Professional Documents
Culture Documents
The AMMI model first calculates genotypes and environment additive (main) effects
using analysis of variance (ANOVA) and then analyses the residual from this model
(namely the interaction) using principal components analysis (PCA). Fisher (1918)
discovered (ANOVA). The salient feature of ANOVA is that it computes a genotype
deviation (difference from the grand mean) and an environment deviation whose sum
(plus the grand mean) estimates the yield for that genotype in that environment. Hence
ANOVA is an additive model. An ANOVA model leaves the non-additive residual,
namely the genotype environment interaction. Pearson (1901) discovered PCA. He
envisoned a two way matrix with rows and coloumns geometrically. Let us replace the
genetic row and coloumn labels with present contexts genotypes and environments.
Then the salient feature of PCA from the present perspective is that it computes a
genotype score and an environment score whose product estimates the yield for that
genotype in that environment. Hence PCA is a multiplicative model Usually the grand
mean is removed before calculating PCA. More than one PCA axes may be calculated.
Note the conceptual simplicity and similarity of PCA and ANOVA. The salient
difference is that the PCA the genotype and environment parameters are multiplied,
where as in ANOVA they are added.
Williams (1952) and Pike and Silverberg (1952) invented the AMMI model. Being a
combination of ANOVA and PCA, and being true to its name, the additive Main effects
and multiplicative interaction (AMMI) model is an additive and multiplicative model.
Thus AMMI model is a hybrid model, the result is the least square analysis, which with
further graphical representation of the numerical results (biplot anlysis) often allows a
straight forward interpretation of the underlying causes of G X E interaction.
AMMI is applicable and useful for data (i) structured in a two-way factorial, with at least
three rows and three columns replicated or not, (ii) containing one kind of data
quantitative rather than categorical and (iii) fitted by the AMMI model reasonably well as
ordinarily happens when main effects and interaction are both significant. The third
requirement is fulfilled when all the three factors viz., genotype, environment and GX E
interaction are significant in the pooled analysis of variance table.
The AMMI model mathematical equation is
Yger=μ+α+β+Σnλnγgnδen+ρge+tger
First the additive part of AMMI model uses ordinary ANOVA. ANOVA leaves a non-
additive residual, the interaction. Second the multiplicative part of the AMMI model
uses PCA to decompose the interaction into PCA axes 1 to N, and a residual ρge remains if
not all axes are used. These PCA scores are termed as interaction PCA scores or IPCA
scores. Each IPCA axis is assigned G+E-1-2n as degrees of freedom where n is the
number of axis.
The member of the AMMI family with 1 IPCA while relegating all higher axes to the
residual is denoted AMMI 1., while AMMI 2 retains 2IPCA axes and so on. In general
AMMIN denotes the AMMI model with IPCA axes 1 to N. AMMIO has no IPCA axes
and is identically ANOVA.
ANOVA for Stability- AMMI model
Source df SS Ms F ratio
Repln r-1
Total (ger-1)
Treatment (ge-1)
Genotype g-1
Environment e-1
Interactions (g-1)(e-1)
IPCA1
IPCA2
Residual
Error (r-1)(ge-1)
If IPCA MS are significant and residual Ms is non-significant, the steps may be continues
for development of biplot. The AMMI biplot is developed using genotype and
environment mean (on X axis) and the respective IPCA values (on Y axis).
Interpretation of Biplot:
Displacements among the X axis indicate differences in main (additive) effects,
where as displacements along the ordinate (y-axis) indicate differences in
interaction effects.
The main effect for genotypes reflects for breeding advances (superiority) and
main effects of environment reflects the overall superiority of the environment.
From the values of mean and IPCA 1, the genotypes are classified under four
distinct classes
Class I: Genotypes with high mean and positive IPCA 1
Class II: genotypes with high mean and negative IPCA1
Class III: genotypes with low mean and negative IPCA1
Class IV: Genotypes with low mean and positive IPCA1
Sometimes there is interest in more than just the first IPCA axis. Accordingly a different
type of biplot, having IPCA 1 on the X axis and IPCA 2 on the ordinate may be
developed. This biplot does not show the additive main effects, but by showing two
IPCA axes, it is highly informative on interaction. When IPCA 2 is sizable and
significant such a graph is a useful to the usual biplot.
CASE STUDY: AMMI analysis of farmers’ participatory variety selection trials in
groundnut (Arachis hypogaea L.)
Conclusion:
All genotypes recorded significantly higher pod yield than the local cultivar TMV
1 (G1).
The local cultivar TMV 1 recorded low interaction with environments.
The genotype ICGV 87846 (G5) recorded high pod yield but with high
interaction. However this genotype has similar type of interaction with four out
of nine villages.
Hence this genotype can be recommended in the villages namely Kothur (E5),
Pranthagam (E7), Chinnathambi palayam (E8) and Karumapuram (E9).
The genotype ICGV 98369 (G6) which followed ICGV 87846 (G5) for pod yield
showed medium level of interaction. These genotypes also can be recommended
in all the above villages to improve the productivity of groundnut in Namakkal
District.
1600 G5
E6 E8 E7
1400
G6
E3 G4
E5 E9
Pod yield (kg/ha)
1200
-1.00 -0.80 -0.60 -0.40 G7G2 -0.20 G1 0.00 0.20 0.40 0.60 0.80 1.00
G3
E2
E1 1000
G8
800
E4
600
IPCA 1 score