You are on page 1of 8
veld 2 est Eng mY, al of 378. WY IE AEN | MOGANG OA DECADE AG DIC TON LEST MAC ORY. ent by Comparison of nonparametric and estimates to evaluate alf. CHEMIE DE mera teu) J. Genet. & Breed. $0: 67-74 (1996) arametric stability a cultivars G. Yue!, W.F. Heer’, J.L. Moyer’, L.D. Maddux’, ELL. Sorensen’ and GH. Liang* * Liaoning Academy of Agricultural Sciences, Statistical Laboratory, Shenyang, China, 2 South Central Experiment Field, Hutchinson, KS 67501, * Southeast Kansas Experiment Station, Parson, KS 67357, ‘ Kansas River Valley Experiment Field. Topeka, KS 66618, S USDA, Kansas State University, Manhattan, KS 66506-5501 and * Kansas State University, Manharian, KS 66506-5501, USA, Fax 913-532-6084 Recsived March 20, 1996 ABSTRACT Five parametric and two nonparametric stability estimates were obtained from alfalfa perfor- ppuance trials conducted in 11 environments (year by location combinations) to characterize six syn thetic cultivars and two germplasm. The coefficient of variation (CV) and regression coefficient (b) ‘were significantly correlated but were not correlated with other stability estimates. Wricke’s ecova- lence (W)), Eberhart and Russells 544, Shukla’ o%, and Nassar and Huhn’ §|” and §"' nonparamet- ric measures were significantly and positively correlated with each other, indicating that they were functionally similar in providing stability estimates to characterize the cultivars and germplasm. The onparametric and W, estimates are simple to calculate and do not require data from individual rep- ications, thus they appear to be better suited to data from performance trials. In addition to being distribution free, the nonparametric model requires no assumptions such as homogeneity of varianc- es, normality, and additivity or linearity of genotypic and environmental effects, “Kanza", KS. 196 germplasm, and “Garst 630° were unstable with high values of W,, 547, 0, §\° and S©, wheceas “Ri ley" and Cimarron” were most stable because they were characterized with low values of the stabil ity estimates. To provide additional guide for selection, a rank-sum approach was applied where both yield (In rank) and the nonparametric stability measures (in rank) were considered jointly. The data indicated that selection for yield only would sacrifice stability to some degree: likewise. selection for stability only would sacrifice a certain amount of yield ‘Key words: Ecovalence, Coefficient of variation, Nonparametric measures, Regression coefficient INTRODUCTION indicates that the mean performance of'a geno- type over all environments is less meaningful Elite lines from a breeding program are usu- | lly tested over multiple environments for yield performance and consistency. Likewise, yields of cultivars or hybrids and their relative ranking or consistency in performance trials form the basis, | ‘or recommendations to growers. With the pres- ence of genotype-by-environment (GE) interac- tion, especially of the crossover type (Baker, '988), the effects of genotypes and environments become nonadditive, indicating that differences in yield performance among genotypes depends on the environment. Significant GE interaction * Corresponding author and complicates the selection of superior culti vars or synthetics across environments. Many parametric stability models have been developed to analyze and interpret crop respons- es to unpredictable growing environments (e.g. Wrickie, 1962; Eaeruart and Russet, 1966; SHUKLA, 1972; LIN et al., 1986; LEON and BECK- ER, 1988; KANG, 1990). Parametric stability mod- els have been applied to various crops, with in- ferences drawal about their effects on selection or cultivar recemmentadion (KANG et al., 1991; JALALUDDIN and HARRISON, 1993; LINK er al., 1994; ’ 68. STELUNG et al., 1994), However, information on stability estimates of slf'% . athetics, cultivars, or germplasm are not > ilable. Also, the non- parametric .cability estimates have rovely been used in cting crop performance. Genotypic mean, regression coefficient and coefficient of variation for yield are heritable and reflect the ability of genotypes to perform over a range of environments, while mean squares for deviation from regression and variances from partitioning of GE interactions that are assign. able to each genotype are measures of stability of a cultivar or hybrid, but may be less heritable (Leox and Becker, 1988). To reduce the risk of yield loss when selec- tion is based on mean yield only, a simultane- ‘ous selection approach for high yield and stabil- ity was proposed (KANG, 1993). A statistic that ntegrated yield and stability was presented (Bacutreppy ef al., 1992; Macart and KANG, 1993), which identified fewer unstable geno- types that when yield alone was considered However, the mean yields of the selected geno- types were lower than those genotypes selected based on yield performance only. To be useful, further refinement of the integrated model ap- pears to be necessary, because high yield is of- ten the most important trait in most breeding programs, Nonparametric measures of _ phenotypic stability, based on the mean of the absolute rank differences of a genotype over the environments and on variance among the ranks over the envi- ronments, provide a simplified but useful alter- native, Individual observations {rom replications are not essential. Only the mean or total of each genotype in each environment is required. Also, no assumption about the distribution of the ob- served values are necessary. Nassar and HORN (1987) suggested the use of two nonparametric measures of stability, $,") (mean difference from ranks) and §,° (variance among the ranks), where smaller values should indicate more stable genotypes. Repeatibility of stability parameters in differ- ent subsets of environments and correlations among various stability parameters from differ ent models have been studied (BECKER, 1981; Le ‘ON and BECKER, 1988; Pam and Kane, 1988; Lin and Binns, 1991). Stability measures computed from data of individual years could be compared and correlated. Coefficients of correlation ob- tained for pairs of stability parameters from sin, gle years may reveal discrepancies or inconsis. tencies among years. 4 In this resort, we analyze ‘ucage yield of six alfalfa cultivars and two germplasm from perf mance trials conducted in 11 environments in, eastern Kansas where alfalfa is commonly grown, We used five statistical models to analyze the data and obtained seven statistics to determine the re lative stability of the cultivars and germplasm. The usefulness of the statistics in determining each entry's yield stability was evaluated and com- pared, and their mutual relationships investigat- ed. In addition, a rank-sum approach (KANG, 1988) was employed, where both yield and the nonparametric stability estimates in ranks were combined for each cultivar and germplasm. MATERIALS AND METHODS: Statistical analyses were peformed using the tu- age yield data from Kansas Agricultural Experiment Station alfalfa performances tests (Kansas Alfalf Performance Tests, 1987-89). The data were collect: ed from each of four counties (Shawnee, Riley, Re- no and Labette) in eastern Kansas. Eleven environ: ments (ail the yearx location combinations, except Riley County in 1989) were sampled. Six cultva: (Kanza, Riley, Cimarron and Garst 630, 636 and 65: and two germplasm (KS 196 and K 82-21) were planted in a randomized complete block design with four replications at each location (Table 1). Individ- ual cuttings (3-4) were combined, and the total for- lage yield from each replication was analyzed. The entries were considered as a fixed variable in the analysis of variatice., TABLE 1 Yields (Meg/ha) by year and averaged across year 1 Den of ike average in alata onesies ‘used for the stabity analysis Year cultivar Average Rank 4 195719881989 7 Kanza 2159 20.79 17451995 Riley 346 20191750 197 KS 196 2252-2018 1773201 a1 140 168718 2133212 1833 d 3 2195 -2ksp 1887209 2221 24d 1835 20.72 2139 1969 «170394 sd 0.05) 029025035 ‘Means doo slg fle Coun fate Stability analyses were performed using five models to obtain seven stability measures: the gen, Sigpic coefficient of variation (CV) (Francis and TUNNENSERG, 1978); regression coefficient 19 Miaion mean square (542) fom Esereagt aad gusset (1966); variance (6) associated with each we | EeSStype from partitioning the GE interaction into : setters (SuURLA, 1972), ecovalence (W,) (WRICKE, E | (eo); and two nonparametric statistics, Sand a (Nassak and HOHN, 1987, HOHN and Nassar, ne 389). asim | | From the nonparametncstabiliy mode S| the ining | mean of the absolute rank differences of a cultivar "nine. | Nor the N environments, is expressed as stigatt Kot ry (MIN D1, d thé ‘wee | where ny i the rank of cultivar é in environment j. $0, the! variance among the ranks over the N envi- ronments, was expressed as: YS (ry ar)? (=D, he for: riment IN here 1, = Dry Cultivar i was considered stable over environ~ ments if its rank was similar over environments. For fcultivar with maximum stability, §,°" = 0; likewise, zero variance also indicated maximum stability "The data set from the 3-yr trial was also divided nto three subsets, consisting of locations in each of the 3 ye (four locations in 1987 and 1988, and three locations in 1989), because seasonal variation is often ‘uch greater than that of locations in Kansas Spearman's rank correlation (Sokal and Rohlf, 1981) between pairs of stability measures was calculated for each of the three subsets. Likewise, correlation coef- ficients of ranks between pairs of stability measures ‘were calculated for the total data set consisting of 11 environments 69 RESULT AND DISCUSSION For the combined analysis (1987-89), the mean squares of entries, eavironment plus (en vironment x entry), and environment (linear) ‘were all higly significant (Table 2), indicating the existence of genetic differences among entries and variantions in environmental effects and GE interactions. The combined analysis also pro- vides further partition of the sum squares due to entry x environment into sum squares due to entry x environment (linear) which is in fact sum squares due to regression and sum squares due to deviations from linearity of response (i.e. sum. squares due to pooled deviation). The entry x en- vironment (linear) was not significant suggest- ing entries did not differ for the regression on the environmental index, as shown for the b, val- ues (Table 3). However, individual deviation from linear regression (542) was higly significant for three entries, Kanza, KS 196, and Garst 630 (Table 3). From Eberhart and Russell's model, stable entries are those with unit regression co- efficients (bj = 1) and deviations (s,;") not signif- icantly different from zero. Thus, five of the en- tries, Riley, K28-21, Cimmaron, Garst 636, and Garst 635 were stable genotypes. The differen- tial response of entries to environments and vari- ation in sq? indicated that recommendation of cultivars should be made on the basis of the in- dividual environments. Because of the unpre- dictable nature of seasonal fluctuations among years, stability measures help to characterize the entries under evaluation. Stability estimates estimated for each cultivar were obtained from the 11 environments (Table 3). Cultivars, Riley and Cimmaron, appeared to bbe the most stable, having the smallest values of TABLE 2 ‘Analysis of variance for each of 3 years and for the combined alfalfa forage yield data 7 1987-98 1987 1988 1939) 5 ‘Source of variation — st - oe a MS a af MS a MS 8 Entry 7 oe 7 06 7 oe 7 O18 3 Environment + (Eav. x Entry) eo Lsert 21.94 57) 122 160.83. U vironment (linear) toouniges 1 aT. Tae 1 asese LR Tae 2 Cultivar xenvironment (linear) 7 940 1 005 1 025 7 005 7 Pooled deviations mon 16 OOS NEBL Pooled error Bt 0.06 84 0.04 a 007 63004 TGignifcant or the 0.01 love of probe. ? 70 TABLE 8 Seven stability measures estimated from alfalfa yield test in 11 environments (four location: Stability estimate 3 3 Entry a ov W, % 1 3 3 so E Kanza 0.160 1.960 1.007 ages 192 3.055 soo 3 Riley o.aa 0.987 0929 -9.025 0.031 2.182 3618 KS 196 0.160 1.608 1.026 0123" 199 3.164 7418 3 K82-21 0.148 1221 Ossi 0.088 0135 2.909 6273 og Cimarron ous 0.560 0965 = -0,002 0.050 2.036 7 Garst 630 0.155 1.600 1023 0125" 9.201 3.091 7 Garst 636, 0.172 1319 Liss, 0.052 0.139 2618 e Garst 635, 0.180 out 1913 0.032 0.009 2.764 ** Highly significant at 001 loel of probabily W, sy?, 6%, §,, and §. The highest-yielding cultivars were usuaily not the most stable (Tables 1 and 3). These data are in contrast with maize (Zea mays L.) (Yur, et al, 1990) and common bean (Duarre and ZimMERMANN, 1995), which showed that a high-yielding hybrid or cultivar can also be stable, Such differences may reflect dif- ferences between species, or the nature of envi- ronments and yield potential. Garst 630 had the highest mean yield from the 3-yr data, but also had high s4?, ¢2, 8," and 8, Riley, on the oth- er hand, had low values of the stability estimates and below average yield. Cimarron ranked third in yield and was stable, appearing to be a logical choice for growers in areas where environmental effects are diverse and unpredictable. The lowest yielding entry, K82-21, was not necessarily un- Stable as indicated by sa? and o?, However, it ap- eared to be less responsive to environment, be- cause it had the smallest b; (0.881) among all en- ties, In other words, a genotype with a regres- sion coefficient significantly less than unity may be considered as “overstabilized” and does not re- spond to favourable environmental changes. Although b, does not measure stability in a strict sense, it describes the response of an en- try to changing environments, if the response is linear Entries with b, values less than unity, sug- gesting a lack of response to changing environ. ments, are overstabilized, whereas those with b, values larger than unity are more sensitive to en- vironmental fluctuations and tend to produce more in better environments but less in poor en- vironments. Regression coefficients were used recently to measure stability in faba bean (Vicia faba L.) breding programs (Link ef al, 1994; ‘STELLING ef al., 1994) and in simulation models (Bary and Geno, 1990) because they provide in- formation characterizing genotypes and offer an additional selection criterion. Spearman's rank correlation indicated that was significantly correlated only with CV (Tablé 4), suggesting that regression and CV measure the entries’ response to environmental changes rather than stability. Most of the alfalfa cultivars and germplasm had b, values of approximately 1.0, indicating an average response among these entries in those environments (Table 3), The high and significantly positive correla- tions between pairs of stability estimates, W, Sa? , §\ and S."), suggest that these statistics are functioning in a similar manner in charac- terizing these entries (Table 4). A significant cor- relation coefficient was also reported between W, and s,* in common bean (Duarte and Zimmer. MANN, 1995). The statistics, W,, S\°", and $,), are simple to calculate, using total yield data derived from individual replications, locations, and year and are especially useful for data obtained fro. performance trials, where only the total or aver. age yields are given. An examination of relationships between pairs of stability estimates was made for data from in- dividual years to provide more information on their consistency (Table 5). Soil properties, such as pH, structure, depth, or topology, normally do not change in a short time period. However, sea- sonal fluctuations that are related closely to mois- ture availability, temperature, and occurrence of insects and diseases could cause GE interactions, thus complicating the recommendation of entries or interpretation of the data. ’ TABLE 4 Spearman's rank correlations coefficients between pairs of stability estimates for alfalfa yield trials (1987-89) Stability estimates nm i 7 we 2 ” rr 091 ost 0.90s++ oz 0.524 037s 0333 18 0524 0.905" 0305+ 0833+ oss naa ge] 0619 019 0381 0381 9273 oe 1Lo0a+* oasis Dazgee 3353 o Ossi o9a9% cp Sey 0952+ ao Highly significant at OD) level of probaly 3 The CV and b, values were significantly and — Unlike the combined data 5,3? was not signifi- 4 | _ positively correlated to each other in each of the cantly correlated with the nonparametric mea- models | 3 years and in combined data. Ecovalence (W) sures in each of the 3 individual years nor with vide in: | was significantly correlated with sq’, 0°, S"", — o? in 1988. Shukla’s (1972) o? was not signifi- offeran | and $,° in 1987 and the combinated data but cantly correlated with the nonparametric meas- not in 1988 or 1989 (Tables 4 and 5). Ecovalence ures in 1989. The reason for discrepances | that b, was not significantly correlated with sy? in 1988 between individual years, particularly 1989 and ‘(Table | or with the nonparametric measures in 1989. the combined data, is unknown, neasure anges sltivars TABLE 5 imately Spearman's rank correlations coefficients between pairs of stability measures from individual years g these for six alfalfa cultivars and two germy ‘Sealy eatiates correla. 2s, W, w b a : sh se atistics uae 0387 0.690" 0071 0.405 na 0327 0310 ona ore ostr oust 0.595 0.286 0.343 0232 0.738" 0583 0354 0.869% oasis ooo" 0.238 1.000 01048, 0.238 0.190 0.190 0.238 08 1.000" ‘920+ 0.905 0.048, 0.238 0.190 0.190 0.088 0.095 0143 ‘sz 0.903" 0.952" such ly do naa eae cv osz4 agro o2ia 0.137 0025 - 0.405 0,786" O4tt 0.333 bse 0.143 0.214 O13 0.048 ree of aaa 0.082 0.048, @ 0.308 0310, ntries si” 0.970** ST" Signuticantiy different ar the 0.05 and O01 levels of probability respectively , Computing stability estimates of a group of commercial entries, either from a performance trial or from a breeding program, could provide valuable information for growers who normally judge the yield potential of a cultivar on data of past years. A stable entry with high mean yield is considered desirable by growers. Likewise, breeders can select elite lines with stable perfor mance for advanced trials. In fact, many breed- ers unconsciously could have selected high-yield- ing materials with stability over environments without using a well-defined statistic model. We believe that the stability estimates from all pled models provide valid measures on yield per- formance. However, the ecovalence (W,) and the nonparametric statistics, either S/" or $,%, ap- pear to be more suitable to the data obtained from performance trials, which normally provide only the total or average yields. The rank-sum approach, where yield and stability in ranks are summed for each entry and then ranked, provided a joint evaluation of yield and stability. KS 196 germplasm not only yield~ ed less than other entries but appeared to be less stable (Table 6). Garst 630 had the highest mean yield and ranked third in stability, whereas Ci- marron ranked first in stability and third in yield. It seems that selecting for yield only would Sacrifice the stability to some degree, Likewise, selecting for stability only would sacrifice yield somewhat, When yield and stabil- ity are considered jointly, Cimarron and Garst 636 appeared to be the logical choices. Thus, the rank-sum method offers a simple, quick, and reasonable way to select promising cultivars, germplasm, or synthetics for release or recom mendation to growers. In some cases, different= weights may be applied to vield and stability esd timates depending on the relative importance o| the two variables. In this report, alfalfa cultivars and germ plasm were distinguished based on yields and stability parameters. Among the models evaluat-§ ed, the nonparametric methods were not only: simple to calculate, but required no assumptions ® concerning additivity of genotype and environ- mental effects, homogeneity of variances, and? normality. Nonparametric methods appear to be readily adaptable to data from performance ! trials and can be combined with yield to identi- fy entries with both high and stable yields. ACKNOWLEDGEMENTS, We sincerely thank the United Nation Develop- ment Program in China for the financial support by sending Professor G.L. Yue to our laboratory conduct ing the stability analysis on alfalfa data collected from various experiment stations in Kansas, USA. REFERENCES BACHIREDDY, VR., PRUNE, RJR., CHIN, K.L, and KANG, MS, 1992, Conventional selection versus methods that ase genotype x environment interaction in sweet corn trials, Hort 2 436-438, Barny, TA. and GEG, S., 1990. A simulation approach to analyze and interpret genetic-environmental interaction, In: MS. Kang (ed.) Genorype-by-environment interaction and plant breeding. Louisiana State Unix. Agric, Cts, Bar ton Rouge, LA, pp. 49-107, TABLE 6 Ranks for yield, the nonparametric stability estimates (8, and §2"] and rank-sum for six alfalfa cultivars ‘and two germplasm Nonparametric estimates Rank Sunt Calivare Yield ry = ee 5 6 6 n® 6 2 1 72) 4 8 7 ne® 8 5 5 BS) Cimarron 3 1 2 sm Garst 630 1 1 5 9a) Garst 636 2 3 3 5a) Garse 655 7 4 4 n@ "The numbers iv parentheses indicate the ranking of rank-sum approach, ort by nduct: from MS, ch to *, Ba. ‘naa. actin Selenite BECKER, H.C, 1981, Correlations among some statistical ‘measures of phenotypic stability. Euphytica, 30: 335-340. oARTE, J.B. and ZIMMERMANN, MLJ., 1995, Corzelation among, yield stability parameters in common bean. Crop. Sct. 35: 903.912 eoeiutaer, S.A. and Russet, WA., 1966, Stability parameters for comparing varieties. Crop. Sci, 6: 36-40. ranets, TR, and KaNnevnens, LW, 1978. Yield stability studies in short-season maize. IA descriptive method for grouping genotypes. Can J. Plant Sci, $8: 1029-1034, Hows, M. and Nassar, R., 1989. On test of significance For onparamesric measures of phenotypic stabil. Biomer- res, 45: 997-1000. JALALUDOIN, M.D. and HARRISON, S.A., 1993. Repeatability of stability estimators for grain yleld of wheat. Crop. Sei., 33: 720-725, KANG, M.S., 1988, A rank-sum method for selecting. high- yielding, stable com genotypes. Cereal Res. Commun. 16: 113-115, KANG, M.S., 1990, Understanding and utilization of genorype- by-environmeatal interaction in plant breeding. In: MLS. Kang (ed.) Genorype-by-environmental interaction and plant breeding, Louisiana State Unix, Agric. Ctr, Baton Rouge, LA, pp. $2-68. Kanc, M.S., 1993. Simultaneous selection for yield and stability in crop performance trials: consequences for growers, Agron, 4, 88: 754-757. KanG, MS. Goewan, D.P. and PHAM, H.N., 1991. Application fa stability statistic ¢o international maize yield tials. Theor. Appl Genet.. 84: 162-165. Leon, J, and BECKER, H.C,, 1988, Repeatability of some sta- tistical measures of phenotypic stability-cortelations: between single year results and multi-year results. Plant Breeding, 100: 137-142. B Lay, C.S., Buns, MIR, and LerKovtrcH, LP, 1986. Stability analysis: where do we stand? Crop Sci, 26: 894.900, Lin, CS. and BINNS, M.R., 1991. Genetic properties of four types of stability parameter. Theor. Appl. Genet,, 82: 505+ 509, LINK, D., STELUNG, D. and Esweven, E., 1994. Yield stability ‘in faba beans, Vicia faba L. L Variation among inbred lines, Plant Breeding, 112: 24-29. Macanl, R. and Kavo, M.S, 1993. Genotype selection via a new yield-stability in maize yield trials. Euphycica, 70: 105-111 Nassan, R, and Hun, M., 1987. Studies on estimation of phenotypic stability: test of significance for nonparamet- ric measures of phenotypic stability: Biometrics, 43: 45: 33. PUAN, H.N. and KaNG, M.S., 1988. Interrelationships among, and repeatability of several stability statistics estimates from international maize trials. Crop Sci,, 28: 925.928, Y SHUKLA, G.K., 1972, Some statistical aspects of partitioning genotype-environmental components of variability. He- rediry, 29: 237-245. SoKAL, RR and ROWE, FJ., 1981. Biometry - the principles ‘and practice of statistics in biological research. WH. Free+ ‘man and Co., San Francisco, CA. SreLUxe, D,, EaweveR, E. and Link, W,, 1994, Yield stabil- ity in faba bean, Vicia faba L, 2. Bifects of heterozygos- ity and heterogeneity, Plant Breeding, 112: 30-39, Waucxs, G., 1962. Uber eine Methode zur Ergassung der ‘okologischn Strebreite in Feldversuchen. 2. Pflancen- elchig, 47: 92-96. ‘Yur, C., PEANG, S.K., WALTER, TL., Wassom, C.D. and Lunc, ‘G.H., 1990. Stability analysis of yield in maize, wheat, and sorghum and its implications in breeding programs. Plant Breeding, 104: 72:80,

You might also like