PertanikaJ. Trop. Agric. Sci. 18(3): 201-207(1995) ISSN: 0126-6128© Universiti Pertanian Malaysia Press

Component Analyses and their Implicationon the Breeding of Soya Bean (Glycine max (L.) Merr)

OJ. ARIYODepartment of Plant Breeding & Seed Technology

University of AgricultureAbeokuta, Nigeria

Keywords: soya bean breeding, correlation, factor analysis, selection index

ABSTRAK

Sepuluh dri yang digabungkan dengan hasil biji benih dalam 20 genotip kacang soya, Glycine max (L) merr,telah dianalisis menggunakan teknik-teknik korelasi, regresi dan analisis-analisis faktor. Korelasi linearberganda (R) 0.99 dengan koefisien penentuan 0.97 adalah direkodkan. Analisis regresi menunjukkan bahawahari-hari mengeluarkan bunga, hari-hari kematangan, ketinggian untuk menuai, bilangan dahan setiap pokok,bilangan biji setiap lenggai, pembintilan dan panjang lenggai yang banyak dihasilkan kepada variasi adalahdisebabkan oleh regresi. Hanya ketinggian untuk menuai dan hari-hari kematangan, secara positif dansignifikan berkaitan dengan hasil biji benih. Analisis faktor menghasilkan keputusan yang sama dengananalisis regresi dan kolerasi dri-dri tanaman. Empat faktorpertama diambil kira untuk 82.53% variasi dalamstruktur bergantungan.

ABSTRACTThe characters assodated with seed yield in 20 genotypes of soya bean, Glycine max (L) Merr. were analysedusing techniques of correlation, regression and factor analyses. The multiple linear correlation (R) of 0.99 witha coeffident of determination of 0.97 was recorded. The regression analysis indicated that days toflowering} daysto maturity, hdght at harvest, number of branches per plant, number of seeds per pod, nodulation and pod lengthcontributed substantially to the variation due to regression. Only hdght at harvest and days to maturity werepositively and significantly correlated with seed yield. Factor analysis produced a similar result to those of plantcharacter correlation and regression analysis. The first four factors accounted for 82.53% of the variation inthe dependence structure.

INTRODUCTIONSeed yield in soya bean is a complex characterinfluenced by the interplay of many othercharacters. Knowledge of the relationship ofyield with its main components is important inplant breeding, particularly for indirect selectionfor quantitative traits, such as seed yield, thatexhibit low heritability. Beside the yieldcomponents, physiological and morphologicalcharacters of soya bean plants are known to playa major and interdependent role in determiningseed yield (Denis and Adams 1978; Bartual et al1985).

Correlation studies between characters andthe use of multivariate analysis to determine therelative contribution of different characters to

the total variation are of great value indetermining the most effective breedingprocedures (Bhatt (1976) in wheat; Ghaderi etal (1979) in mung bean; Broich and Palmer(1980) and Ariyo (1995) in soya bean; Ariyo(1991a, 1991b, 1993 in okra).

The breeder has a number of desirablecharacters in mind when carrying out selection.To maximize improvement in the character ofchoice, selection is generally appliedsimultaneously to several other characters thatinfluence the character of choice. Falconer(1960) reported that the most rapid improvementof economic value was expected from selectionapplied simultaneously to all the components.Use of selection index gives adequate weight to

OJ. ARIYO

each of the desirable characters identified. Hazeland Lush (1942) and Falconer (1981) reportedthat selection based on such an index was moreefficient than selecting individuals on variouscharacters. In addition, Ariyo (1991a) reportedthat the selection index should be used inconjunction with yield data preferably obtainedacross contrasting environments to producevaluable results.

The objectives of this study were:(a) to determine the relative importance of

various characters of soya bean and therelationship among them, and

(b) to construct selection indices for seed yield.

MATERIALS AND METHODS

Twenty genotypes, consisting of early, mediumand late maturing varieties of soya bean fromthe International Institute of Tropical Agriculture(IITA), Ibadan were grown in a randomizedcomplete block design with three replications.The planting was done at the University ofAgriculture, Abeokuta, in July, 1991. Each entrywas grown in four-row plots of 6 x 3m but onlythe competitive plants in the two inner rowswere observed. Following planting, a mixture of4 I Galex and 1 1 Gramaxone in water wassprayed per hectare to control weeds. Subsequentweeding was done manually.

The number of days to flowering wasrecorded as the date the plants of a genotypeattained 50% flowering. Maturity was when thepods had turned brown just before shattering.Shattering was taken as the proportion of guardrows that shattered two weeks after the twoinner competitive plants had been harvested todetermine seed yield; nodulation was assessedon the size and number of nodules at full bloom.Lodging was scored on the proportion of theplants that fell down at harvest. Genotype means,averaged across replications, were used forstatistical analysis.

Phenotypic a coefficients of correlation werecalculated among all the characters evaluatedfollowing the procedure of Steel and Torrie(1968). Step-wise multiple regression analysis(forward selection) was performed as outlinedby Draper and Smith (1966) by which themultiple-regression equation and multiplecoefficient of determination (R2) were obtainedby adding independnet variables, one at a timedepending on their relative importance, indetermining dependent variables. Analysis was

terminated when the proportion of dependentvariance explained by adding each of theremaining variables was not significant at 0.05level of probability. In this study, seed yield wasfitted as a linear function of the other tencharacters. The sequential contribution of eachcharacter to the total variation in seed yield wasdetermined by the forward selection procedure.On the basis of the relative importance of eachcharacter to seed yield, selection indices werecontructed.

Data were also subjected to factor analysisaccording to the procedure of Cattell (1965).The analysis produced factor loading as well ascommunality for each character from thevariance-covariance matrix of the 20 genotypes.The factors were ranked on the basis of themagnitude of variability explained in thedependent structure. When the contribution ofa factor to the variability was less than 10%, theprocess was terminated. The particularcombination of variables that form a factoraccounted for more of the variance of the dataas a whole than any other linear combinationvariable. Therefore, Factor 1 was the bestcombination of the linear relationships in thedata. Factor 2 was the best linear combination ofvariables that accounted for most of the residualvariance after the effect of Factor 1 had beenremoved. Subsequent factors contributedprogressively less to the total variance.

Communality is the amount of variance of avariable accounted for by all factors collectivelyand it is the R2 value obtained by regressing avariable on all other variables in the model (Leeand Kaltsikes 1973; Eckert and Westfall 1975).

RESULTS

Table 1 presents the correlation coefficientsbetween the various characters evaluated. Onlyheight at harvest was strongly correlated withseed yield (r=0.65), while number of days tomaturity was moderately correlated with seedyield (r=0.44). Height at harvest was also positivelycorrelated with days to flowering (r=0.78) andmaturity (r=0.57) while pod shattering wasnegatively correlated with days to maturity(r=0.75) and height at harvest (r=0.58). Numberof branches per plant exhibited a positiverelationship with days to flowering (r=0.59), daysto maturity (r=0.55), height at harvest (r=0.58),and lodging at harvest (r=0.44) but was negativelycorrelated with pod shattering (r^0.62). On the

202 PERTANIKAJ. TROP. AGRIC. SCI. VOL. 18 NO. 3, 1995

TABLE 1Evaluation of correlations among the eleven characters of soya bean

V

3

Characters Seed Yield/ Days to Days to Height at Lodging at Pod Number of Number of Number of Nodulationplot Flowering Maturity Harvest Harvest Shattering Branches/ Pods/Plant Seeds/Pod(g) (cm) Plant

Days toFlowering 0.30Days toMaturity 0.44* 0.35

g2o

G

o00

§00

s

Harvest (cm)

Lodging atHarvest

Shattering

Number ofBranches/Plant

Number ofPods/Plant

Number ofSeeds/Pods

Nodulation

Pod Length(cm)

0.65**

0.09

-0.35

0.22

-0.10

0.34

0.28

0.29

0.78**

0.22

-0.30

0.59**

0.52*

0.08

0.12

-0.14

0.57**

0.34

-0.75**

0.55*

0.38

0.07

0.46*

-0.45*

0.06

-0.58**

0.58**

0.29

0.19

0.28

0.10

-0.01

0.44*

0.61**

-0.49*

0.01

-0.10

-0.62**

-0.28

0.00

-0.53*

0.14

0.80**

-0.13

0.44*

-0.27

-0.36

0.39

-0.31

-0.19

-0.27

8

i

03

2z

s

-0.24

* Significant at P = 0.05; ** Significant at P = 0.01

OJ. ARIYO

other hand, number of pods per plant showed acorrelation of r=0.52, 0.61 and 0.80 with days toflowering, lodging at harvest and number ofbranches per plant respectively. Nodulationwas positively correlated with days to maturityand number of branches per plant, butnegatively correlated with shattering. Podlength was negatively correlated with days tomaturity (r=-0.45).

Four factors were obtained by factor analysis(Table 2); together these accounted for 82.53%of the variance for all the 11 characters. Thisimplies that the factor-analysis model used in thisstudy was effective in illuminating the uniquevariance of each variable. Communalities rangedfrom 0.6728 to 0,9360. Generally, factor loadingsof 0.7 - 0.9 would be considered high loadings andthose from 0 to 0.2 as low loadings (Denis andAdams 1978). In this study, however, for thepurpose of interpretation only characters exhibitingfactor loadings of 0.5 or larger were consideredimportant and no character was loaded on morethan one factor. It should be noted that the choiceof loading of 0.5 or greater is arbitrary and doesnot imply biological significance. Biological

interpretation of factors depends largely on thegenotypes evaluated, the particular sets of charactersmeasured, and how well the researcher understandsthe biology of the organism (Fakorede 1979).These are limitations of factor analysis.

Factor 1, which accounted for 39.44% ofthe total variance, contained days to flowering,days to maturity, height at harvest, shattering,number of branches per plant and number ofpods per plant. The influence of a factor on atrait is determined by the square of the factorloading for that trait (Lee and Kaltsikes 1973).Therefore, Factor 1 accounted for 33% of thevariance due to seed yield in this study. Inaddition, Factor 1 related essentially to thephysiological aspect of the crop. Apart fromnumber of pods per plant, other charactersloaded in Factor 1 had direct bearing on acrop's growth and development.

Lodging at harvest and number of seeds perpod were grouped under Factor 2, and thisfactor accounted for 19.34% of the variance inthe data as a whole. Factor 3 comprised onlypod length, while Factor 4 was nodualtion; mostcharacters in Factor 1 were correlated with each

TABLE 2Evaluation of communalities and the factor loadings of eleven characters

Traits

Factor 1Days to FloweringDays to MaturityPlant HeightShatteringNumber of Branches/PlantNumber of Pods/Plant

Factor 2Lodging at HarvestSeed Yield/PlantSeeds/Pod

Factor 3Pod Length

Factor 4NodulationPercentage of Total VariationCumulative PercentageEigen Value

Communalities

0.83800.76940.93310.79380.82580.8759

-0.61020.78850.8923

0.9359

0.7518

1

0.69400.81300.7724

-0.75300.86780.7017

0.41360.4371

-0.0681

-0.2913

0.579439.4439.44

4.33

Factor

2

0.10880.12520.4738

-0.2327-0.2480-0.5956

-0.23270.67150.7234

0.2312

-0.026019.3458.78

2.13

Loadings

3

0.2378-0.26800.29290.1695

-0.00340.0384

0.34020.3542-0.4168

0.8469

-0.248413.0771.85

1.43

4

0.5371-0.14410.16210.37920.10610.1654

0.1172-0.14500.4333

-0.2834

-0.594810.6882.53

1.17

204 PERTANIKAJ. TROP. AGRIC. SCI. VOL. 18 NO. 3, 1995

COMPONENT ANALYSES AND THE BREEDING OF SOYA BEANS

other while correlations between characters wereobserved across factors.

Table 3 presents linear regression analysisof yield components. The analysis showed thatall except lodging at harvest, shattering andnumber of pods per plant contributedsignificantly to the total variation due toregression. The multiple linear correlationcoefficient between seed yield and the other tencharacters was R=0.99 given a coefficient ofdetermination of 0.97.

Table 4 gives the selection indices, usingvarious character combinations and their relativeeffectiveness. The high value of the coefficientof determination indicated that the tencharacters accounted for most of the variation

TABLE 3Multiple linear regression analysis of

components of seed yield in soya bean

Source

RegressionDays to floweringDays to maturityHeight at harvest (cm)Lodging at harvestShatteringNumber of branches/podsNumber of pods/plantNumber of seeds/podNodulationPod length (cm)

DF

101111111111

MS

205,820'*194,393**271,429**673,161**46,32555,814

179,221**4,476

188,524**252,840**192,021**

in seed yield. The step-wise regression indicatedthat nine characters accounted for 97% of thetotal variation. The remaining character,shattering, did not meet the a • 0.5 significancelevel for entry into the model. Plant heightalone accounted for 42.46% of the total variationdue to regression. By inclusion of number ofdays to flowering, 53.62% of the total variationwas explained. Adding number of branches perplant, lodging, number of seeds per pod,nodulation and pod length, 92.47% of the totalvariation was accounted for. Inclusion of days tomaturity and number of pods per plant separatelyoccasioned a marginal increase of about 2% ineach case. It is generally accepted that correlationmay not always adequately indicate therealtionship among variables. This is whymultivariate rather than bivariate statisticalmethods are preferred.

DISCUSSIONThe high correlation between days to flowering,days to maturity and height at harvest suggeststhat short varieties flowered earlier and maturedearlier. However, high correlation betweenheight at harvest and seed yield indicates thattall varieties yielded more than short varieties.The correlation between branches per plant,pods per plant and days to flowering indicatesthat late flowering plants produced more seed-bearing branches, which were possiblyresponsible for correlation between height atharvest and seed yield as only tall varietiescould accommodate many branches. Lateflowering varieties matured late and grew taller,

TABLE 4Selection indices in soya bean for yield. Multiple regression equation

Yl =-11.62 HY2 = 12.02 -Y3 = 13.07 +Y4 = 6.86 +Y5 = -8.29 +Y6 = -24.47 -Y7 = -96.54Y8 = -222.87Y9 = -243.86

h 0.45X,0.74X,- 0.86X20.85X, - 0.69X2 - 1.64X,

0.9X, - O.82X2 - 2.61X3 + 12.99X4

0.85X1 - 0.84X2 - 2.44X3 + 18.54X4 +h 0.76X, - O.62X2 -+ 0.43X, - 0.15X2+ 0.06X, + 0.35X2- 0.08X, + 0.64X2

3.62X3 + 23.39X4 4- 2.83X3 + 17.35X4

- 2.54X, + 22.06X^- 1.41X3 + 24.80X,

4.95X5

- 6.85X5 + 2+ 11.92X5 H, + 13.17X5

, + 13.01X5

U3X6

- 4.92X6 +• 4.83X6 •+ 5.39X6

9.70X7

f 15.73X7 4+ 16.38X7 H

1.03X8

- 1.01X, + 13-OlX^

R2(%)

42.4653.6259.0866.7573.1082.0492.4795.1397.10

Xj= Height at harvest; X̂ • Days to flowering; Xj = Number of branches/plant; X4 = lodging;X5 = Number of seeds/pod; X6 = Nodulation; X7= Pod length; Xs = Days to maturity; X^ = Number of pods/plant

PERTANIKAJ. TROP. AGRIC. SCI. VOL. 18 NO. 3, 1995 205

OJ. ARIYO

besides producing more branches. That suchvarieties are susceptible to lodging suggests thata compromise must be struck between highyield and loss due to lodging at harvest.Correlation between pods per plant and days toflowering indicated that late flowering varietieswhich were also late maturing produced morepod-bearing branches. Of interest is thecorrelation between pods per plant and lodgingat harvest. This implied that top-heavy varietieswere likely to lodge under the weight of the pods.

The study identified days to flowering, daysto maturity, number of branches per plant, heightat harvest and pods per plant, some of thesecharacters correlated among themselves, asimportant yield components in soya bean.

The results of regression analysis werecomplementary to the correlation studies byhighlighting the relative weight of the variouscharacters to the total variation. Height atharvest contributed the highest variance to yieldfollowed by days to maturity, days to flowering,pod length and number of branches per plantin diminishing order of contribution. Similarly,the factor model identified the contribution ofall these characters to the dependence structure.Days to flowering, days to maturity, height atharvest, number of branches per plant, numberof pods per plant and pod length had highcommunalities, suggesting their importance inthe dependence structure.

The various selection indices showed thatheight at harvest alone accounted for 42.46% ofthe total variation due to regression,demonstrating a multiple correlation coefficient(R) of 0.66. A selection index based on height atharvest and days to flowering gave the coefficientof determination of 53.62% of the total variable.The coefficient of determination continued toincrease as more characters were entered, andterminated at R2 of 97.10%. However, there wasa marginal decrease in R2 value as morecharacters were progressively added, indicatingthat earlier characters were more important thanlater entries. Although number of pods per plant,days to maturity and pod length were rankedhigher in importance than lodging at harvest,nodulation and number of seeds per pod, thesewere nonetheless also important components ofpod yield.

In breeding soya bean for high yield,therefore days to flowering, which to a largeextent determines days to maturity and height at

harvest, is a premium character. Other characterssuch as number of branches per plant, podlength and nodulation should also be considered.Similar observations were also reported by Denisand Adams (1978). Since it is better to applyselection generally to several characters thatinfluence yield, a knowledge of the inheritanceof the various characters in the selection indexis required.

This study gives an insight into the associationamong traits observed in a set of genotypes. Astatistical demonstration of association, whetherby correlation, regression or factor analysis,however, does not provide information on thecausative agents (genetic, physiological,morphological or environmental) (Fakorede1979). Genetic analysis is necessary whenever thedeterministic relationship is of interest.

Factor analysis has the limitations of thearbitrary and subjective nature of interpretationof factors and the dependence on the numberand loadings of factors on the particular set ofgenotypes and variables. Despite theselimitations, however, its data-reducing capacitygives it an advantage over correlation andregression analyses.

ACKNOWLEDGEMENT

The author acknowledges the contribution ofthe Grain Legume Improvement Programme,International Institute of Tropical Agriculture(IITA), Ibadan, Nigeria towards the successfulcompletion of this study.

REFERENCES

ARIYO, OJ. 1991a. Effectiveness of classical selec-tion index in discriminating desirable varietiesof okra (Hibiscus esculentus). Indian]. Agric. Sri.60: 793-795.

ARIYO, OJ. 1991b. Regression analysis of pod yieldand yield components in okra (Abelmoschusesculentus (L). Moench)./ Agric. Sri. TechnoL 1:72-74.

ARIYO, OJ. 1993. Genetic diversity in West Africanokra (Abelmoschus caillei (A. Chev.J stevels -Multivariate analysis of morphological andagronomic characteristics. Genet Res. Crop. Evol40: 25-32.

ARIYO, OJ. 1995. Correlation and path-coefficientanalysis of components of seed yield insoybeans. African Crop. Sri. J. 3: 29-33.

206 PERTANIKAJ. TROP. AGRIC. SCI. VOL. 18 NO. 3, 1995

COMPONENT ANALYSES AND THE BREEDING OF SOY\ BEANS

BARTUAL, R., E.A CARBONELL and D.E. GREEN. 1985.

Multivariate analysis of a collection of Soybeancultivars for southwestern Spain. Euphytica 34:113-123.

BHATT, G.M. 1976. An application of multivariateanalysis to selection for quality characters inwheat. Austral. J. Agric. 27: 11-18.

BROICH, S.L. and R.G. PALMER. 1980. A cluster

analysis of wild and domesticated soyabeanphenotypes. Euphytica 29: 23-32.

CATTELL, R.B. 1965. Factor analysis. An introduc-tion to essentials 1. The purpose of underly-ing models. Biometrics 21: 23-32.

DENIS; J.C. and M.W. ADAMS. 1978. A factor analysisof the plant variables related to yield in drybeans. 1. Morphological traits. Crop. Sci. 18:74-78.

DRAPER, N.R. and H. SMITH. 1966. Applied RegressionAnalysis. 2nd edn. New York: Wiley.

ECKERT, R.T. and R.D. WESTFALL. 1975. The factor

analysis of multivariate data systems Northwest-ern Forest. Tree Improvement Conf. Proc. 22: 41-52.

FAKOREDE, M.A.B. 1979. Inter-relationship amonggrain yield and agronomic traits in a syntheticpopulation of maize. Maydica 24: 181-192.

FALCONER, D.S. 1981. Introduction to QuantitativeGenetics. 2nd edn. London: Longman.

GHADERI, A., M. SHSHEGAR and B. EHDAIE. 1979.

Multivariate analysis of genetic diversity foryield and its components in mung bean. / .Amer. Soc. Hort. Sci. 104: 728-731.

HAZEL, L.N. and J.L. LUSH. 1942. The efficiency of

three methods of selection. / Hered. 33: 393-399.

LEE, J. and PJ. KALTSIKES. 1973. Multivariate statisti-cal analysis of grain and agronomic charactersin durum wheat. Theor. Appl Genet. 43: 226-231.

STEEL, R.G.B. andJ.H. TORRIE. 1968. Principles andProcedures of Statistics. 2nd edn. New York:McGraw Hill.

(Received 26 June 1995)(Accepted 3 January 1996)

PERTANIKAJ. TROP. AGRIC. SCI. VOL. 18 NO. 3, 1995 207