Lecture 6 Heritability and Field...
-
Upload
nguyennhan -
Category
Documents
-
view
289 -
download
0
Transcript of Lecture 6 Heritability and Field...
Lecture 6 Heritability and Field Design
Lucia Gutierrez lecture notesTucson Winter Institute
1
Selection Response
0µSµ
Y)(selection toresponse R
X)( ldiferentiaselection S
pointn truncatio c
sindividual selected ofprogeny a ofmean
population R.M. from sindividual selected a ofmean
population Mating Random initial theofmean
1
0
∆=∆=
====
µµµ
S
1µ
S
R
2,
,
)(
hS
Rb
SbR
XbY
yx
yx
==
=∆=∆
Heritability: “Expected proportion of selection differential to be achieved as a gain from selection” (Hanson, 1963).
Holland et al. 20102
c
Heritability
REALIZED HERITABILITYThe relation between the observed response to selection and the selection differential.
NARROW SENSE HERITABILITYThe fraction of all trait variation due to variation in breeding values (additive variance).
BROAD SENSE HERITABILITYThe fraction of all trait variation due to variation in genotypic values (genetic variance).
P
A
V
Vh =2
P
G
V
VH =2
S
Rh
ˆ
ˆ2 =
3
Holland et al. 2010
Heritability
FUNCTION OF POPULATION
o They are functions of both genetic and environmental variance, therefore a property of the population.
o Suppose one inbred line with a Mendelian inherited trait. How much is the genetic variance? Will the trait segregate? How much is the h2? Does it mean that the trait is not genetically determined? Since there is no genetic variation within this population, h2=H2=0. However, it does not mean that the trait is not genetically determined.
o Additionally, it is unreliable to predict future response to selection because while conducting selection the population changes.
4
Holland et al. 2010
Heritability
ENVIRONMENTAL VARIANCE
Since h2 is also a function of environmental variance, and decreasing environmental variance increases h2, controlled conditions would be optimal for identifying superior genotypes (predicting breeding values). However, caution should be exercised because GxE is important for many traits and therefore selecting in a non-targeted environment could be detrimental.
5
Holland et al. 2010
Parent-Offspring Regression
( ) ipipooi ezbz +−+= µµ |
deviationse
offspringth i ofparent of phenotypez
slope offspring-parent regression b
mean populationµ
offspringth -i theof phenotypez
i
pi
po|
oi
=
−=
==
=
LW 17
Offspring
Parent
-20
-15
-10
-5
0
5
10
15
20
-20 -15 -10 -5 0 5 10 15 20
6
Parent-Offspring Regression
( ) ( )( )
( )po
z
AAApo bhh
EpEobE |
222
24
122
1
p2
po| 2 ,
2
1,
z
z,zˆ ≅≅++≅=σ
σσσσ
σ
Regression one parent on offspring – no environment correlation among parent and offspring.
( ) ( )( )
( )( ) po
z
AAApo bhhbE |
222
21
24
122
1
P221
P1212
P221
P121
o
p2
po| ,
zz
zz,z
z
z,zˆ ≅=+=++==
σσσ
σσ
σσ
Regression one parent – offspring (one offspring or the mean of multiple offspring).
( ) ( )( ) po
z
AAApo bhhbE |
222
24
122
1
p2
po| 2 ,
2
1
z
z,zˆ ≅≅+≅=σ
σσσ
σ
Regression mid parent on offspring – no environment correlation among parent and offspring.
( ) ( )( ) 0|1:0
222
2212
122
12
S02
1:S0S00|1:0 ,
z
z,zˆSS
z
AADDASS bhhbE ≅≅+++==
σσσσσ
σσ
Regression parent -offspring inbreeding – no environment correlation.
LW 177
Mating Designs
FULL-SIB DESIGN: N full-sib families with n offspring each.
ijiij wfz ++= µ
iancefamily var-ithin w
on)contributi talenvironmen dominance, on,(segregatierror residualw
familyth -i theofeffect f
mean populationµ
familyth -i theof offspringth -j theof phenotypez
ij
i
ij
===
=
SoV df SS MS EMS
Among-families N-1 SSf/df(f)
Within-families n(N-1) SSw/df(w)
( )∑ −=i if zznSS 2
...
∑ −=ji iijw zzSS
,
2. )( 2
)(FSwσ
22)( fFSw nσσ +
LW 188
Mating Designs
FULL-SIB DESIGN: N full-sib families with n offspring each.
ijiij wfz ++= µ
( )( )( ) ( ) ( ) ( )2
, ,, ,
,
,)(
f
ijijiijijiii
ijiiji
ijij
wwfwwfff
wfwf
zzFSCov
σ
σσσσµµσ
σ
=
+++=
++++=
=
224
122
12EcDAf σσσσ ++=
( )( )
2224
322
1
224
122
1222
224
122
122)(
EcEDA
EcDAEDA
EcDAPFSw
σσσσσσσσσσ
σσσσσ
+++=
++−++=
++−=
2)(
22FSwfP σσσ +=
LW 189
Mating Designs
FULL-SIB DESIGN: N full-sib families with n offspring each.
( )( )( ) ( ) ( )wfz
w
f wf
VarVarVar
MSVarn
MSMSVar
w
+==
−=
( )( )
( )[ ] [ ])1(211)1(2)(
2
2
1
Var
Var
2
2
2
224
12
2
224
122
1
−−+−≅
≅
++=++==
nNntnthSE
th
hz
ft
FSFS
FS
z
EcD
z
EcDAFS σ
σσσ
σσσ
ijiij wfz ++= µ
SoV df SS MS EMS
Among-families N-1 SSf/df(f)
Within-families n(N-1) SSw/df(w)
( )∑ −=i if zznSS 2
...
∑ −=ji iijw zzSS
,
2. )( 2
)(FSwσ
22)( fFSw nσσ +
LW 1810
Mating Designs
HALF-SIB DESIGN: N half-sib families with n offspring each.
ijiij wfz ++= µ
iancefamily var-ithin w
on)contributi talenvironmen dominance, on,(segregatierror residualw
familyth -i theofeffect f
mean populationµ
familyth -i theof offspringth -j theof phenotypez
ij
i
ij
===
=
SoV df SS MS EMS
Among-families N-1 SSf/df(f)
Within-families n(N-1) SSw/df(w)
( )∑ −=i if zznSS 2
...
∑ −=ji iijw zzSS
,
2. )( 2
)(FSwσ
22)( fFSw nσσ +
LW 1811
Mating Designs
HALF-SIB DESIGN: N half-sib families with n offspring each.
ijiij wfz ++= µ
iancefamily var-ithin w
on)contributi talenvironmen dominance, on,(segregatierror residualw
familyth -i theofeffect f
mean populationµ
familyth -i theof offspringth -j theof phenotypez
ij
i
ij
===
=
SoV df SS MS EMS
Among-families N-1 SSf/df(f)
Within-families n(N-1) SSw/df(w)
( )∑ −=i if zznSS 2
...
∑ −=ji iijw zzSS
,
2. )( 2
)(FSwσ
22)( fFSw nσσ +
( )( )( ) ( ) ( )wfz
w
f wf
VarVarVar
MSVarn
MSMSVar
w
+==
−= ( )
( )HS
z
AHS
th
hz
ft
4
4
1
Var
Var
2
22
24
1
≅
===σσ
LW 1812
Mating Designs
NORTH CAROLINA DESIGN I: Each male (N sire) is mated to several unrelated females (M dams) to produce n offspring per dam.
ijkijiijk wdsz +++= )(µ
)deviations iancefamily var-(withinerror residualw
sireth -i the tomated damth -j theofeffect d
sireth -i theofeffect s
mean populationµ
damth -j and sireth -i theoffamily thefrom offspringth -k theof phenotypez
ijk
ij
i
ijk
=
===
=
LW 1813
Mating Designs
NORTH CAROLINA DESIGN I: Each male (N sire) is mated to several unrelated females (M dams) to produce n offspring per dam.
ijkijiijk wdsz +++= )(µ
SoV df SS MS EMS
Sires N-1 MSs/df(s)
Dams(Sire) N(M-1) MSd/df(d)
Sibs(dams) T-NM MSw/df(w)
( )∑ −=ji is zzMnSS
,
2...
∑ −=ji iijd zzSS
,
2. )(
222sdw Mnn σσσ ++
( )
( )( )( ) ( ) ( ) ( )wdsz
w
d
s
Wd
dS
VarVarVarVar
MSVarn
MSMSVar
Mn
MSMSVar
w
++==
−=
−=( )( )
( ) ( )( )
PHS
z
EcD
z
EcDAFS
z
APHS
th
hz
dst
hz
st
4
2
1
Var
VarVar
4
1
Var
Var
2
2
224
12
2
224
122
1
22
24
1
≅
++=++=+=
===
σσσ
σσσσ
σσ
∑ −=kji ijijkw zzSS
,,
2. )(
22dw nσσ +
2wσ
LW 1814
Mating Designs
NORTH CAROLINA DESIGN II: A group of sires (NS sires) are mated to an independent group of dams (ND dams) to produce n offspring
.
ijkijjiijk wIdsz ++++= µ
)deviations iancefamily var-(withinerror residualw
damth -j theand sireth -i ebetween thn interactio theofeffect Iij
damth -j theofeffect d
sireth -i theofeffect s
mean populationµ
damth -j and sireth -i theoffamily thefrom offspringth -k theof phenotypez
ijk
i
i
ijk
=====
=
LW 2015
Mating Designs
NORTH CAROLINA DESIGN II: A group of sires (NS sires) are mated to an independent group of dams (ND dams) to produce n offspring
.ijkijjiijk wIdsz ++++= µ
SoV df SS EMS
Sires Ns-1
Dams
Interaction
Nd-1
(Ns-1)(Nd-1)
Sibs NsNd(n-1)
∑ −−−=ji jiijI zzzzSS
,
2........ )(
222sdIw nNn σσσ ++
2
24
1
z
APHSt
σσ=
∑ −=kji ijijkw zzSS
,,
2. )(
22Iw nσσ +
2wσ
( )∑ −=j jsd zznNSS 2
.....
( )∑ −=i ids zznNSS 2
.....
222dsIw nNn σσσ ++
2
2224
1
z
EcGmAMHSt
σσσσ ++=
2
24
1
z
DIt σ
σ=
LW 2016
Full Diallele (all selfed and reciprocal crosses are made)Incomplete Diallele – no selfed crosses Incomplete Diallele – no selfed, no reciprocal crosses
Mating Designs
DIALLELS: A group of individuals (N) are mated to the same set of individuals (N) to produce n offspring
ijkijjiijk wsggz ++++= µ
)deviations iancefamily var-(withinerror residualw
th-j andth -i parents ofability combining specific s
th-jparent ofability combining general g
th-iparent ofability combining general g
mean populationµ
damth -j and parentsth -j andth -i thefrom offspringth -k theof phenotypez
ijk
ij
j
i
ijk
=
=
===
=
LW 2017
Mating Designs
DIALLELS: A group of individuals (N) are mated to the same set of individuals (N) to produce n offspring. Analysis for incomplete diallelewithout selfed or reciprocal crosses.
ijkijjiijk wsggz ++++= µ
SoV df SS EMS
GCA N-1
SCA N(N-3)/2
Sibs (n-1)[N(N-1)/2-1]
( ) 222 2 GCASGAw Nnn σσσ −++
2
24
1
z
AGCAt
σσ=
∑ <−=
kji ijijkw zzSS,
2. )(
2wσ
( ) GCAji ijSCA SSzznSS −−= ∑ <2
....
( ) ( )∑ −−−=
i iGCA zzN
NnSS 2
.....
2
2
1
22SCAw nσσ +
2
24
1
z
DSCAt
σσ=
LW 2018
Genotypic Means
GENOTYPIC MEANS:
The environment includes non-genetic factors that affect the phenotype, and usually has a large influence on quantitative traits. o Micro-environment. Environment of a single plant. Need to be
controlled with experimental design.o Maco-environment. Environment associated to a location and
time. GxE is the norm and not the exception in plants. Therefore defining the target environments is a crucial part in plant breeding, both for variance component estimation and identifying superior genotypes.
ijklijklijjiijkl pGEEGz ε++++=
B819
Micro -environmental variation
HERITABILITY: For related individuals, heritability can be calculated as previously shown. The previous calculation assumes individual plants are measured, and heritability on an individual plant basis is estimated. However, because quantitative traits measured on individual plants have large non-genetic effects, heritability on a mean basis is higher.
22222weFEFP σσσσσ +++=
Individual plant basis, n=1, r=1, e=1
2222
2
2
2
2
ˆˆˆˆ
ˆ1
4
ˆ
ˆ1
4
ˆweFEF
FP
p
FP
F
FFh
σσσσ
σ
σ
σ
+++
+=
+=
Plot basis, n=n, r=1, e=1
n
hw
eFEF
F
p
FF 2
222
2
2
22
ˆˆˆˆ
ˆ
ˆ
ˆ
σσσσ
σσσ
+++==
nw
eFEFP
22222 σσσσσ +++=
Plot basis, multiple env, n=n, r=r, e=e
ernere
hweFE
F
F
p
FF 222
2
2
2
22
ˆˆˆˆ
ˆ
ˆ
ˆ
σσσσ
σσσ
+++==
ernereweFE
FP
22222 σσσσσ +++=
Holland et al., 2010, B620
Experimental Design and Analysis
So we need good experimental designs to test genotypes!
FISHER’S PRINCIPLES. Statistical theory for efficient estimation (i.e. unbiased and of minimum variance) of treatment mean differences are based on 3 principles:
– Randomization , random assignment of treatments to experimental units provides valid estimation of experimental error, unbiased comparisons of treatments, and justifies statistical inference.
– Replication allows estimation of experimental error variance, and decreases mean variances (i.e. variance of a mean = variance of the observation divided by the number of replications).
– Local control is the grouping of homogenous experimental units. Well chosen blocks will decrease error variance.
21
Experimental Design and AnalysisCLASSIFICATION OF EXPERIMENTAL DESIGNS: 1. Experimental Units.
1. Homogenous –Complete Randomized Design (CRD)2. Heterogenous in one way – Randomized Complete Block Design (RCBD)3. Heterogenous in more than one way – Latin squares or latinized designs.
2. Large number of treatments.1. Incomplete Block Designs (IBD or Alpha)2. Unreplicated experiments (or Federer)
3. Modeling (post-blocking, spatial analysis)
CRD RCBD IBD
13
12
243
4 14
42
213
3 14
42
213
3
ijiijy εαµ ++= ijjiijy εβαµ +++= 22ijkjkjiijky εγβαµ ++++= )(
Complete Randomized Design (CRD)
TREATMENT ASSIGNMENT: Each treatment is assigned atrandom to the experimental units.� Experimental unit: one plot.
RICE EXAMPLE. YIELD COMPARISON OF 4 RICE CULTIVARSTreatments: 4 cultivars
Replications: 4 homogenous experimental plots per treatment
Experimental design: CRD
Dependent variable : Y = Grain yield (Kg ha-1)
1 2 3 4
5 6 7 8
9 10 11 12
13 14 15 16
33
33
2 22
2
11
11
444
4
EXPERIMENTAL PRINCIPLES1. Randomization2. Replication3. Local control
23
Complete Randomized Design (CRD)
ijiijY εαµ ++=
(residual)error alexperimentε
ntth treatme-i theofeffect α
mean populationµ
repth -j on thent th treatme-i theof responseY
ij
i
ij
==
=
=
ASSUMPTIONS:
1. TO THE MODEL:• Is correct (in relation to the experimental units)• Is additive
2. EXPERIMENTAL ERRORS:• Are random variables• εij ~ N• E(εij) = 0 for every i, j• V(εij) = σ2 for every i, j• Are independent
3. “BY DEFINITION” ααααi = µi - µ
24
Complete Randomized Design (CRD)
T1 T2 T3 T4
47 50 57 54
52 54 53 65
62 67 69 74
51 57 57 59
53 57 59 63
SoV df SS MS F p
Geno (t-1) SST/gl(T) MST/MsE
Error t(r-1) SSE/gl(E)
Total tr-1 SSTOT/gl(TOT)
2... )(∑ −=
iiT yyrSS
∑ ∑= =
−=t
i
r
jiijE yySS
1 1
2. )(
2.. )(∑ −=
ijijTOT yySS
SoV df SS MS F p
Geno 3 208 69.3 1.29 0.323
Error 3 646 59.8
Total 15 854
58
25
Randomized Complete Block Design (RCBD)
TREATMENT ASSIGNMENT: Within a block, each treatment isassigned randomly to the experiemntal units within a block.Randomization in each block is independent.� Experimental unit: one plot.
WHEAT EXAMPLE. PLANT HEIGHT OF 5 ADVANCED LINES OF WHEAT IN 5 BLOCKS
Treatment: 5 advanced wheat linesBlock: 5 diferent blocksExperimental design: RCBDDependent variable: Y = cm
3 3
3
3
2 22
2 11
11
4
44
4
3
2
14
55
5
5
5B 1 B 2 B 3 B 4 B 5
EXPERIMENTAL PRINCIPLES1. Randomization2. Replication3. Local control
26
Randomized Complete Block Design (RCBD)
ijjiijY εβαµ +++=
ASSUMPTIONS:
1. TO THE MODEL:• Is correct (in relation to the experimental units)• Is additive• There is NO treatment by block interaction
2. EXPERIMENTAL ERRORS:• Are random variables• εij ~ N• E(εij) = 0 for every i, j• V(εij) = σ2 for every i, j• Are independent
3. “BY DEFINITION” ααααi = µi - µ
(residual)error alexperimentε
blockth -j theofeffect
ntth treatme-i theofeffect α
mean populationµ
blockth -j on thent th treatme-i theof responseY
ij
j
i
ij
=
==
=
=
β
27
Randomized Complete Block Design (RCBD)
T1 T2 T3 T4 T5
B1 8 10 8 9 10 9
B2 7 9 8 8 9 8.2
B3 6 8 9 8 8 7.8
B4 6 7 9 8 7 7.4
B5 7 9 10 7 9 8.4
6.8 8.6 8.8 8.0 8.6 8.16
SoV df SS MS F p
Block (r-1) SSB/gl(B)
Geno (t-1) SST/gl(T) MST/MSE
Error (t-1)(r-1) SSE/gl(E)
Total tr-1
2... )(∑ −=
iiT yyrSS
∑ ∑= =
−−−=t
i
r
jjiijE yyyySS
1 1
2.... )(
2.. )(∑ −=
ijijTOT yySS
SoV df SS MS F p
Block 4 7.36 1.84 2.77 0.0637
Geno 4 13.36 3.34 5.02 0.0081
Error 16 10.64 0.665
Total 24 31.36
2... )(∑ −=
jjB yytSS
28
Number of treatments
HOW TO DEAL WITH HIGH NUMBER OF TREATMENTS?
1. STRATIFICATION: Group genotypes with similar characteristics(maturity, color, family), compare within groups. NO BETWEEN GROUP COMPARISONS.
2. PRODUCE HOMOGENOUS EXPERIMENTAL UNITS: Make everyeffort to homogenize experimental area (look for soil similarity, fieldconditions to reduce variation, choose seeds of similar vigor).
3. USE REPEATED CHECKS: You may use checks in a systematicway to control or model soil heterogeneity.
4. EXPERIMENTAL DESIGN WITH SPATIAL CONSIDERATIONS. Use experimental designs that include a large number of treatmentswhile controling variability (i.e. alpha designs, unrep, etc.).
29
Incomplete Block Designs (IBD)
IBD CLASIFICATION.1. Based on their balance.
A. Balanced: each treatment is compared with another one the samenumber of times in an incomplete block (usually once). A large numberof reps are needed.
B. Partially unbalanced: Not all pair of treatments are compared inan incomplete block the same number of times.Precision for mean comparisons is differentdepending on the pair. Statistical analysis morecomplex.
BI 1 BI 2 BI 3 BI 4 BI 5 BI 6 BI 7 BI 8 BI 9 BI 10 BI 11 BI 12
1 4 7 1 2 3 1 2 3 1 2 3
2 5 8 4 5 6 5 6 4 6 4 5
3 6 9 7 8 9 9 7 8 8 9 7
BI 1 BI 2 BI 3 BI 4 BI 5 BI 6
1 4 7 1 2 3
2 5 8 4 5 6
3 6 9 7 8 930
Incomplete Block Designs (IBD)
2. Based on their structure.A. Resolubles: incomplete blocks could be grouped in complete
replications.
B. Non resolubles: incomplete blocks cannot be grouped in completereplications.
BI 1 BI 2 BI 3 BI 4 BI 5 BI 6 BI 7 BI 8 BI 9 BI 10 BI 11 BI 12
1 4 7 1 2 3 1 2 3 1 2 3
2 5 8 4 5 6 5 6 4 6 4 5
3 6 9 7 8 9 9 7 8 8 9 7
BI 1 BI 2 BI 4 BI 5 BI 7 BI 8 BI 10 BI 11
1 4 1 2 1 2 1 2
2 5 4 5 5 6 6 4
3 6 7 8 9 7 8 9 31
Incomplete Block Designs (IBD)SUNFLOWER EXAMPLE. OIL YIELD COMPARISON OF 20 ADVANCED INBRED LINES.
Treatment: 20 sunflower ILExperimental design: IBDResoluble with r=3 and s=5Dependent variable : Y = L ha-1
Incomplete block1 2 3 4 51 2 3 4 5
R1 6 7 8 9 10 11 12 13 14 1516 17 18 19 20
1 2 3 4 5R2 7 8 9 10 6
13 14 15 11 1219 20 16 17 18
1 2 3 4 5R3 8 9 10 6 7
15 11 12 13 1417 18 19 20 16
TREATMENT ASSIGNMENT:Each treatment is assigned randomly to the experimental units in the first rep. In the following reps, restrictions in the randomization are conducted such that each pair of treatment is compared the same number of times within an incomplete block.
32
Incomplete Block Designs (IBD)
ijkjkjiijkY εγβαµ ++++= )(
(residual)error alexperimentε
repth -j in theblock with incompleteth -k theofeffect
repth -j theofeffect
ntth treatme-i theofeffect α
mean populationµ
block incompleteth -k theand repth -j on thent th treatme-i theof responseY
ijk
k(j)
j
i
ijk
=
=
==
=
=
γβ
33
Field Designs and Heritability
ijkijkijkjijk PEPEy εβµ +++++= )(
SoV df EMS
Environment (e-1)
Block(Env) e(r-1)
Progeny (n-1)
ProgenyXEnv (n-1)(e-1)
Pooled error (n-1)(r-1)e
ProgenyPEεProgeny reVrVVMS ++=
FACTORIAL DESIGNS (HS, FS, RIL, DH, Clones)
PEεPE rVVMS +=
εError VMS =
( )D
2
AProgeny V4
1V
2
1V:FS
FF +++=
AProgeny V4
1V:HS
F+=
AProgeny V2V:RIL =ernere
hweFE
F
F
p
FF 222
2
2
2
22
ˆˆˆˆ
ˆ
ˆ
ˆ
σσσσ
σσσ
+++==
B634
Other Considerations (Design )
1. PLOT SIZE:a. Big enough to have plants growing normally (i.e. one-plant does
not work in crops but works fine for trees).b. Big enough to represent trait variation (i.e. you might need larger
plots for yield than for maturity).c. Not too big as to have considerable within-plot variation (i.e.
increasing plot size increases within-plot variation).d. Balance between cost of increasing plot size and increasing
number of plots.
EXAMPLE. Mohammad et al. (2001) compared plot size and number of replications to detect differences in wheat. Bigger plots require less reps.
r=2
r=3
r=4
r=5
Difference to be detected
Plo
t siz
e
35
Other Considerations (Design )
2. PLOT SHAPE: a. Balance between plot border and plants within plot (i.e.
rectangular plots have more border than squared ones).b. Take into account gradients (i.e. if unidirectional gradient like
fertility long and narrow plots might be better).
EXAMPLE. Haapanen(1992) compared plot size and shape for height in pines
36
Other Considerations (Design )
3. ROW vs. HILL PLOTS:
EXAMPLE. Tragoonrung et al., 1990 compared row and hill-plots for different traits in barley. Hill-plots are ok for most traits but not for yield.
37
Other Considerations (Design )
4. BORDER ROW. It is a good idea to include a row of plants as a border row to avoid having plots in the margins of the experiment performing differently due to lack of competition.
5. OPERATORS NOISE. If measurements on the experiment will be performed by different people it is a good idea to consider operators noise by having different people measuring for example in different blocks.
6. EARLY TESTING. Small amount of seeds are available in early generations. Therefore replicating is a challenge. Taking this into consideration when deciding which traits will be measured early is important.
7. OTHER DESIGNS AND SPATIAL CORRELATIONS. More efficient designs exist for testing large numbers of genotypes in fields that might not be completely heterogeneous. Additionally, spatial corrections might improve estimations of genotypic means. 38
Other Considerations (Estimation )
1. MIXED MODELS. Careful interpretation of Expected Means Squares should be exercised especially when using Mixed Models. EMS is different if factors (i.e. environments, genotypes, etc. ) are defined as random or fixed effects:
a. E, G, GxE random:
within env:
a. G, GxE random, E fixed:
across env.:
222)( GIeG nNnEMS σσσ ++=
22)( GeG nNEMS σσ +=
( )222
222
2 IGe
IG
Fh
σσσσσ
+++=
( )2222
22
2 EIGe
G
Fh
σσσσσ
+++=
LW 2239
Other Considerations (Estimation )
2. NON-BALANCED DESIGNS . When designs are not balanced due to their design, missing plots or missing data, MS calculations are not straight-forward. There are four ways to estimate MS.
MIXED MODELS - INCOMPLETE BLOCKS – UNBALANCED
With unbalanced data, mixed models, or correlation among genotypes, variance component estimation of heritability is not related to the gain from selection. Using the concept of “effective error variance” (Cochran and Cox, 1957), Cullis et al (2006) proposed to use a generalized method for heritability:
22
21
G
BLUPhσ
υ−=
Piepho and Mohring 2007
BLUP twoof difference a of ncemean varia:BLUPυ
40