2.7 Sire Evaluation for Type .................. 36
2.7.1 BLUP Approach . ............... . ..... 37
3. METHODS AND PROCEDURES ................. ........ 41
3.1 Data ... ................................... 41
3.1.1 Data Source and Preparation .. ...... 41
3.1.2 Merger of Data ...... ...... . ........ 42

iv

3.2 Examination of Randomness of Sires'

3.2.1 Goodness-of—Fit Procedure

Mates

3.3 Mates' Influence on Sire Ranking Using

All Sire's Progeny ......................
3.3.1 Models

2 Editing the Data ...................
3 Absorption of Herd-Year Groups .....
4 Adjustments to Random Portion of
Mixed Model Equations ............
5 Solutions of Mixed Model Equations .
6 Disconnectedness ...................
7 Comparison of Sire Rankings ........
8' Influence on Sire Ranking Using

Initial Daughter Samples ................
3.4.1 Editing the Data ...................
3.4.2 Disconnectedness Among Sires Having
Few Progeny ......................

3.4.3 Comparisons Made Using Results of

4. RESULTS AND DI

Evaluations Based on Few Progeny .

SCUSSION ......OOOCOCOOOOO ........

4.1 Extent of Non-Random Mating for Final

Class ...................................

4.2 Effect of Mate's Scores on Sire Rankings

Based on Many Progeny ...................

4.3 Mates' Effect on Sire Rankings Based on

Initial Daughter Samples ................

4.4 Examinati
Coeffic

5. CONCLUSIONS ......OOOOOOOOOOOOOOOOO0.0.0.0000...

LIST OF REFERENCES

on of Covariate Regression

ients .0......COOOOOOOIOOOOOOOOOO.

Page

43
45

46
46
48
49
49
51
53
54

55
55

55

58

58
60
62
67
70

74

10.

LIST OF TABLES

Page
Summary of Heritabilities for Conformation
Traits in Five Dairy Breeds ................ 17
Genetic Correlation of Official Final Score
to Production in Five Dairy Breeds ......... 26

Tabulation of Frequencies for Final Classifi-
cation of Guernsey Dams .................... 44

Heritabilities of Guernsey Type Traits and
Resultant Variance Ratios .................. 52

Proportion of Sires Having Mates Distributed
Differently from the Population at Various
confidence Levels O.......OOOOOOOOOOOOOOOOOO 58

Standard Deviations of Sire Estimates Using
All Progeny of Sires ....................... 61

Rank Correlations Between Models Including
and Excluding Dam Covariate Using All
Progeny 0f Sires .........OOOOOOOOOOOOOOOOOO 63

Weighted Rank Correlations Between Models
Including and Excluding Dam Covariate
Using Initial Progeny of Sires ............. 65
Weighted Rank Correlations of Both Models
Using Initial Sample of Daughters to Model
Without Covariate Using All Daughters ...... 66

Regression Coefficients for Dam Covariates ... 68

vi

Figure

1.
2.
3.

LIST OF FIGURES

Page

Dairy Cow Unified Score Card ................ 6

Holstein Descriptive Type Traits ............ 8
Portion of American Guernsey Cattle Club

Scorecard ................................. 9

Traits Evaluated for All Animals .... ........ 42

Examples of Differing Distributions ......... 59

vii

1. INTRODUCTION

Sire evaluation has been an essential tool for
dairy cattle selection since 1936, when the United States
Department of Agriculture (USDA) released its first sire
summary for milk production. This first ranking was based
mainly on a survey reporting the percentage of daughters
who exceeded their dams' production, and the percent by
which a sire's daughters' average exceeded their dams'
average, making use of only primitive corrections for herd
differences. Since then, gradual improvements in statis-
tical methods have enabled, and refutations of certain
distribution assumptions have necessitated, changes in
the methods by which sires are evaluated. Also, sires
have been evaluated for other traits, such as body con-
formation and dollar returns from milk and its fat per-
centage.

The advent of artificial insemination (AI) enabled
breeders to use much more intense selection than previously
possible. However, to make Optimum use of AI, genetically
superior bulls must be identified at an early age, thereby
decreasing the generation interval and increasing the rate
of progress. For sex-limited traits, such as milk yield,

this is accomplished by using available information on

1

female relatives, in particular daughters. The purpose of
statistical procedures is to eliminate variation not
caused by differences among sires, and thus to rank the
bulls according to their breeding values.

The evolution of sire evaluation procedure is a
history of recognizing, and adjusting for, simplifying
assumptions such as no herd differences, no selection
among sires, no genetic progress, and others. One recent
consideration which may not hold true is that mates are a
random sample from a static population. Therefore, some
researchers have looked into the inclusion of daughter's
maternal grandsire as a factor contributing to phenotypic
expression of a characteristic. Incorporation of
daughters' maternal relatives in sire evaluation should
have the advantage of reducing the potential source of
bias due to a sire being bred to mates not truly repre-
sentative of the pOpulation. This bias is more likely to
occur in the evaluation of young sires, because they have
few progeny and the inclusion of a dispr0portion of
superior or inferior mates can result in an unreasonably
high or low progeny test. Additionally, inclusion of
relatives on the daughter's maternal side could somewhat
increase the information available on phenotypically poor
daughters, who might have disproportionately few records.

Bias caused by non—representative selection of

mates may be of greater concern when evaluating a bull's

genetic ability for body conformation than for milk pro-
duction. Because sale of milk is the primary source of
revenue for most herds, bulls might be chosen for their
conformational ability only when it is necessary to cor-
rect a type trait within a herd. Mates for these bulls
would generally be phenotypically poorer than if they
were a random sample of cows.
Body conformation records of Guernsey cows were
used to examine three objectives in this study:
1. to determine the extent of non-random
mating for body conformation within
the Guernsey breed;
2. to determine the effect of mates'
conformation scores on sire evaluations
and rankings; and
3. to determine whether the inclusion of
mate's score can improve the ranking

of sires based on a small number of
daughters.

2. REVIEW OF LITERATURE

The body conformation of dairy cows long has been

of interest to dairymen as well as dairy breeders. In

addition to the economic value of type classification in

the promotion and sale of cattle, its relationship to milk

and fat production, herdlife, health, and total merit, to

name a few, make it imperative that cows be logically and

consistently scored and that sires be accurately evaluated

and ranked for genetic merit of type traits.

2.1 Type Classification of Dairy Cows

Over the years, various methods have been proposed

to classify cows in the most objective and descriptive way

possible.

Many methods share several common features,

necessary for widespread acceptance:

1.

Simplicity. To be readily understood, a
classification system must include only a
small number of the almost limitless body
characteristics of a dairy cow. Each trait
should have some economic value to a dairy-
man and be at least moderately heritable,
so that selection can occur.

Objectivity. If cows are to be measured
consistently, the same set of objective
guidelines must be applied, which enable a
trait to be accurately evaluated. This
also implies that traits be sufficiently
variable to ensure that differences are
distinguishable.

3. Logic. If a scoring method is to be
accepted, it must be easily understood.
Most currently employed methods have an
ordinal system of grading, with a limited
range of classification scores for the
inferior through the most favorable animals.
Whereas many different systems of classification
exist today, two of the most common will be described as
examples typical of all. One of the first wideSpread
methods used to classify cattle was through use of the
Purebred Dairy Cattle Association's (PDCA) Dairy Cow Score
Card (Figure 1). Being so widespread, the PDCA scorecard
could be applied across all breeds, and today is the
basis for many classification systems, such as those for
the Guernsey and Jersey breeds. As outlined by the score-
card, cows are evaluated for traits within four major
categories (general appearance, dairy character, body
capacity, and mammary system), each having a maximum
number of points awarded. The sum of the points for all
traits determines an animal's final score, which in an
ideal cow is 100.
The Holstein-Friesian Association of America
(HFAA) makes use of a descriptive method of type classifi—
cation. The HFAA does not define ideals of conformation
traits in general terms, as in the PDCA Score Card.
Rather, traits are judged categorically, from 1 through

5 for most traits. Lower scores indicate more favorable

conformation than higher scores, except for miscellaneous

 

AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAALAAAAAAAAAAAAAAAAAAAA‘AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
YYYYYYYYYYYYYYYYYYYYYYYYVYYYYYYYYYYYYYYYYYYYYYYYYYYYVYYYYYVYYVYYYYYYVYYYYYYYYYYYYYVYYYYVYYY

 

DAIRY COW UNIFIED SCORE CARD

canwgnmdbyInsfiuehodOUanuflaAuanuon."I: Io«ul,anl(amnqbndl99’vdIO?!
Aorond-IhsAmuemsbunHkomnAmanunthfl

 

“WMthIeWeaelhissseree-d

 

Order at abservatIen
I. GENERAL APPEARANCE

(Attractive Individuality saithJeminity. vigor. stretch. scale, harmonious blending of all parts, and
impressive :tyle and carriage. All parts d a cost- should be considered in evaluating a oose's general
appearance

BREED CHARACTERISTICS — (see reverse side)

HEAD — clean cut. roportionate to body; broad munle with large. o n nostrils; strong jaws; large.

brightdeyes; forehead. road and moderately dished; bridge of nose straighI: ears medium sise and alertly

carrie

SHOULDER BLADES - set smoothly and tightly ms: the body

BACK - straight and strong; loin. oad and nearly evel

RCMP — Ion . wide and nearl level from HOOK BONES to PIN BONES: clean cut and free from
patchiness; T L'RLS. high an wide apart; TAIL HEAD. set level with bachlioe and free from coarse.
ness; TAIL. slender

LEGS AND FEET -- bone flat and stro . pasterns short and stro . hooks cleanly moulded. FEET.
short. compact and well rounded with dfl eel and level sole. FOREHLEGS. medium in le h. straight.
wide apart. and squarely laced. HIND CS. nearly perpendicular from hock to pastern. rom the side
view. and straight from rear view

2. DAIRY CHARACTER

(Eeidenee o] milking ability. angularity. and general openness. without mistress; freedom from
coarseness. giving due regard to period of lactation)
NEC -— ong. lean. and blending smoothly into shoulders; clean cut throat. dewlap. and brisket
WITHERS. sharp. RIBS. wide apart. rib bones wide. flat. and long. FLANKS. deep and refined. THIGI‘IS.
incurring to flat. and wide apart from the rear view. providing ample room for the udder and its rear
attachment. SKIN. loose. and pliable

3. BODY CAPACITY

(Relatively large in proportion to else of animal. providing ample capacity. strength. and vigor)
BARREL - strongly su rted. long and deep; ribs highly and widely sprung; depth and width of barrel
tendin to increase towar rear
HEA T CIRTH — large and deep. with well sprung fore ribs blending into the shoulders; full crops;
full at elbows; wide chest floor

4. MAMMARY SYSTEM

(.4 strongly attached. well balanced. capes-ions udder affine texture indicating heat-y production
and a bnslperiod of usefulness)

L' ER — symmetrical. moderately long, wide and deep. strongly attached. showing moderate cleavage
imweend halves. no quartering on sides; soft. pliable. and well collapsed after milking; quarters even y
l ance
IFORE UDDER — moderate len h. uniform width from front to rear and strongly attached
REAlgdUDDER — high. wide. ightly rounded. fairly uniform width from top to floor. and strongly
CHIC
TEATS — uniform site. of medium length and diameter. cylindrical. squarely placed under each quarter.

lamb. and well a ced from side and rear views
AMMARY VE NS - Is e. long. tortuous. branching
"Because of the natural un eveloped mammary system in heifer calves and yearli a. less emphasis is
on mammary system and more on general ap arance. dairy character. and y capacity. A slight
to serious discrimination applies to overdeveloped. atty udders in heifer calves and yearlings. '

 

IO

~

0

131

~
Ml ml «sot °

 

Subscores are not used in breed type classification. TOTAL

100

 

 

 

 

 

V I
.e---—ane---s e—oou --ne—--oea---s “K
I

 

 

. ' ' .. -v. lb
u\ . m . Ir ’53" . "-‘-' a "
uten- , I-.~,,= ~. ~. , ‘ . .6 “A: n
\4 - _ ' t ‘ a 1" ' m!

. 4.}
' . ‘ - arenas-

 

 

 

' 5
It! .v ‘ roe-ransom

IIII CI’S ‘ III! GINO.

 

 

 

AAAAAAA AAAAAAA‘AALAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA‘AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAA ALL A
YYYYVYYVYYYYYYYYYYYYYYYYYYYVYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYVYYYYYYYYYYYYYYYYYYYVYYY

 

 

Figure 1.--Dairy Cow Unified Score Card.

 

traits (Figure 2). The advantage of this method is that
for all traits except final score, no ideal need be used
by each judge against which to compare each cow. Classi-
fiers merely designate the category for each trait which
most aptly describes a cow.

Currently, the American Guernsey Cattle Club
(AGCC) employs in its classification system the scorecard
found in Figure 3. The AGCC Type Committee rates in six
different categories (excellent, very good, desirable,
acceptable, fair, and poor) the thirteen different char-
acteristics of type shown. Final score and rating are
arrived at by balancing the strengths and weaknesses of
each trait, those that indicate a long useful life against
those that impair an animal's usefulness. Descriptions of
desirable prOperties of the thirteen traits can be found
in Sowerby (1978).

2.2 Sources of Variation in Body

Conformation Scores:
Environmental

 

When sires are compared, it is essential to elimi-
nate as nearly as possible all sources of variation which
can affect the evaluation. At the same time, the inclusion
in models of factors which have no effect on a character-
istic is of no use to an investigator, wastes computing
time and core memory, and may actually decrease the

accuracy of comparisons. Of the many factors which have

.maHnuB mama 0>Humwuomoo cHoumHomii.N mudmwm

 

 

own no“ HHnEm 008 m
HQEMHU v
mcuounoa xooz m uanouun 009 m
uanH oou moon 4
use no» mmmH acoum ~ 300: on omoHu HOxpcn ponoam m
muocHsonm pomcHz H msoocoHHoomHz oHnoumooo< m
manna mHnouHmopca m poooHa mHono9Um .cooHo .ocouum H moon ch=
none» aconu mow: v
xoon now 00» none» unmm m mchon m
oHnouomou< w mch on HHHnHooomo .30uuoz e
acoEwonHa nQH: cos» nonon nch m
can .oNHm .sumcoH oHnnanop .nE5Hm H momma mmmcHo>mH .numcmH .nuoH3 auwpoz N
Ho>oH HHuoo: .osz can mcog H mesa
ucoeomHH >u0ncoaman .coxOnm m
m
30H ocupHWme m xoon nexpco cHoH goo: v
men>Hae essence we gone N was ozone 364 m
mcH> m can u .>u0ncmamsn vacuum uncomnm Hoop: snow: can nuocouuo EDHcoz m
. H a. H H cHoH oUH3 .oconun .usmHnuum H xoom
coxoun no omooq m
pococHQ 0cm aouuoz v xoo: can sounoz v
30H m xooc can nopHsocn onumoo m
nucHz one unwed; :H enuncmsumucH ~ coeHa can nauseous suave: ~
mow: .an: .vonoouuo >Hsuwm H Hoop: unom unoco oUH3 .uopHaozm suooem H can unonh
coxonm m
goo: m
omooH no wwwmm M onunoo no cHon v
aoHsn sHuanHm uzuocoH demeanor w oHsum mxomH .mwwmum M
How: 3oHHmnw q
moo» powwow m new 304 m
oHnouowoud N muoHpoEuoucH N
cashew HHo3 .ocouum H Doom mchcoummD H ouauoum
:oHumHuoooo pouow>ounn< opoo awoke :oHumHuonoo vouow>ounn¢ 0000 uHouB

 

 

.Unmomuoom QSHU mHuumU Hmmcuosw cmowuosm mo :oHuHomri.m ousmfim

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ZOHP<U—h_mm<au flm>._. no CREEK I QDHD air—LED >umz¢flac Z<OENZ< £15.

I 3.3-a {so In

8

on

'8

an

«a

a

n.

a.

w.

0'

O

m

h

o

m

w

n

n

I u

t .- -t
:83 all. N H human-h. MUM.“ H l m Slut as .5 a. Elli-l... .22... .3 use:
‘5 ‘82!

I a“! illnfl ...-gels. a6: sea-lie luau“ la“ “

.-II teethinlsm In ......mm Emmiimm HR 4....“ a
rise! donned.» “n... In“ ...ill ..Hnu .un o

10

been studied, some of the more significant are reported

below.

2.2.1 Herd

In an examination of nearly 23,000 appraisals of
all breeds, Van Vleck (1964) performed an analysis of
variance components which demonstrated that herd effects
comprise less than 10% of the total variance for most
traits. Legates (1971) used single evaluations of over
130,000 Holstein cows to obtain herd components of vari-
ance of between 13% and 25% for final score and the four
major categories of traits (general appearance, dairy
character, body capacity, and mammary system). For
descriptive traits estimates of prOportion of variance
caused by differences among herds were 6% to 14%. Carter
et a1. (1965), analyzing the sources of variation of type
scores within Canadian Holsteins, gave a comparable
figure for prOportion of total variance caused by herds
(8%) . Norman and Van Vleck (1972b) also reported that
variation among herds accounted for less than 10% of total
variance in most body conformation traits. They used more
than 16,000 appraisals of Holsteins obtained through the
New York type appraisal program. However, for milk pro-
duction, as well as some health-related traits (such as
incidence of edema) the herd component approached 25%.

Hanson et a1. (1969) obtained 4,656 appraisals of 2,518

11

Holstein cows in 41 Wisconsin herds. They found that for
final score and each descriptive trait, herd accounted for
not more than 10% of the variance. In analyses of Brown
Swiss classification data, Moreno et al.(1979) found that
herd-year variation accounted for 8% to 17% of total vari-
ance. In the Jersey breed the same factors accounted for
14% to 21% of the variation, as reported by Norman and
Cassell (1978). In a recent study of Holsteins, Vinson

et al. (1976) determined that, for 78,151 cows in 2,117
herds, herd component of variance was less than 14% for

all body conformation traits.

2.2.2 Age at Classification

 

Hansen et a1. (1969) found age at classification
to be a significant source of variation among classifica-
tion scores of Holstein cows, in final score as well as
for the four major categories. Rennie et a1. (1971) accounted
for age in their analysis of environmental and genetic
parameters for 51,044 Holstein first calf heifers. He
found age at classification to be a significant factor for
most traits. Norman and Van Vleck (1972a) uncovered large
age differences in about a third of the classification and
management traits they considered, especially incidence of
mastitis, body weight, and depth of udder. A significant
effect of age of classification score was found in New

Jersey Holsteins by Wilcox et al. (1958), and in North

12

Carolina institutional herds of Holsteins by White et a1.
(1967). Based on these reports, Cassell et al (1973)
computed age correction factors for conformation similar
to those already in existence for milk production. These
factors were derived from first available classification
scores of 336,253 registered Holstein cows, classified
between 1967 and 1971. Within the Jersey breed, Norman
(1978) found age effects significant for all traits,
eSpecially on body capacity and dairy character. Mao

et al. (1977) also found significant age differences in
the Guernsey breed, and calculated age-season adjustment

factors to compensate for these differences.

2.2.3 Stage of Lactation

 

The effect of stage of lactation is considered in
analyses of classification scores in several studies.
Hansen et a1. (1969) found stage of lactation highly sig-
nificant in accounting for some of the variance of
Holstein scores. Classification scores appeared to be
somewhat higher if a cow were scored near the beginning
or end of her lactation than during the middle. An inter-
action of stage of lactation with age in the same study
was insignificant. Norman and Van Vleck (1972a) reported
that differences caused by stage of lactation in Holsteins
were smaller than those caused by age, and they uncovered

no age by stage interaction. Udder quality appeared most

13

affected by stage of lactation in their study. Analyses
by Wilcox et a1. (1958) and White et al. (1967) reported
similar significant effects of stage of lactation.

Purohit et al. (1973) found stage of lactation to be sig-
nificant for all traits in a population of 12,030 Canadian
Holstein daughters of 200 AI sires. When evaluating the
sires, records were adjusted for effects of stage.

Norman and Cassell (1978) found that effects of stage and
interaction of age with stage of lactation were small in

the Jersey breed.

2.2.4 Season of Classification

 

Month of classification effect was examined by
Carter et al. (1965) and determined to have a small effect
on final type score of Canadian Holsteins. For United
States Holstein cows, White et al. (1967) reported that
date of appraisal had a significant effect on type classi-
fication scores for many traits. Wilcox et a1. (1958)
published evidence of seasonal variation among type
ratings of Holstein cows, although only two seasons,
Spring and fall, were included in their analysis. In the
Guernsey breed, Mao et a1. (1977) noted definite rises in
classification scores in several months, notably August
and February. It was Speculated that the August increases
were the result of preparation of cows for showing activi-
ties; no explanation was offered for the rise in February

scores .

14

2.2.5 Year of Classification

 

Evidence of trend in classification scores over
time has been moderate in most previous studies. Carter
et al. (1965) found that, for Canadian Holsteins, rounds
(or groups of years) accounted for less than 1% of total
variation between 1961 and 1963. In a later investigation
of Canadian Holsteins by Schaeffer et al. (1978), annual
changes in genetic ability varied from .13% for final
score to 1.28% for general appearance. Their research
used records collected from 1971 through 1975. In United
States p0pulations, Norman and Van Vleck (1972b) reported
that up to 3% of variation in type scores could be
explained by year differences, though for most traits
this figure was below 1%. Studying the Guernsey pOpula-
tion, Mao et al. (1977) discovered a small but definite
upward trend in final score between 1959 and 1971. While
working with Jerseys, Norman and Cassell (1978) found 1%
to 4% of variation in type traits explained by differences

among years.

2.2.6 Classifier

 

Differences among classifiers are an unavoidable
source of variation in analyzing conformation scores. As
early as 1942, Johnson and Lush (1942) demonstrated a
repeatability of type scores of only .34, which varied
little with interval between classifications. Though this

is small, undoubtedly partly because of changes in the

15

animal, the subjectivity of the judges cannot be dis-
counted. Touchberry and Tabler (1951) investigated the
changes in type ratings of Holstein and Guernsey cows when
rated by the same judge at two different times. They
found that differences among classifiers contributed sig—
nificantly to total variance only for general appearance
and body capacity. The effect of time on variability of
conformation score, though statistically significant in
several traits, was small. Investigating appraiser effect
on classification scores of the Iowa State Holstein herd,
McGilliard and Lush (1956) stated that, except when
appraising heifers, the effect of classifier was negli-
gible. Wilcox et al. (1959), using records of 233 cows
scored by 13 judges, found rather large variation between
judges. An analysis of variance was not done, however,
because all effects were treated as fixed. Vinson et al.
(1976) reported that the prOportion of variation caused

by differences among classifiers was small (from 0.7% to
5%) for the first-available Holstein appraisals. However,
herd by classifier interaction appeared to account for
twice the variation of classifier alone in most traits,
perhaps indicating some variation in classifiers' standards

from herd to herd.

16

2.3 Sources of Variation in Body
Conformation Scores: Genetic

 

 

Variation among phenotypic characteristics is of
use to dairy breeders only if sufficient genetic variation
exists. Because of the widespread use of AI, the sire is
the main vehicle by which most genetic improvement is
passed from generation to generation. The fundamental
relationship in the study of dairy populations is that of
parent-offspring, between a sire and his daughters. There-
fore, in analysis of variation among half-sibs, measures
of prOportion of sire variance and estimates of heri-

tability are associated and will be discussed together.

2.3.1 Sire

 

In studying the role of sire in contributing to
total variation, genetic theory holds that, in a half-sib
p0pulation, differences among sires account for one-fourth
of the additive genetic variance, along with small frac-
tions of variances caused by various inter-locus inter—
actions of additive effects. Because heritability (hz)

can be expressed as the ratio of additive to total vari-

O2

ance (-%0, the proportion of variance caused by sires in
o
T

half—sib populations usually is assumed to be one-fourth

of the heritability. Heritabilities estimated by many
researchers for various breeds and traits are shown in

Table 1. The research will be reviewed by breeds.

 

 

 

 

vN. ON. muopHaocm
Hm. NN. 9N. an. ounce
aH. HN. mH. aH. 0N. mN. mm. mH. Home: noon
NH. HN. mH. HN. NN. on. mN. HN. Hoop: ouch
HN. NH. NN. NN. NN. NH. NN. «N. NN. NN. NN. seams» Nausea:
HH. 4N. mH. oN. oH. NH. noon one uoom
em. mN. ON. Hm. NN. He. on. mH. can:
mN. NH. ow. vN. NN. ov. mN. NN. on. MN. vH. auHoonoo >00m
oN. mo. om. oN. mH. mN. oH. oH. oN. on. ON. nouonnono >uHon
SN. mH. NN. HN. NN. em. «N. 4N. mm. NN. NH. mocuunoaaa Huuocou
NN. NH. NN. Nn. nouoouoso pooum
NH. nH. auHHoso Honocoo
no. NN. suHHoso .mcHo> .auuoa
ma. mo. nucoanoouu< Hope:
we. no. oNHm can madam Home:
NN. Nn. nanna can mean
71 SH. Hm. :HOH one oprH:
11 wH. mH. umozu can unopHsozm
NN. on. 3002 can one:
Hm. ON. vN. wN. Hm. an. NN. NN. no. mm. mN. NN. oH. mH. Hm. ououm anHm
H. m. .... N .... H. N H N ... N m a a ..
H 3 r. u s .o o J 1 u. n W a uHous
e o o a s e o a u m r. A o
a u. o a a 3 a n u s e n a u. m
3 a x u r. a 3 s o o u .A a e
I. T. T. s a u 1 u 8
a r. a a e x. a a m. e a
z 3 3 a I. I 3 e 3 u a e s
a 3 T. . a u D. 3 u a
a 3 e o .o 3 e o. e WW He 9
.... T. I P L l .l T. n... P 1
D. . . T. T. T. P . J. . n T. nu D
O T. I . I 6 .... o s . m. u.
I s l I ..L s \l n l u. a
e I 6 ‘I L I 1 I ) 1
l T. 9 6 6 Ts 8 I. T. 6 1
I 6 z ( L 6 o ) 8 \II 6 9
6 9 I E C. I I I H S r. l
L I Ila 6 6 z ( I
L i. i. 9 K" I 6
Ila 0 l‘ S
i. m_
:HounHoz momcuonu anzm moaned oannna<
caonm

 

.mpowum huHoo 0>Hm cw muHona coHuneu0mcoo now noHuHHHnoanom mo muoessmiu.H manta

18

Ayrshire

Freeman and Dunbar (1955) used 1,180 daughter-dam
pairs to estimate heritabilities of Ayrshire body conforma-
tion traits. They obtained an h2 for final score of .31
i .15, while the heritabilities of other traits varied
between .06 for udder attachments to .32 for breed char-
acter and rump and thighs. Several years later, Butcher
et al. (1963) used a larger sample Ayrshire daughter-dam
pairs (8,165), and obtained an h2 of .18 i .05 for final
score. The range of heritabilities was from .05 (udder
attachments) to .27 (head and neck). In that study, the
authors adjusted each record for age, but ignored differ-
ences among years and classifiers. Because of the small
numbers in the Ayrshire breed and the limited use of AI
in the time periods of these studies, an estimate of the
sire component of variance as gé-may not be accurate;
there may be too many full-sib relationships among
daughters of natural service sires. A confounding effect

between natural service sire and herd also may occur.

Jersey

In an early study, Harvey and Lush (1952) reported
an h2 of .14 for final score. Tabler and Touchberry
(1956) used 2,810 daughter-dam pairs of Jersey cows to

obtain twice the former estimate, or .28. Recently,

Norman et a1. (1978) calculated heritabilities from 4,539

19

pairs, appraised between 1968 and 1976. These records
were adjusted by age and stage of lactation, and provided
a heritability figure of .23 1 .03 for final score, with
other h2 ranging from .12 to .26. The latter heritabili-
ties corresponded to a prOportion of sire variance of 3%

to 6.5%.

Brown Swiss

In two major studies, estimates of genetic
parameters for body conformation in the Brown Swiss breed
were reported. The first, by Johnson and Fourt (1960),
included type records from 3,161 daughter-dam pairs,
appraised between 1950 and 1953. Their estimates among
traits varied from .19 for feet and legs to .36 for rump,
with an h2 of .35 for final score. Standard errors of
these figures were .05 to .07. The second study was com-
pleted by Moreno et al. (1979), using 12,838 records in
an analysis of variance. Heritabilities of .24 (feet and
legs) to .43 (final classification) were reported. These
investigations indicate that generally between 5% and 10%

of the variation of classification data is attributable

to sire effect.

Guernsey
A report by Berousek et al. (1959), based on
daughter-dam regression within sire and herd of 1,981

pairs of Guernsey cows attributes to final score a

20

heritability of .28. Dairy character was found to be the
least heritable trait, while teat placement had the highest
h2 of .30. Mao et al. (1977) reported heritabilities cal-
culated from results of an analysis of variance on more
than 200,000 Guernsey appraisals. They found a similar
heritability of final score, .27, and heritabilities for
each trait similar to those reported by Berousek et al.

(1959). These h2 evince pr0portions of variance caused

by variability among sires of 4% to 8%.

Holstein

More reports of heritabilities of conformation in
Holsteins have been published than for any other dairy
breed. Mitchell et al. (1961) used 14,727 daughter-dam
pairs from Wisconsin herds to estimate heritabilities.
Dairy character had the lowest hz, .09, whereas final
score had the highest, .20. Hanson et al. (1969) also
used animals from Wisconsin herds to estimate heritabili-
ties. From 4,656 records on 2,518 cows, they obtained a
heritability of final score of .26, with extremes of .20
for dairy character and .27 for mammary system. Least
squares procedures were used to obtain variance components
in this investigation. Sire differences accounted for up
to 7% of the variance.

Wilcox et al. (1962) analyzed the conformation

records of 233 daughters of 51 bulls in one herd, as

21

appraised over a period of 13 years. A heritability of
.24 was computed for final score, comparable to previous
estimates. Heritabilities of .50 and .66 were reported
for dairy character and body capacity, respectively,
figures which were well above any estimates reported
previously for any breed. These high estimates may have
been caused by sampling errors resulting from the small
amount of data considered. Legates (1971) calculated
variance components on single records of 44,030 three
year old Holsteins. He reported a wide range of hz, from
.58 for final score to .19 for fore udder and rear udder.
Cassell et al. (1973) reported heritabilities
between .19 (for dairy character) and .31 (for final
score). These were computed by daughter-dam regression
of 30,714 pairs of Holstein cows. From Canadian Holstein
type data, Schaeffer et a1. (1978) analyzed 182,608
classification records of daughters of 10,360 sires. The
highest heritability calculated from variance components
was .34 for rump; the lowest was .11 for feet and legs.

An h2 of .31 was published for final score.

Using records from all breeds, Van Vleck (1964)
found that the fraction of variance caused by variation
among sires was less than 10% for most of 34 traits. This
agrees with results previously cited, in which very few

heritabilities exceed .40.

22

2.3.2 Group of Sire

 

By nesting effects, sires within genetic group in
a model, additional variation can be eliminated. However,
to date no studies have been reported in which this reduc-
tion has been quantified using conformation records.
Using milk production data, Rothschild et al. (1976)
found that consideration of genetic groups of sires can
remove a substantial portion of bias favoring older sires
as well as natural service sires. Using simulated pOpula-
tions, Tong et al. (1980) have shown that ignoring genetic
groups of sires can increase the variance of a sire's

predicted breeding value by 9% to 12%.

2.3.3 Relationships Among Sires

 

When sires are evaluated using mixed model tech—
niques develOped by Henderson (1974), provisions exist for
the inclusion of variance-covariance (VCV) matrices for
random effects. If sire is the only random factor in the
model, represented by a sample of n sires, the dimension
of its VCV matrix is n_x 2. The VCV matrix can be
expressed as §_A 02, where §_is a scalar and A is a matrix
of genetic relationships among sires. Residual variance
is deonted by 02.

To incorporate the VCV matrix for sire, the rela-
tionship matrix A must be inverted, which can be unwieldly

and expensive. Henderson (1975a, 1975b) introduced a

23

method of directly computing the desired inverse of the
relationship matrix. He also pointed out some of the
potential advantages of incorporating these relationships:

1. improvement in the accuracy of prediction,
particularly for sires with few progeny;

2. need for fewer groups to account for
genetic differences among sires; and

3. earlier evaluation of sires due to the
use of additional information not pre-
viously included in the evaluation
(Henderson, 1975c).

In a comparison of evaluations of bulls grouped by
year of entry-into-service and bulls related by a VCV
matrix, Kennedy and Moxley (1975) found correlations of
.88 to .93 for evaluating Predicted Differences (PD's) of
milk, fat, and protein production. They also showed the
variance of prediction error to be 18% larger when con-

sidering genetic groups than when considering sires'

relationships, all other factors remaining the same.

2.3.4 Cow

 

No investigations have been found examining the
effect of cows' multiple classification records on sire
evaluation for type. There are several, however, which
use milk production to examine that effect. The flexi-
bility of mixed model equations (MME's) allows as many
lactations to be included as are desirable, but there may
be some question whether inclusion of several or all of a

cow's records in her sire's evaluation is necessary or

€V€I

24

even beneficial to accurate results. Ufford et a1.
(1979), using records from the Ayrshire, Jersey, Guernsey,
and Brown Swiss breeds, found the variance of estimated
effects of sire reduced 7% to 14% by including all lacta-
tions over using only first records, with AI sires. The
difference was much greater (20%-24%) for sires used in
natural service. These reductions enable a decrease of
40% in the number of daughters needed to prove a sire at
a given accuracy if all lactations are used rather than
only the first. They indicated that, because of the
improved accuracy of evaluation, potential genetic progress
in the selection of bulls could be 10%-15% greater when
all lactations are included.

However, caution is also advised when using more
than the first lactation. Computational difficulties may
offset the theoretical advantage of including all records.
Further, Wickham and Henderson (1977), comparing sires'
evaluations based on their daughters' second lactation
and the same daughters' first lactations, found a rank
correlation of only .77; this could indicate the presence
of some selection bias from including second lactation

daughters.

25

2.4 Relationship of Body Conformation
to Production

 

 

It has long been of interest to breeders whether
selection of dairy cattle for production traits influences
the body conformation of future progeny. Many studies
have been conducted to determine the genetic relationships
of conformation traits to milk production in the Jersey,
Guernsey, Brown Swiss, and Holstein breeds. In the
Ayrshire breed, butterfat production's relationship to
conformation has been emphasized in studies. These
investigations will be reviewed by breeds. Table 2

summarizes the work discussed in the following sections.

2.4.1 Ayrshire

 

Studies of the butterfat production-type score
relationship within the Ayrshire breed have been incon-
sistent. Tyler and Hyatt (1948) found a genetic correla-
tion between butterfat production and final type score of
.16, using intrasire regression of daughter on dam for
5,177 pairs. Using the same method, Freeman and Dunbar
(1955) obtained a correlation of -.52. They made use of
only 729 daughter-dam pairs in 165 herds, and although a
standard error was not reported, it may have been larger
than those found in comparable studies, due to the fewer
pairs studied. This would make their high negative esti-
mate somewhat more unreliable than other estimates within

the breed. Butcher et al. (1963) reported a correlation

26

.coHuosooum HHHE ou wmuMHmHHoo

.coHuonooum umm o» mum mCOHumHmHHoo .muwnmn>¢ may you
mH whoom Hmch .mHHan>< ummoxm mommun HHm Homo

 

 

Nm. hamH on: wmmzmmoo
NN.- aka .H8 um emnunmuo
NN.: oemH .Hm um uumum>m
mo. NomH .Hm um xooHHz
mN.u vo.n HomH .Hm um HHmnoqu
NH. mmmH .Hm um nmuumo
oo. HmmH sunmnnosoe szEmHom
hm. ommH .Hm um HomH> cm>
«N. ommH punch can acmcnon mmHsm 230mm
HUHmhuocwzm. mN. nmmH mummmom can mGMHH
No. mmmH Hummnnosoa can HmHnma
mH. NmmH smug new Nm>umm Hammmn
No. mmmH .Hm um umnousm
Nm.l mmmH HMQCfim flfim SMEmmHh
oH. maaH Hum»: new umHme mmHmmmwa
GOHuMHmHHoo Ham» mumnoummmmm woman

 

muHmo 0>Hm CH GOHuoscoum ou muoom

m.mvmmnm

HMGHE HMHOHMMC MO GOHHMHOHHOU UHfiUCGUIlcN WdeB

27

of .02 from 8,165 daughter-dam pairs in 2,924 sire-herd
groups. Other type traits varied in their relationship
to butterfat production, having correlations ranging from

-.15 to .16.

2.4.2 Jersey

 

Harvey and Lush (1952), using daughter-dam
regression techniques, obtained a genetic correlation
between final type score and milk yield of .18 in Jersey
cows. Tabler and Touchberry (1956), with 2,810 pairs of
Jersey daughters and dams, estimated the same correlation
at .07. More recently, Liang and Fosgate (1967) used a
multiple regression technique on a herd of 382 Jerseys to
obtain a phenotypic correlation of .25. They computed
other phenotypic correlations to milk production, which

ranged from .12 for fore udder to .32 for dairy character.

2.4.3 Brown Swiss

 

Using 3,161 daughter-dam paired records supplied
by the Brown Swiss Cattle Breeders Association, Johnson
and Fourt (1960) obtained a milk-final score genetic cor—
relation of .24. The lowest genetic correlation to milk
production they reported was .21 for rump; correlations
ranged as high as .48 (for rear udder). Van Vleck et al.
(1980) analyzed nearly six thousand Brown Swiss records
with a two-way, cross-classified model (herd-year and sire

factors). They reported a genetic correlation of .27 for

28

final score and milk production. The lowest genetic cor-
relation reported was .03 for milk production with feet
and legs, while the highest was .29 for milk production

with rump.

2.4.4 Guernsey

In work with Guernsey cattle, Mao et al. (1977)
analyzed 54,638 type-production records. They found the
genetic correlation between final score and milk production
to be quite high, .46 to .63, depending on the year of
classification. Similar high correlations with milk pro-
duction range from .83 for dairy character to .25 for
mammary system. All the cows were scored on only one
classifier, and it was speculated that his philosophy of
a desired positive relationship between conformation and
production may have subjectively affected type evaluations
in favor of higher producing cows and thus led to high

correlation values.

2.4.5 Holstein

 

Records of 187 daughter-dam pairs in the Iowa
State University Holstein herd provided data for a study
by Touchberry (1951). He reported zero correlation between
final score and milk production, although he warned that
with the relatively small number of records, sampling
errors could mask any correlations. Using 8,287 Canadian

Holstein type and production records, Carter et al. (1965)

29

found genetic correlations with milk production of from
-.06 for body capacity to .49 for dairy character. Final
score and milk production were correlated at .12. Mitchell
et al. (1961) divided cows into three levels of milk pro-
duction, then calculated correlations of various type
traits to production of cows at each level. By far the
highest correlations were for dairy character with pro-
duction, at .61 to .82. Dairy character was correlated
with fat production at .34 to .84. They stated that
progress in level of milk production would be one-third
to one-half as fast if selecting for dairy character alone
than if selecting for milk production, based on their
analysis of the records of 14,727 Holstein cows.

With 671 lactation records and 2,272 type classi-
fication records of 233 Holsteins from one herd, Wilcox
et al. (1962) found a correlation of .08 between final
score and milk production. This was the lowest correla-
tion they found of any type trait with milk production.
The highest was .16, for mammary system. More recently,
Everett et al. (1976) evaluated 2,390 sires used in
artificial insemination in the northeastern United States
using the milk and fat production records of their 558,654
daughters. Solutions obtained were correlated to Pre-
dicted Difference Type (PDT) of the same sires, obtained
from the Holstein-Friesian Association of America.

Genetic correlations were measured by a formula develOped

30

by Calo et a1. (1973). Predicted Difference Type was
found to be negatively correlated to sire solutions for
both milk and fat production. Schaeffer and Burnside
(1974) examined PDT and Predicted Difference Milk (PDM)
computed from Canadian Holstein data and found a much
smaller negative correlation between the two of -.05.

In an often cited study, Grantham et al. (1974)
obtained genetic correlations between PDM, as computed by
USDA and included in the USDA sire summaries, and final
score as calculated by daughter averages. More than
750,000 official Holstein cow appraisals were used to
obtain the daughter average for body conformation traits.
Genetic correlation of PDM with final score was -.23 for
sires having at least 20 daughters in 10 herds. Correla—
tions of PDM with scorecard traits ranged from -.22 to
-.24, except for .41 for dairy character.

2.5 Effects of Body Conformation on Herd
Management Traits

 

 

Though undeniably a major consideration, the rela-
tionship between conformation and production is not the
only important correlation studied when evaluating the
effect of breeding for body conformation. Its relation-
ship to traits of herd management, such as herdlife or
total merit, has also been investigated.

In a study of six institutional herds, Berger

et al. (1973) found that type score ranged between

31

one-third and nearly twice as important as milk yield in
determining herdlife. This illustrates the significant
role of type in influencing management decisions. They
also reported that, if one were to construct an index of
type and production designed to predict herdlife, the
efficienty of selection for lifetime production on that
index rather than production alone would be 31% to 52%.
O'Bleness and Van Vleck (1962) investigated the reasons
for disposal of dairy cows from New York herds and dis-
covered that two to four percent of disposals were for
undesirable dairy type. This figure may be somewhat mis-
leading, however. Udder trouble and mastitis, grouped
together in the survey as a reason for disposal, accounted
for 14% to 20% of the diSposals within a herd. Unfortu-
nately, the proportion of these animals without mastitis
but having poorly conforming or severely broken-down
udders is not known. In a later study, Van Vleck and
Norman (1972) found udder (22%), inabilities or disease
(8%), and type (3%) all to be significant reasons for dis-
posal of a cow from a herd. The survey fails to distin-
guish, however, whether the cause of an udder problem or
an inability or disease, is originally due to some short—
coming in the animal's conformation. This may under-
estimate the proportion of diSposals attributable to
faulty type. Subsequent analysis of these data indicated

that few traits measured before 49 months of age have any

32

significant value in predicting the eventual reason for a
cow's disposal. These results, when coupled with the
medium to low heritability of many type traits, indicate
that selection for longevity based on an index of con-
formation scores in first lactation would have only
limited effectiveness.

Studying the daughters of AI sires, Schaeffer and
Burnside (1974) reported significant correlations of sire's
PDT to survival rate of two year old daughters to year
three (.34) and year four (.31), among Canadian Holsteins.
These figures are only slightly lower than those for PDM
(.39 and .44).

Honnette et al. (1980), studying the records of
34,675 Holsteins, concluded that selection for conforma-
tional improvement in udder traits was consistent with
objectives of increasing herdlife and lifetime milk. They
found udder traits, as well as miscellaneous traits such
as winged shoulders, to exceed other traits by far in pre-
dictive value for lifetime milk yield and herdlife. A
correlation between first type score and herdlife of .20
was determined by Specht et al. (1967), although as a
source of variation the effect of first score was rather
minor.

In close relationship to longevity is the value of
type in reducing labor or health costs. Blake and McDaniel

(1979), working with 522 Holstein and Jersey cows, found

33

significant positive correlation between the conformation
scores of mammary traits and decreased cost of milk harvest
in Holsteins. A similar correlation of Jersey cows, how-
ever, did not exist, perhaps owing to the somewhat smaller
scale of the Jersey mammary system. The predictive value
of these correlations, which range as high as .35 for the
correlation of volume of rear teats to average flow rate,
is rather low.

Gilmore and McDaniel (1977a) develOped an annual-
ized income model including such factors as values of
calves, milk revenue, health costs and feed costs. They
also calculated the relationships between several factors
and annualized income, including milk revenue, mastitis
cost, dairy character, and other variables. A correlation
of .25 was found between dairy character and annualized
income. In addition, dairy character was found to be
nearly half as important as milk production in first
lactation for predicting annualized income, and about 40%
as important as fat in first lactation. In their study,
other type traits did not significantly affect annualized
income. Gilmore and McDaniel (1977b) also studied the
association between milk yield, type score, and health
costs. A correlation of -.14 was reported between first
mammary system score and overall health cost.‘ These

results suggest that costs of treatment of mastitis or

34

other health problems can be decreased by improving
mammary conformation.

It can be seen, on review, that many studies have
shown a positive value for conformation by improving pro-
duction and herdlife and decreasing costs associated with
dairy production. A consideration for type score improve-
ment through breeding, therefore, would be of benefit to
producers of dairy cattle in increasing production and
income, and decreasing some related expenses.

2.6 Prior Consideration of Mates in
Evaluation of Bulls

 

As early as 1944, Lush (1944) determined the
expected optimum emphasis on sires' mates' records in
their proofs. However, this was prior to the advent of
widespread AI, and most sires were bred to relatively few
mates. Inclusion of any additional information to a sire's
proof was of great benefit to more accurate evaluation,
and Lush (1944) reports an expected genetic improvement
from sire selection of 12%-20%.

As AI became widespread, little additional benefit
was seen from including a sire's mates in his evaluation.
In one of two studies reported in 1965, Bereskin and
Freeman (1965) conclude that, if deviations from contem-
porary herdmates are used to evaluate bulls, little addi-
tional information is gained by including the records of

mates in bull proofs. Miller and Corley (1965) found that

35

rank correlation between PDMs calculated with and without
mates' records was .998. Sires in this study had an
average of more than a hundred daughters each, and both
daughter and mate records were weighted as a function of
their respective number of herdmates. They also concluded
that inclusion of production data of sires' mates' does
not alter the relative standing of individual bulls.

Hillers and Freeman (1966) investigated sources of
genetic error in sire proofs for production traits. They
found significant differences among sires in their mates'
level of milk and fat production when ten mates were con-
sidered, but not when considering twenty or forty mates.
They asserted that this bias is relatively unimportant in
the selection of sires, but in certain situations it could
cause inaccurate selections.

Bodoh and Shook (1972) computed mean bias of 13.2
kg milk in Sires' PDMs due to non-random selection of
mates. Rank correlations between PDMs corrected and
uncorrected for this bias were .996, for 126 Holstein bulls
with published proofs. Also, they concluded that bias due
to non-random mating of sires for milk production was too
small to consider at that time.

Everett et al. (1979) designed a model wherein the
effect of non-random mate selection was not directly
measured, but was determined through the genetic value of

mates' sires. The sire solutions from two models were

36

compared, one including maternal grandsire (MGS) effect
and one excluding it, using 808,796 production records
from all five dairy breeds. Rank correlations ranged from
.91 for both the Ayrshire and Guernsey breeds when all
sires are evaluated, to .99 when only sires having more
than a hundred daughters in the Holstein, Guernsey, or
Jersey breeds were evaluated. They conclude that includ-
ing MGS in the sire evaluation model is of benefit both
in decreasing prediction error variance and reducing bias
in young sire proofs caused by non-random mate distribu-
tion of young sires. They also point out a 25% increase

in computing time if the model includes MGS.

2.7 Sire Evaluation for Type

Many methods have been devised to evaluate sires
for genetic ability for body conformation traits. The
simplest may be the average score of a sire's progeny, or
Daughter Average. Daughter Average takes into account no
herd differences nor environmental or genetic differences
over time, and assumes a random distribution of sires'
mates. It is also subject to sampling error by failing to
account for the number of daughters of a sire. Obviously,
the evaluation of sires by this method is subject to many
inaccuracies when any of its necessary conditions is

violated.

37

A second method has been used by the Holstein-
Friesian Association of America to determine Predicted
Difference Type (PDT). Predicted Difference is the
expected difference between the performance of a sire's
daughters and the daughters of sires of breed average at
some base period, under similar herd management conditions
(Vinson, 1981). The method uses the formula:

PDT = blfiefi - .15(D¥B)] where b is the repeatability of
the PDT (regression of future on past daughters), P is

the daughter average for final score adjusted for age,

B is the breed average adjusted for age, .15 is the cor-
relation between the classification scores of daughters
and dams, and D is the average dam score, adjusted for
age. While this method accounts for a sire's mates and
his number of daughters, it cannot adjust for herd dif-
ferences nor variation in scores over time due to environ—

mental and genetic changes.

2.7.1 BLUP Approach

 

The shortcomings of both of these methods can be
overcome using a method developed by Henderson (1950,
1966). Originally called Maximum Likelihood technique,
this method was later termed Best Linear Unbiased Predic-
tion (BLUP). However, it was almost two decades before
the development of improved computers made his method
cost-efficient for large dairy populations (Henderson,

1974).

38

Using this procedure, as many fixed and random
effects can be included in a model as are desired or
computationally feasible. Mixed model equations (MME's)
are constructed, and adjustments are made within the
random effects for their variance-covariance structure.
Solution of the MME's yields constants for every level of
each fixed effect in the MME's, as well as estimates for
the random factors. Three investigations will be reviewed
which use BLUP techniques to evaluate sires for type
traits.

Norman et al. (1979) evaluated 726 Jersey bulls
for conformation and then examined various Production -
Type Indices (PTIs) for recommendation to the American
Jersey Cattle Club. Their model contained fixed effect
for herd-year sub-classes and genetic groups for sires,
and random sire, herd by sire, and cow effects. Although
they did not state, they assumed that categorical type
scores have an underlying continuous relationship to the
cow's conformation. The sire estimates obtained, when
added to group constants calculated, were PDTs, which were
correlated to prior estimates of the bulls' genetic ability
based on Daughter Averages. These correlations ranged
among traits from .48 to .76, with most between .6 and .7.

Predicted Difference Milk (PDM) and Predicted
Difference for Fat Percentage (PD%) were obtained for all

sires, and used to calculate a Predicted Difference in

39

Dollar Income (PD$). Various PTIs were computed using
combinations of PD$ and PDT. They investigated the effect
of the PTIs on expected genetic improvement for milk pro-
duction, and noted very small differences between expected
PDMs for weights of 3:1 and 1:0 (PD$:PDT). Desiring to
improve milk production as well as maintain adequate con-
formation within the breed, they recommended relative
weights of 3:1 in calculating PTIs for Jersey sires.

A study of Brown Swiss cattle by Karner and Van
Vleck (1980) presented several differences from the afore-
mentioned study. The most striking is their treatment of
categorical data without assuming continuity. The results
are predicted categorical frequencies on every trait for
each sire, rather than PD's for each sire. The researchers
evaluated 712 Brown Swiss bulls for conformation in twelve
descriptive traits with a model having random herd by year
subclass and sire effects, and fixed sire by category sub-
classes. The data consisted of the official classification
scores of 12,838 Brown Swiss cows, younger than 47 months
of age when classified between 1971 and 1976. Each sire
was required to have daughters in more than one herd. No
relationships were considered between sires, and herd by
sire covariances also were assumed zero. One major diffi-
culty was the inclusion of categories with very low fre-
quencies, causing computing difficulties due to core

storage limitations. Therefore, categories occasionally

40

were merged together to yield a new combined category with
greater frequency of occurrence. It was recommended that
a comparison be made between the effectiveness of this
system and that of linearized systems of scoring in pre-
dicting sire genetic values for conformation.

Schaeffer et al. (1978) evaluated more than 14,000
Canadian Holstein sires based on the records of 230,000
cows and records on 180,000 of their dams. Fixed factors
in the model included age at classification, year, classi-
fication of the dam, and genetic group of sires. The only
random factor in the model was sire effect. Though most
studies have found herd effect to be a significant source
of variation, herd effect was not included in this study.
This may result in somewhat greater variance of prediction
error than if it were included.

Classification of the dam was included as a cate—
gorical variable in the model. The dam's classification
for final class was used for the analysis of final score
and final class, and for all other traits the correspond-
ing classification of the dam for the same trait was used.
The inclusion of this factor would adjust resulting sire

proofs for any differential mating which might exist.

3. METHODS AND PROCEDURES

3.1 Data

 

3.1.1 Data Source and Preparation

The data, as supplied by the American Guernsey
Cattle Club, consisted of 301,321 records of body conforma-
tion scores. The appraisals were made on purebred Guernsey
cows in United States herds by ten classifiers, between
1956 and 1977. In addition to conformation data, each
record contained identification numbers of the appraised
cow, her sire and dam, the herd in which she was classified
(several cows were classified in more than one herd at
different times) and the appraiser. Also included in each
record were the cow's date of birth and the date of the
appraisal.

All cows were appraised for twelve conformation
traits, as listed in Figure 4. The evaluation of three
other traits was either discontinued or begun within the
period covered by the data. These were shoulder (discon-
tinued in 1977), suspensary ligament (begun in 1973), and
stature (begun in 1972). For each trait, every cow was
assigned one of six classifications: excellent, very good,
desirable, acceptable, fair, or poor. A numerical final

score between 40 and 99 was assigned each animal, based on

41

42

the evaluations of its component traits. This numerical
final score determined a cow's final classification. If
its score was at least 90, a cow received a final classi-
fication of excellent; if at least 80 but not more than 89,

very good; and so on.

 

Final Classification Fore Feet and Legs
Breed Character Rear Feet and Legs
Dairy Character Rump

General Appearance Fore Udder
Mammary System Rear Udder

Chest and Body Teats

 

Figure 4.--Traits Evaluated for All Animals

3.1.2 Merger of Data

 

To incorporate the relationship between a cow's
body conformation with that of her dam, the records of
both animals were paired. This merging was enabled by
sorting the data into two separate files. The first was
sorted by dam identification number, while the second was
sorted by cow number and, within cow number, by date of
classification. Often a cow in this second file was

appraised more than once; when merging a record of such a

43

cow to those of its daughters, the appraisal was used
whose date was closest to the date of the breeding which
produced each daughter. Use of this criterion for pairing
maximized the resultant number of daughter-dam records,
and resulted in 192,640 paired records from 132,318
daughters of 14,310 sires.

3.2 Examination of Randomness
of Sires' Mates

 

Before studying mates' effect on sire evaluation,
the extent of non-random mating for type was examined. To
determine the degree of non-random mating, the distribu-
tions for final classification of sires' mates and that
of the dam population were compared, using a Chi-square
goodness of fit test. If few sires were bred to biased
groups of mates, consideration of mates in sire evaluation
for final class might be unnecessary. Distributions in
this test were Specified by frequencies of animals' final
classifications into each of the six categories previously
described. A tabulation of the paired data file was made
for dams' final classification, and the frequencies for
each category are shown in Table 3. Similarly, the fre-
quencies of each sire's daughters' classifications were
tabulated, as well as the number of his progeny in the
paired data.

Broffitt and Randles (1977) published formulae to

determine the minimum sample size necessary to obtain a

44

TABLE 3.--Tabulation of Frequencies for Final Classifica-
tion of Guernsey Dams.

 

 

Code Category Frequency
1 Excellent .0319
2 Very good .4831
3 Desirable .3664
4 Acceptable .1069
5 Fair .0114
6 I Poor .0003

 

valid goodness-of-fit test, for given expected frequencies.
This would be interpreted as the minimum number of a
sire's daughter appraisals needed, given the parameters

of the dam population. For all six categories, solution
of the Broffitt-Randles equations indicated that only
sires with more than a thousand daughter appraisals could
be included in the test; there were only five such sires
in the paired data. However, when the fair and poor
categories were combined, yielding five categories, the
necessary number of daughter appraisals was reduced to 40,
Because more than 300 sires each had at least 40 daughters,
addition of the two least favorable categories was per-

formed for the daughters of these sires. Daughter records

45

for all sires having fewer than 40 progeny were eliminated

from the data.

3.2.1 Goodness-of-Fit Procedure
For the remaining data, the Chi-square goodness-
of-fit test was performed for each sire; a test statistic,

g, was obtained:

 

where n is the number of the sire's daughter appraisals,
Oi is the observed frequency of a sire's daughters being
classified in category i, and Ei is the frequency of the
dam population in category 1. Only five categories were
used, as a result of combining the fair and poor classes.
A level of confidence was determined by locating
q in a Chi-square table, with 4 degrees of freedom. For
each of the 306 sires having at least 40 daughters, a
indicated the probability of Type I error when stating
that a sire's distribution of daughter appraisals for

final class differs from the dam pOpulation distribution.

46

3.3 Mates' Influence on Sire Ranking
Usinqull Sire's Progeny

3.3.1 Models

 

Two models were used to study the effect of a
sire's mates on his evaluation for each of the twelve
body conformation traits.

The first model can be expressed as:

+ HY +

= u + Si + Aj + M 1m eijklmn.

Yijklmn k

In this equation, is a discrete classification

Yijklmn
score for a cow on one of the traits. These scores

varied from 6 (excellent) through 1 (poor), and were
assumed to represent an underlying continuous distribution
of phenotypes within each trait.

The factors included in the model to explain

variation in Y.. were:
ijklmn

l. p, the pepulation mean;

2. S-, the effect of the i£2 sire. It is con-
Sidered a random effect with mean zero and
homogenous variance,02. Covariances between
sires were assumed zero.

3. A-, the effect of the jEh age group at
classification. It is considered a fixed
effect, with eight classes. The upper limits
for these categories, in years, were: 3,4,5,
6,7,8,10, and 20. Each cow was included in
exactly one of these classes;

4. Mk, the fixed effect of the K-EE month of
classification, with k ranging from 1 for
January through 12 for December;

47

5. HYlm: the fixed effects of the lEB herd, mEB

year group of classification, and the inter-
action between the two. Years from 1956
through 1973 were grouped into six sets of
three, while the final four years formed the
seventh group. These groupings were used to
account for any trends in scores over time,
while resulting in fewer empty subclasses
than if single years had been used to form
22 year groups; and

6. eijklm , the random residual effect unique
to each observation. This factor was assumed
to have mean of zero and homogenous variance
of oé, with zero covariance between residuals
of any two observations.

This model will be referred to as NOCOV.

The second model used in this study was:

Yijk1mn= “+ Si +Aj +Mk+HY1m+bDijklIm + eijlkmn

Each factor represented in NOCOV was included in
this model, which will be termed WITHCOV. All definitions
of these factors and assumptions pertaining to them are
unchanged. In addition, consideration was given to a

covariate of each cow's dam score, for the trait

Dijklmn'
being considered. This factor was described in the same

method, and subject to the same assumptions, as Yijklmn’

discussed above in this section. The covariate is

incorporated into WITHCOV as a product of D1 and a

jklmn
regression coefficient, b.

48

3.3.2 Editing the Data

Before analyses by these two models, the paired
daughter-dam data file was edited. Because body conforma-
tion records made by animals less than thirty months of
age are not likely to accurately reflect genetic ability
for mature conformation, these appraisals were eliminated.
Because use of only first records would reduce permanent
environmental effects and eliminate the need for a cow
factor in the model, all second and later appraisals were
removed from the data.

From the remaining records, 80 sires having the
largest number of appraised daughters were tabulated.
The progeny of these sires were selected to make up the
new set of data, called ALL. In this file were 33,534
daughter records, comprising 3,454 herd-year groups. Each
sire had at least 140 daughters in the data. Limiting the
records to progeny of these 80 sires served two purposes:
it restricted the number of equations to be solved to a
reasonable size while maximizing the amount of information
per sire, and the need was foreseen to use small, equal-
sized subsets of progeny from each sire in the later
evaluations. Use of daughters of sires with the most
progeny would produce meaningful contrasts between results
of the current evaluation and evaluations to be described

later.

49

3.3.3 Absorption of Herd-Year Groups

Obtaining solutions using these data for every
level of all factors would necessitate solving a system
of 3,553 equations for NOCOV model and 3,554 equations
using the WITHCOV model. To reduce the number of equations
to a more manageable size, the nuisance factor of herd-
year group was absorbed into the remaining terms as
follows: after sorting the data by herd and year within
herd, the mixed model equations (MME's) were constructed
using the records of one herd-year group at a time. After
all of its records were read, the nuisance factor was
immediately absorbed into the rest of the equations. In
this way, it was never necessary to store more than one
herd-year group at a time in computer core memory, nor
was execution time spent to obtain solutions for 3,453
levels of a nuisance factor.
3.3.4 Adjustments to Random
Portion of Mixed Model Equations

In solving mixed model equations, the inverse of a
random factor's variance-covariance (VCV) matrix must be
added to the portion of the MME's corresponding to that
factor. After some manipulations of MME formulae using
matrix algebra, it can be shown that this addition is some-
what simplified when all random factors are assumed to have
homogeneous variance and no covariance, as in this investi-

gation. Because sire is the only random factor in MME,

50

only diagonal terms of the sire portion of MME need to be

adjusted by the addition of a variance ratio,

2
0
error

2
sire

o
Mao et al. (1977) determined the heritabilities (hz) of
each trait, based on a subset of the same Guernsey popula-
tion.
In half-sibs from randomly mated populations,

genetic theory indicates that 02 is equal to one-quarter

sire
of the additive genetic variance (assuming zero epistatic

2 + 02 expresses the total pheno-

r'a ce a d .
va 1 n )' n OSire error

typic variance. Because the ratio of additive genetic

variance to total phenotypic variance is the definition of

heritability in the narrow sense, and assuming any existing

relationship between cows in this Guernsey population to

be half-sib, the heritability therefore can be expressed as
4 O:ire

2 2 °
. + o
Slre error

 

0

The desired variance ratio to be added to the diagonal of

the sire portion of the MME can be calculated as

_._.1,
h2

51

given the heritability for each trait. Heritabilities as
calculated by Mao et al. (1977) and the resultant variance
ratios are shown in Table 4.

3.3.5 Solutions of Mixed
Model Equations

 

 

Mixed model equations were constructed using the
set of data ALL, and in the process, herd-year group
factor was absorbed into the remaining effects. After the
diagonal of the sire portion of the MME matrix was adjusted
by the variance ratio for the trait being evaluated, four
rows and columns corresponding to levels of fixed effects
were deleted to eliminate linear dependencies within the
MME's. This was because a singular matrix, that is, one
which contained linear dependencies, could not be solved
as efficiently as a non-singular matrix. However, the
deletion of dependent rows and columns had no effect on
the sire solutions.

Solutions were found within all traits using each
model. Sire estimates as well as age group and month
constants were determined; results of using WITHCOV model
included a regression coefficient for dam covariate. Age
group and month were considered nuisance variables and
their solutions were not of interest in this study.

The International Mathematical and Statistical
Library (IMSL) of Fortran subroutines was used in part to

solve the MME's. All computations were performed on a CDC

52

TABLE 4.--Heritabilities of Guernsey Type Traitsa and
Resultant Variance Ratios.

 

 

0error
2
Trait Heritability Osire
General appearance .24 15.67
Breed character .22 17.18
Fore feet and legs .18 21.22
Rear feet and legs .06 65.67
Rump .31 11.90
Dairy character .16 24.00
Chest and body .25 15.00
Fore udder .21 18.05
Rear udder .19 10.05
Teats .24 15.67
Mammary system .23 16.39
Final score .24 15.67

 

aMAO, 1977.

53

Cyber series computer at Michigan State University.
Because the order of the matrix was nearly 100 for each
evaluation, the cost of obtaining an inverse would have
been many times higher than the cost of obtaining only
solutions, and for this reason no inverses were calculated.
The VCV matrix for solutions of fixed effects can
2

be calculated as P Oerror’ where P 18 the fixed portion

of the inverse of the MME coefficient matrix. Also, the

VCV matrix of sire prediction error would be S 02 ,

~ error
where S is the random portion of the inverse of the MME
coefficient matrix. Without an inverse of the coefficient
matrix, neither of these sets of variances and covariances

could be computed.

3.3.6 Disconnectedness

Disconnectedness among groups of sires must be
considered if sire solutions are compared. For one group
of sires to be disconnected from another group in this
study, all of the progeny of sires in the first group must
be appraised in different herd-year groups, and/or differ-
ent age groups, and/or different months than all of the
daughters of the second group of sires. If this condition
is not true, the two groups of sires are actually one
connected group.

For example, suppose all of the daughters of one

sire were classified in December, while no other cows were

54

classified in that month. It would be impossible to
separate the effect of that month from the sire effect,
and hence that sire could not be compared to other sires.
In this study, there were only eight age groups and twelve
months, so disconnectedness by month or age group would
be extremely unlikely. Because there were so many herd-
year groups, however, it was possible that a sire or group
of sires was disconnected through that factor.

A program written by the author was used to search
for disconnected groups of sires. All sires were com-

pletely connected within the herd-year group factor.

3.3.7 Comparison of Sire Rankings

 

For each trait, bulls were ranked twice, based on
sire estimates obtained using both models, NOCOV and
WITHCOV. Spearman's rank correlation was calculated for
each trait to determine correspondence between the two
sets of rankings. If the rank of sire i resulting from

the NOCOV model is ti and from the WITHCOV model is ti

1 2'

then correlation between the ranks of the 80 bulls

determined from two models is:

80
6.2 Itil - tiZ)
_ 1-1
r—1- 3 °
80 - 80

2

 

55

3.4 Mates' Influence on Sire Rankinngsing
Initial Daughter Samples

The influence of a sire's mates on his evaluation
for body conformation was studied when a much smaller
number of progeny was available. The models used to
investigate this effect are identical to those described

in section 3.3.1.

3.4.1 Editing the Data

From the set of data ALL, the first 20 daughters
of each sire were included in the data file termed FIRST.
There were 1,600 appraisals within this file, comprising
557 herd-year groups.

Absorption of herd-year group, addition of vari-
ance ratios to the diagonal of sire portion of the MME,
and solution of mixed model equations were performed as
described in sections 3.3.3 through 3.3.5, respectively,
for each trait.

3.4.2 Discgppectedness Among
Sires Having Few Progeny

One of the effects of reducing the number of
progeny per sire to 20 is the occurrence of disconnected
groups. Using the program described in section 3.3.6, it
‘was determined that 35 groups of sires in the FIRST data
file were disconnected by herd-year group. Single dis—
<:onnected sires made up 28 of these disconnected groups;

tsolutions for these sires cannot be separated from

56

constants of herd-year groups in which their daughters
were appraised, and therefore are not meaningful. Of the
remaining seven disconnected groups, only three contain
more than three sires.

Comparisons should not be made between sires in
disconnected groups, although within connected groups
comparisons are valid. Therefore to obtain rank correla-
tions between two sets of sire solutions, at least one of
which was calculated from the set of data FIRST, a
weighted average is used. Rank correlations are calculated
as described in section 3.3.7, except that instead of
including all 80 sires, the calculations employ only the
number of sires within a connected group. The three major
connected groups consist of 9, 11, and 23 sires. If the
rank correlations obtained within these groups are denoted
r1, r2, and r3, then the weighted rank correlation as

described by Donner and Rosner (1980) is:

9r + llr + 23r

1 2
9 + 11 + 23

3.

 

3.4.3 Comparisons Made Using Results
of Evaluations Based on Few Progeny

 

 

Weighted correlations as described in the previous
section were used to make three comparisons. First,
'weighted correlations were calculated between results of

‘using both models, NOCOV (without covariate) and WITHCOV

57

(with a covariate), with only the sires' first 20
daughters. Then, the same two sets of sire estimates were
correlated, one at a time, to estimates obtained using all
the data (ALL data set) with the NOCOV model. This was in
an attempt to determine the accuracy with which either
model using few progeny records could predict a sire's
evaluation when more records on daughters are available.
In calculating the latter two weighted correla-
tions, sire estimates resulting from using the ALL file
were considered within the connected groups of the smaller
FIRST file. This is because connected sires, if Split
into groups, would maintain their relative ranks within
groups. On the other hand, if disconnected sires are
combined into one group, there is no assurance that a

ranking based on their estimates would be correct.

4. RESULTS AND DISCUSSION

4.1 Extent pron-Random Mating for
Final Class

 

 

A Chi-square goodness—of-fit test was performed
on each of 306 sires having at least 40 daughters.
Results, reported. in Table 5, indicate the proportion of
sires whose mates are not representative of the population

at three levels of confidence.

TABLE 5. Proportion of Sires Having Mates Distributed
Differently from the Population at Various
Confidence Levels.

 

 

Level of Confidence Proportion of Sires
90% 45.4%
95% 37.2%
99% 28.1%

 

This procedure does not quantify the amount nor
direction of mates' bias. For example, the three distri-
butions shown in Figure 5 might be judged significantly

different. However, this procedure will not indicate

58

59

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

a.

.10 .20 .40 .20 .10
b.

.28 .24 .20 .18 .12
c. —_————

.....r
.12 .18 .20 .22 .28
1 2 3 4 5

-frequencies

-frequencies

-frequencies

CLASS

Figure 5.--Examples of Differing Descriptions

60

whether a distribution is biased up (figure c) or down
(figure b) or to what degree it differs, from figure a.

From these results, then, it seems that many
sires may have been bred to samples of mates which were
not representative of the population. This could be the
result of deliberate non-random mating, such as corrective
mating for type, practiced by Guernsey breeders. It also
could be the incidental result of any preferential mating
for other traits which may be positively correlated to
body conformation traits.

If left unadjusted for mates' scores, provided
that mates were a selected non-random sample from the
population, bulls' rankings based on daughters' scores
could be biased.

4.2 Effect of Mates' Scores on Sire
Rankings Based on Many Progeny

Sire solutions were obtained for each of the
twelve traits using both models (NOCOV, without covariate,
and WITHCOV, with covariate) and all of the sires'
progeny (ALL). The expected values of all means of sire
estimates and the observed means of those estimates were
zero. The standard deviations of estimates are shown in
Table 6. They are virtually identical between the models
used, indicating comparable ranges of estimates. Because
of differences in heritabilities among the various traits,

the standard deviations among traits varied quite widely.

61

TABLE 6.--Standard Deviations of Sire Estimates Using All
Progeny of Sires.

 

 

 

Model
Without Covariate With Covariate
Trait (NOCOV) (WITHCOV)
General appearance .140 .139
Breed character .154 .155
Fore feet and legs .124 .125
Rear feet and legs .077 .077
Rump .153 .153
Dairy character .129 .129
Chest and body .137 .175
Fore udder .163 .163
Rear udder .145 .145
Teats .134 .134
Mammary system .153 .152

Final score .147 .146

 

62

If the distributions of daughter scores are similar
between two traits, one would expect to find greater dif-
ferences among sire estimates for the trait with a higher
heritability. For traits with higher heritability, a
greater portion of phenotypic variation is caused by sire
effect, and it is subsequently easier to distinguish
genetic ability among sires.

Rank correlations were calculated on all traits
between the two sets of sire solutions corresponding to
both models. These are shown in Table 7. All correlations
between results of these two evaluations, for all traits,
exceeded .995, which indicates little meaningful differ—
ence between the two sets of solutions; their rankings are
not significantly changed due to the addition of dam's
score.

4.3 Mates' Effect on Sire Rankings Based
on Initial Danghter Samples

 

In practice, sires having few mates are more sub-
ject to random sampling variation, which could result in
a bull having a non-representative sample of mates.
Because young sires routinely are selected on evaluations
based on few progeny, it is important to eliminate as much
sampling variation as possible.

Three sets of correlations were computed to assess
the value of including a dam covariate in a sire evaluation

model when few data are available. First, for sire

63

TABLE 7.--Rank Correlations Between Models Including and
Excluding Dam Covariate Using All Progeny of

 

 

Sires.
Trait Spearman's r
General appearance .997
Breed character .998
Fore feet and legs .998
Rear feet and legs .998
Rump .995
Dairy character .999
Chest and body .996
Fore udder .998
Rear udder .999
Teats .998
Mammary system .998

Final score .998

 

64

estimates calculated from few progeny, the rank correla-
tions between the two models were examined. These corre-
lations do not indicate whether differences caused by
inclusion of a covariate constitute an improvement in
accuracy, however. Therefore, each set of results was
compared separately to results obtained using many more,
data, to determine whether inclusion of sires' mates'
covariate results in an improvement in accuracy of rank-
ings. It was assumed that rankings of sires as evaluated
using all of their progeny were more accurate than rank-
ings using only the first 20 daughters, so that higher
values in the second set of correlations indicate better
accuracy of the evaluations based on few progeny than do
lower values.

Rank correlations between results of both models
using few data are shown in Table 8. These correlations
are weighted averages as described in section 3.4.2.
Comparison with Table 7 shows that, for each trait, there
is less agreement between rankings of the two models when
fewer data are available than when all daughters are used.
In general, however, the correlations are quite high.

In Table 9 the accuracies of the two models using
few records are contrasted. In general, only moderate
weighted rank correlations were found between evaluations

based on few records and evaluations based on many,

TABLE 8.—-Weighted Rank Correlations Between Models
Including and Excluding Dam Covariate Using

Initial Progeny of Sires.

 

Trait

Spearman's r

 

General appearance
Breed character
Fore feet and legs
Rear feet and legs
Rump

Dairy character
Chest and body
Fore udder

Rear udder

Teats

Mammary system

Final score

.940
.993
.985
.989
.992
.980
.980
.970
.990
.964
.988
.974

 

66

TABLE 9.--Weighted Rank Correlations of Both Models Using
Initial Sample of Daughters to Model Without
Covariate Using All Daughters.

 

 

 

Model
Without Covariate With Covariate

Trait (NOCOV) (WITHCOV)
General appearance .436 .348
Breed character .134 .43
Fore feet and legs .327 .365
Rear feet and legs .235 .240
Rump .295 .294
Dairy character .316 .244
Chest and body .640 .656
Fore udder .276 .214
Rear udder .293 .273
Teats .238 .261
Mammary system .172 .169

Final score .512 .475

 

67

indicating that evaluations based on few data provide only
a weak indication of a sire's true genetic rank.

If sire solutions are more accurate when their
mates' covariate is included in the model, we would expect
corresponding correlations to be higher. However, upon
examination of Table 9, little difference can be seen
between models in their ability to predict sire rankings.
Among traits, however, a wide range of correlations can
be seen, suggesting that evaluations of young sires may
be more accurate for some traits, such as chest and body,
than for others such as breed character or mammary system.
While reasons for this are not clear, large phenotypic
variances may influence young sire evaluation because a
more balanced range of category frequencies for daughters
exists, making it easier to distinguish between sires.

4.4 Examination of Covariate Regression
Coefficients

 

 

Regression coefficients of the sires' mates'
covariates using all of the data and only the first 20
daughters of each sire are reported in Table 10. They are
all greater than zero, indicating a positive relationship
between type score for dam and daughter in all traits.
Standard errors were not computed for reasons given in
section 3.3.5. Across traits, 95% confidence intervals
about the mean regression coefficient were .008 t .055

and .094 i .087, using all of the data and the first 20

68

TABLE 10.--Regression Coefficients for Dam Covariates.

 

 

 

Data Set

All Initial 20
Trait Daughters Daughters
General appearance .0989 .1492
Breed character .0923 .0661
Fore feet and legs .0773 .1036
Rear feet and legs .0546 .0557
Rump .1255 .0925
Dairy character .0602 .0431
Chest and body .1428 .1814
Fore udder .0882 .0831
Rear udder .0680 .0639
Teats .0719 .1086
Mammary system .0836 .0895
Final score .0962 .0898

 

69

daughters of each sire, respectively. It appears that
regression coefficients are not zero. Because the vast
majority (nearly 85%) of all animals are classified in
only two categories, very good or desirable, and because
the heritabilities of these traits are generally moderate
to low, these coefficients may be of little predictive
value for daughter classification.

It is interesting to note that despite the
paucity of information in the set of data FIRST (contain-
ing all sires' first 20 daughters) compared to the ALL
data (including all sires' progeny), the correlation
between regression coefficients computed from both data
sets is .735, rather high in contrast to correlations of
sire solutions between sets of data, which range from

.179 for mammary system to .656 for chest and body.

5. CONCLUSIONS

Among Guernsey sires having at least 40 daughter
appraisals, more than one-third (37.2%) are bred to mates
which are not a random sample from the dam population
with respect to final class (p < 0.05). However, it can-
not be concluded from these results that, if left
unadjusted for mates, these sires' evaluations would be
biased. Though a sire's mates' distribution of final
classes may differ from the pOpulation distribution, the
mates' frequencies in each category may be such that there
is negligible prejudice toward either extreme. Also,
despite the bias of a sample of mates toward an extreme,
its effect upon a sire's evaluation may not be signifi-
cant.

After acknowledging these two possibilities, the
goodness-of-fit test is seen to indicate only the poten—
tial for bias in an evaluation. Comparisons were made
between results of evaluations adjusted and unadjusted
for mates.

When sire solutions were compared between the two
models, within each data set, rank correlations were high
(> .96) for all traits. Within the large set of data

(ALL), all correlations exceeded .99. There were

70

71

essentially, no differences between sire solutions
obtained with and without a covariate of mate's score

when all sires had a large amount of daughter information
available. When each sire had only 20 daughters, the
correlations between sire solutions obtained from the two
models were slightly smaller across traits, but still near
unity.

Sire solutions obtained from the smaller set of
data (FIRST) provided much smaller correlations than
those obtained from ALL, regardless of model used. These
correlations varied widely across traits, being highest
for chest and body, and lowest for breed character.
Possible reasons for these discrepancies were included
in the Discussion. From the correlations, one may con-
clude that neither model was superior in predicting a
later, presumably more accurate, sire solution.

If a covariate of mates' scores were to be
included in a model to evaluate more Guernsey sires for
conformation traits, then, we might not expect to see a
great change in the ranking of sires with many progeny.
If a sire having few daughters had been bred to a sample
of mates very biased toward either extreme, however, we
might expect a more noticeable change in his rank. From
the results of this investigation, there is little indi-
cation that this change would improve the accuracy of the

sire's evaluation.

72

When reviewing results of this investigation, it
is imperative to consider its scope, especially in two
areas: the sires included, and the use of somewhat sub-
jective classification scores as records rather than more
objective or continuous traits.

One must be careful when applying results obtained
from examination of only the eighty Guernsey sires with
the most daughters to additional bulls or to other breeds.
When a sire has few daughters, his evaluation is more sub-
ject to sampling errors and therefore less precise. This
imprecision is increased if the sire's progeny are in very
few herds. Differences in philosophies among breed
organizations about body conformation's desired relation-
ship to production traits can affect sire selection among
dairymen; non-random mating may not occur to the same
degree as found in this study. Also, genetic parameters
of body conformation traits can differ markedly among
breeds.

It is also difficult to extend conclusions of this
research to any continuously scored trait, such as milk or
fat production or linearly measured conformation char-
acteristics. Concerning production traits, the extent of
non-random mating may be quite different than for type
traits, possibly eliminating the need to adjust for mate's
ability. When an appraiser uses linear scales for body

conformation, some of his subjectivity may be reduced

73

compared to a system using classification scores. In
addition, by being scored on a more continuous scale, a
more precise adjustment for covariate can be given to
each observation. This may result in more normally dis-
tributed residuals. The increased continuity of produc-
tion records, as well as the more objective assessment
of cows, might have the same effect, but to a greater

degree.

LI ST OF REFERENCES

74

LIST OF REFERENCES

Bereskin, B., and A. E. Freeman. 1965. Genetic and
environmental factors in dariy sire evaluation.
II. Uses and limitations of deviation records
and the role of dams. J. Dairy Sci. 48:352.

Berger, P. J., W. R. Harvey, and E. R. Rader. 1973.
Selection for type and production and influence
on herd life of Holstein cows. J. Dairy Sci.
56:805.

Berousek, E. R., J. A. Whatley, Jr., R. D. Morrison,
S. D. Musgrave, and W. R. Harvey. 1959. Herita-
bility and repeatability estimates of production
and type of Guernsey cattle. J. Dairy Sci.
42:925.

Blake, R. W., and B. T. McDaniel. 1979. Relationships
of udder conformation with labor and machine
inputs to milk harvest in dairy cattle. J. Dairy
Sci. 62:475.

Bodoh, G. W., and G. E. Shook. 1972. Factors affecting
selection of mates of sires in artificial service.
J. Dairy Sci. 55:851.

Broffit, J. D., and R. H. Randles. 1977. A power
approximation for the Chi-square goodness of fit
test: Simple hypothesis case. J. Am. Stat.
Assoc. 72:604.

Butcher, D. F., R. G. Mitchell, I. D. Porterfield, and
R. S. Dunbar, Jr. 1963. Heritability, phenotypic
and genetic correlations between type ratings and
milk fat production in Ayrshire cattle. J. Dairy
Sci. 46:971.

Calo, L. L., R. E. McDowell, L.IL Van Vleck, and P. D.
Miller. 1973. Genetic aspects of beef production
among Holstein-Friesians pedigree selected for
milk production. J. Anim. Sci. 37:676.

75

76

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

Cassell, B. G., W. E. Vinson, J. M. White, and R. H.
Kliewer. 1973. Age correction factors for type
traits in Holstein cows. J. Dairy Sci. 56:1178.

Cassell, B. G., J. M. White, W. E. Vinson, and R. H.
Kliewer. 1973. Sire-son relationships for type

in Holstein-Friesian cattle. J. Dairy Sci. 56:
1164.

Donner, A., and B. Rosner. 1980. On inferences concern-
ing a common correlation coefficient. Applied
Stat. 29:69.

Everett, R. W., J. F. Keown, and E. E. Clapp. 1976.
Relationships among type, production and stay-
ability in Holstein cattle. J. Dairy Sci. 59:
1505.

Everett, R. W., R. L. Quaas, and A. E. McClintock. 1979.
Daughters' maternal grandsires in sire evaluation.
J. Dairy Sci. 62:1304.

Freeman, A. B., and R. S. Dunbar, Jr. 1955. Genetic
analysis of the components of type conformation
and production in Ayrshire cows. J. Dairy Sci.
38:428.

Gilmore, J. A., and B. T. McDaniel. 1977a. Relation of
selection traits and health costs to relative
annual profitability of dairy cattle. J. Dairy
Sci. 60(supp. 1):75.

Gilmore, J. A., and B. T. McDaniel. 1977b. Health costs
of dairy cattle, its variation, and its associa-
tions with milk yield and type scores. J. Dairy
Sci. 60(supp. 1):76.

Grantham, J. A., Jr., J. M. White, W. E. Vinson, and R. H.
Kliewer. 1974. Genetic relationships between
milk production and type in Holsteins. J. Dairy
Sci. 57:1413.

Hansen, L. R., G. R. Barr, and D. A. Wieckert. 1969.
Effects of age and stage of lactation on type
classification. J. Dairy Sci. 52:646.

77

Harvey, W. R., and J. L. Lush. 1952. Genetic correla-
tions between type and production in Jersey
cattle. J. Dairy Sci. 35:199.

Henderson, C. R. 1950. Estimation of genetic parameters.
Ann. Math. Stat. 21:309.

Henderson, C. R. 1966. Sire evaluation method which
accounts for unknown genetic and environmental
trends, herd differences, seasons, age effects,
and differential culling. Proc. Nat. Symp. on
Estimating Breeding Values of Dairy Sires and
Cows.

Henderson, C. R. 1974. General flexibility of linear
model techniques for sire evaluation. J. Dairy
Sci. 57:963.

Henderson, C. R. 1975a. Rapid method for computing the
inverse (If a relationship matrix. J. Dairy Sci.
58:1727.

Henderson, C. R. 1975b. Use of relationships among
sires to increase accuracy of sire evaluation.
J. Dairy Sci. 58:1731.

Hillers, J. K., and A. E. Freeman. 1966. Two sources of
genetic error in sire proofs. J. Dairy Sci. 49:
1245.

Honnette, J. E., W. E. Vinson, J. M. White, and R. H.
Kliewer. 1980. Prediction of herdlife and life-
time production from first lactation production
and individual type traits in Holstein cows.

J. Dairy Sci. 63:816.

Johnson, K. R., and D. F. Fourt. 1960. Heritability,
genetic and phenotypic correlation of type,
certain components of type, and production of
Brown Swiss cattle. J. Dairy Sci. 43:975.

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

Karner, P. J., and L. D. Van Vleck. 1980. Sire evalua-
tion by Best Linear Unbiased Prediction for
categorically scored type traits. J. Dairy Sci.
63:1328.

78

Kennedy, B. W., and J. E. Moxley. 1975. Comparison of
genetic group and relationship methods for mixed
model sire evaluation. J. Dairy Sci. 58:1507.

Legates, J. E. 1971. Components of variance for descrip-
tive type classification of Holstein-Friesian
cows. J. Dairy Sci. 54:775.

Liang, L. K. P., and O. T. Fosgate. 1967. Multiple
regressions analysis of type classification and
milk production in Jersey cattle. J. Dairy Sci.
50:974.

Lush, J. L. 1944. Optimum emphasis on dams' records
when proving dairy sires. J. Dairy Sci. 27:937.

Mac, I. L., R. R. Neitzel, and C. E. Meadows. 1977.
Type traits in Guernsey cattle. I. Sources of
variation and trends in genetic parameters.
Paper presented to 72nd Annual Meeting of American
Dairy Science Association, Ames, Iowa.

McGilliard, L. D., and J. L. Lush. 1956. Changes in
type classification of dairy cattle. J. Dairy
Sci. 39:1015.

Miller, R. H., and E. L. Corley. 1965. Usefulness of
information on mates of sires in artificial
insemination. J. Dairy Sci. 48:580.

Mitchell, R. G., E. L. Corley, and W. J. Tyler. 1961.
Heritability, phenotypic and genetic correlations
between type ratings and milk and fat production
in Holstein-Friesian cattle. J. Dairy Sci.
44:1502.

Moreno, A. M., G. R. Wiggans, and L. D. Van Vleck. 1979.
Genetic and herd-year variation in type traits
of Brown Swiss cows. J. Dairy Sci. 62:486.

Norman, H. D., and B. G. Cassell. 1978. Effect of herd,
year, age, and stage of lactation on Jersey type
classification. J. Dairy Sci. 61:352.

Norman, H. D., B. G. Cassell, F. N. Dickenson, and E. E.
Wright. 1978. Phenotypic and genetic relation-
ships among type classification traits in Jerseys.
J. Dairy Sci. 61:1250.

79

Norman, 11. D., B. G. Cassell, G. J. King, R. L. Powell,
and E. E. Wright. 1979. Sire evaluation for
conformation of Jersey cows. J. Dairy Sci. 62:
1914.

Norman, H. D., and L. D. Van Vleck. 1972a. Type appraisal:
I. Effects of age and stage of lactation on type
ratings. J. Dairy Sci. 55:1706.

Norman, H. D., and L. D. Van Vleck. 1972b. Type appraisal:
II. Variation in type traits due to sires, herds,
and years. J. Dairy Sci. 55:1717.

O'Bleness, G. V.. and L. D. Van Vleck. 1962. Reasons
for disposal of dairy cows from New York herds.
J. Dairy Sci. 45:1087.

Purohit, V. D., E. B. Burnside, T. R. Batra, J. W. Wilton,
and M. G. Freeman. 1973. Specific combining
ability among dairy sires for type traits. J.
Dairy Sci. 56:1085.

Rennie, J. C., T. R. Batra, M. G. Freeman, J. W. Wilton,
and E. B. Burnside. 1971. Estimation of environ-
mental and genetic parameters for type traits in
Holstein cows. J. Dairy Sci. 54:775.

Rothschild, M. F., C. R. Henderson, and R. W. Everett.
1976. Bias of sire evaluations for natural

service sires and age of sire. J. Dairy Sci.
59:2091.

Schaeffer, L. R., M. S. Hunt, and E. B. Burnside. 1978.
Evaluation of Holstein-Friesian dairy sires for

conformation of their daughters. Can. J. Anim.
Sci. 58:409.

Schaeffer, L. R., and E. B. Burnside. 1974. Survival
rates of tested daughters of sires in artificial
insemination. J. Dairy Sci. 57:1394.

Sowerby, M. B. 1978. The WHY of classification rating.
Guernsey Breeders' Journal 148(3):224.

Specht, L. W., H. W. Carter, and L. D. Van Vleck. 1967.
First classification score and length of herd
life. J. Dairy Sci. 50:1690.

Tabler, K. A., and R. W. Touchberry. 1956. Estimating
the heritability of type classification of dairy
cattle. J. Dairy Sci. 39:1550.

80

Tong, A. K. W., B. W. Kennedy, and J. E. Moxley. 1980.
Potential errors in sire evaluation from regional
genetic differences. J. Dairy Sci. 63:627.

Touchberry, R. W. 1951. Genetic correlations between
five body measurements, weight, type, and produc-
tion in the same individual among Holstein cows.
J. Dairy Sci. 34:242.

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 con-
secutive times. J. Anim. Sci. 10:1029.

Tyler, W. J., and G. Hyatt, Jr. 1948. The heritability
of official type ratings and the correlation
between type ratings and butterfat production of
Ayrshire cows. J. Dairy Sci 31:63.

Ufford, G. R., C. R. Henderson, J. F. Keown, and L. D.
Van Vleck. 1979. Accuracy of first lactation
versus all lactation sire evaluations by Best
Linear Unbiased Estimation. J. Dairy Sci. 62:
603.

Van Vleck, L. D. 1964. Variation in type scores due to
sire and herd effects. J. Dairy Sci. 47:1249.

Van Vleck, L. D., P. J. Karner, and G. R. Wiggans. 1980.
Relationships among type traits and milk yield
of Brown Swiss cattle. J. Dairy Sci. 63:120.

Van Vleck, L. D., and H. D. Norman. 1972. Association
of type traits with reasons for disposal. J.
Dairy Sci. 55:1698.

Vinson, W. E. 1981. Sire selection to improve production
and type. Holstein Science Report, February 1981.

Vinson, W. E., J. M. White, B. L. Combs, and R. H.
Kliewer. 1976. Sources of variation in Holstein
descriptive classification traits. J. Dairy
Sci. 59:522.

White, J. M., J. E. Legates, and K. L. Koonce. 1967.
Environmental and genetic factors affecting type
appraisal scores. J. Dairy Sci. 50:974.

Wickham, B. W., and C. R. Henderson. 1977. Sire evalua—
tion by second lactation records of daughters.
J. Dairy Sci. 60:96.

Wilcox,

Wilcox,

Wilcox,

81

C. J., R. E. Miller, K. O. Pfau, R. F. Gabriel,
and J. W. Bartlett. 1959. Changes in type
ratings of Holstein cows due to age, season,
stage of lactation, classifier, and year.

J. Dairy Sci. 42:1967.

C. J., K. O. Pfau, R. E. Mather, and J. W.
Bartlett. 1958. Effects of season, age, and
stage of lactation upon type ratings of Holstein
cows. J. Dairy Sci., 41:1065.

C. J., K. O. Pfau, R. E. Mather, R. F. Gabriel,
and J. W. Bartlett. 1962. Phenotypic, genetic,
and environmental relationships of milk production
and type ratings of Holstein cows. J. Dairy Sci.
45:223.

"I11111111711111.1111“