Article

Comparison of rank-based stability statistics for grain yield in rainfed durum wheat

Authors:
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

Several rank-based stability statistics are in use, but little information is available on their repeatability and accuracy. The main objective was to examine the repeatability, similarity and accuracy of the estimates of mean yields and 10 rank-based stability statistics calculated from 25 rainfed durum wheat genotypes grown in 21 diversified rainfed environments. Spearman's rank-correlation coefficients and Kendall's coefficient of concordance were used to estimate the repeatability and similarity of stability statistics, while the coefficient of variation from resampling correlations obtained by bootstrap method was used to quantify their accuracy. Correlations between the estimates from each of the rank-based stability statistics obtained from single-year results were very low and non-significant, suggesting that single-year results could not serve as a basis to quantify phenotypic stability. The accuracy for the repeatability of stability statistics was found to be poor, whereas the accuracy for similarities of stability statistics was high.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... , Coefficient of concordance, Biplot analysis Significant genotype-by-environment (G×E) interaction had been reported in large number of multi environmental studies (Pour et al. 2019). Cross over interaction masks the correlation between genotypic and phenotypic values of genotypes and hinders the selection of the promising genotypes (Mohammadi et al. 2016). Large number of statistical approaches, based on univariate and multivariate models, have been observed in literature to estimate 324 Print ISSN : 1974ISSN : -1712 Online ISSN : 2230-732X stable performance of genotypes (Vaezi et al. 2018). ...
... The loadings of measures based on first two significant principal components were reflected in table 12 (Mohammadi et al. 2016). Both significant PAC's accounting for 85.9% of the variance of the original variables (Fig. 1) ...
... Multi-environment trials (MET) have been established as inevitable to recommend promising genotypes for different locations of the country (Vaezi et al., 2018). Genotype x Environment (GxE) interaction effects decrease the association between genotypic and phenotypic values and also masks the selection of the desirable genotypes (Mohammadi et al., 2016). Stability analysis methods are categorized into two parametric and non-parametric groups (Huhn & Leon, 1995;Farshadfar et al., 2014;Golkar et al., 2020). ...
... Biplot analysis of nonparametric measures had been carried out to explore any type of association among studied measures. Loadings of the first two principal components axes (PCA) of ranks of nonparametric measures were shown in figure 1 (Mohammadi et al., 2016). Two separate groups of Yield with GAI and MR with Med were displayed in the graphical analysis. ...
... Both genotypes could be regarded as good candidates for other environments of the country. Nonparametric tests for interactions provided more specific information about the nature of GE interactions (Rasoli et al., 2015;Mohammadi et al., 2016). The main purpose of barley improvement programs is to sustain higher yield with stability. ...
Article
The results of testing new Agrostab computer program “Indicators of stability of agricultural crops varieties” based on long-term (2017–2019) field research data in Novosibirsk region are presented. The program implements modern methods for assessing the ecological plasticity of varieties and allows to evaluate the stability of genotypes by a set of phenotypic characters. The research material was common spring wheat varieties from the Institute of Cytology and Genetics of the Siberian Branch of the Russian Academy of Sciences (Novosibirskaya 15, Novosibirskaya 31, Sibirskaya 12), and collection forms of spring hexaploid triticale from the VIR world collection Lt-F6544-6 (k-3992), variety Ukro (k-3644) and Sears 57 × Ukro hybrid received as a result of breeding. The samples had a different ecological and geographical origin and belonged to different ripeness groups. Plants were grown organically without the use of fertilizers and pesticides. They were sown at different times after bare fallow in four repetitions. In the course of the experiment, the following genotype stability indicators were determined: environmental variance, coefficient of homeostaticity, weighted homeostacity index, steadiness of stability index, coefficient of multiplicativity, specific adaptive ability, superiority measure, ecovalence, regression to environmental index, non-parametric stability index. The necessity of using the complex value of the integrated selection index to calculate the parameters of plasticity and genotype stability is shown. Testing of the Agrostab program for breeding of spring triticale made it possible to determine the targeted vector for the selection of varieties in the agroclimatic conditions of Western Siberia and adaptive samples to be included in hybridization. The use of new models and methods of statistical data analysis to determine the environmental plasticity of varieties and hybrids will increase the efficiency of the breeding process.
Article
Full-text available
Durum wheat (Triticum durum desf.) is one of the most important cereal crops used to make pasta, macaroni and other foods. Understanding the complexity of the genotype × environment interaction (GEI) and its impact on determining the most stable genotypes, mega environments and other adaptation goals has been a special concern for the plant breeders. Various methods exist for the stability analysis of grain yields based on different sustainability concepts that can be grouped into parametric, nonparametric and multivariate methods. In this research, 20 genotypes of durum wheat (19 advanced genotypes and Dehdasht commercial variety) were evaluated in five regions during four years (2009- 2013). Experiments were carried out in a randomized complete block design with four replications in each environment (year × location). Based on the results of grain yield stability assessment using 13 nonparametric methods in the methods of mean rank, standard deviation and adaptability percentage, genotypes 9, 10 and 12, in the methods of Nassar and Huhn and sum rank of Kang, genotypes 4, 5, 10, 11 and 17 and in the Thennarasu and Fox methods, genotypes 6, 9, 10, 12, 13 and 17 were selected as the most stable genotypes. Finally, based on summarizing the results of all methods genotypes 10 and 12 were selected as the most stable genotypes. According to the previous studies of the authors, it is suggested that in the next researches, in order to better evaluate the stability of genotypes, one of the sustainability methods such as AMMI, spatial regression or mixed models, along with nonparametric methods, must be considered.
Article
Full-text available
The objective of this work was to estimate the repeatability of adaptability and stability parameters of common bean between years, within each biennium from 2003 to 2012, in Minas Gerais state, Brazil. Grain yield data from trials of value for cultivation and use common bean were analyzed. Grain yield, ecovalence, regression coefficient, and coefficient of determination were estimated considering location and sowing season per year, within each biennium. Subsequently, a analysis of variance these estimates was carried out, and repeatability was estimated in the biennia. Repeatability estimate for grain yield in most of the biennia was relatively high, but for ecovalence and regression coefficient it was null or of small magnitude, which indicates that confidence on identification of common bean lines for recommendation is greater when using means of yield, instead of stability parameters.
Article
Full-text available
Repeatability of mean downy mildew (Sclerospora graminicola (Sacc.) Schroet.) incidence, regression coefficients and deviation mean squares were investigated for 25 pearl millet (Pennisetum typhoides (Burm.) Stapf. & Hubb.) genotypes in 20 environments by correlating arrays of these stability parameters over subsets of the 20 environments arranged according to the year-wise, random, stratified and extreme methods of environmental division. Correlation coefficients between arrays of mean downy mildew incidence from different pairs of subsets ranged from 0.57 to 0.98 and those of deviation mean squares from 0.58 to 0.96 indicating good repeatability of these parameters. Arrays of regression coefficients from different subsets, on the other hand, showed correlation coefficients that ranged from -0.58 to 0.96. Apparently, the regression index of stability was not repeatable for the genotypes and environments studied. Therefore, in order to identify a widely adapted genotype, testing is required to be carried out over a wider range of environments.
Article
Full-text available
(...) Our objective was to compare these methods and to study their relationship to yield and the stability-variance statistic (σ 2 i ). We compared Kang's (1988b) rank-sum method (Index 1; equal weights for yield and σ 2 i ) and four additional rank-sum indices [two (Index 2), three (Index 3), four (Index 4), and five (Index 5) times more weight for yield than for stability variance] with Huhn's (1979) S 3 i and S 6 i statistics, and Lin and Binns' (1988) P i . All statistics were calculated for each of five sets of data from international maize (Zea mays L.) yield trials (...)
Article
Full-text available
Recent review articles in this journal have com- pared the relative merits of two prominent statis- tical models for analyzing yield-trial data: Additive main effects and multiplicative interaction (AMMI) and genotype main effects and genotype × envi- ronment interaction (GGE). This review addresses more than 20 issues that require clarifi cation after controversial statements and contrast- ing conclusions have appeared in those recent reviews. The AMMI2 mega-environment display incorporates more of the genotype main effect and captures more of the genotype × environ- ment (GE) interaction than does GGE2, thereby displaying the which-won-where pattern more accurately for complex datasets. When the GE interaction is captured well by one principal com- ponent, the AMMI1 display of genotype nominal yields describes winning genotypes and adap- tive responses more simply and clearly than the GGE2 biplot. For genotype evaluation within a single mega-environment, a simple scatterplot of mean and stability is more straightforward than the mean vs. stability view of a GGE2 biplot. Diagnosing the most predictively accurate mem- ber of a model family is vital for either AMMI or GGE, both for gaining accuracy and delineating mega-environments.
Article
Full-text available
Phenotypic yield stability is a trait of special interest for plant breeders. Many statistical procedures are available for stability analysis, each of them alowing for different interpretations. The objective of the present study was to determine the degree of correlation among the 13 satatistical parameters that can be used for the analysis of phenotypic stability. Such correlations could be used to assess the extent to which these parameters identify unique genetic effects.
Article
Full-text available
Multienvironment trials (MET) generate two types of two-way data: genotype X environment data for a target trait and genotype X trait data in individual or across environments. These data can be visually analyzed by a GGE biplot and a genotype X trait biplot, respectively. This paper describes a third type of biplot, the covariate-effect biplot, and illustrates its tandem use with the other biplots to achieve a fuller understanding of MET data. The covariate-effect biplot is generated on the basis of an explanatory trait X environment two-way table consisting of correlation coefficients between the target trait (e.g., yield) and each of the other traits in each of the environments. This biplot displays the yield-trait relations in individual environments and addresses whether and how the genotype X environment interactions (GE) for yield can be explored by indirect selection for the other traits. These other traits are treated as genetic covariables and can be replaced by other genetic covariables such as genetic markers, QTL, or genes. The biplot methodology was demonstrated by MET data of barley (Hordeum vulgare L.) conducted across North America. Both the GGE biplot and the covariate-effect biplot showed that the environments fell into two (eastern vs. western) megaenvironments. The covariate-effect pattern explained 81% of the GGE pattern, suggesting that the GE pattern for yield can be effectively explored by indirect selection for these traits. Specifically, barley yield can be improved by selecting for larger kernel weight, earlier heading, and better lodging resistance in the eastern megaenvironment. In contrast, the yield-trait relationship in the western megaenvironment was highly variable, and yield improvement can be achieved only by selecting for yield per se across environments. We suggest that the GGE biplot, the genotype X trait biplot, and the covariate-effect biplot be used jointly to better understand and more fully explore MET data.
Article
Full-text available
Yield trials frequently have both significant main effects and a significant genotype ✕ environment (GE) interaction. Traditional statistical analyses are not always effective with this data structure: the usual analysis of variance (ANOVA), having a merely additive model, identifies the GE interaction as a source but does not analyze it; principal components analysis (PCA), on the other hand is a multiplicative model and hence contains no sources for additive genotype or environment main effects; and linear regression (LR) analysis is able to effectively analyze interaction terms only where the pattern fits a specific regression model. The consequence of fitting inappropriate statistical models to yield trial data is that the interaction may be declared nonsignificant, although a more appropriate analysis would find agronomically important and statistically significant patterns in the interaction. Therefore, agronomists and plant breeders may fail to perceive important interaction effects. This paper compares the above three traditional models with the additive main effects and multiplicative interaction (AMMI) Model, in an analysis of a soybean [Glycine max (L.) Merr.] yield trial. ANOVA fails to detect a significant interaction component, PCA fails to identify and separate the significant genotype and environment main effects, and LR accounts for only a small portion of the interaction sum of squares. On the other hand, AMMI analysis reveals a highly significant interaction component that has clear agronomic meaning. Since ANOVA, PCA, and LR are sub-cases of the more complete AMMI model, AMMI offers a more appropriate first statistical analysis of yield trials that may have a genotype ✕ environment interaction. AMMI analysis can then be used to diagnose whether or not a specific sub-case provides a more appropriate analysis. AMMI has no specific experimental design requirements, except for a two-way data structure. Contribution from the USDA-ARS, U.S. Plant Soil and Nutrition Lab., Tower Rd., Ithaca, NY 14853. Please view the pdf by using the Full Text (PDF) link under 'View' to the left. Copyright © . .
Article
Full-text available
Southeast Europe is Europe's major production area of soybean of maturity groups (MG) 0, I and II, but genotype by environment interaction (GEI) for soybean traits in this region is still not characterized. The objective of this study was to characterize GEI in soybean grown in Croatia for seed yield, protein and oil content. Fourteen soybean cultivars of MG 0 and I were evaluated at 15 environments. In the combined anova, genotypes, environments and GEI were significant. All three seed traits were more affected by year than by location and seed yield and protein content were more sensitive to environmental changes than oil content. Genotype by environment interaction is generally of less importance than effects of genotypes and years, especially for oil content. High-yielding genotypes had values of regression coefficient (b) close to 1, indicating that they were not responsive to extreme environments, either positive or negative. Means and b values were not correlated for yield and negatively correlated for protein and oil content, probably because of low-input practice in the region. The absence of recognizable differences in means, b values and principal component scores of ammi analysis between two MGs for all traits indicated that soybean cultivars of MG 0 may be recommended for growing in the region because they responded similarly to environmental changes as do full-season genotypes of MG I.
Article
Full-text available
There is an increasing number of stability parameters for genotypes grown in different environments. It is therefore useful to study the statistical relations between these parameters. One approach is the calculation of rank correlations between different stability parameters in empirical data sets. In the data analysed there are high rank correlations between ecovalence Wi, deviation mean square s2 di, the nonparametric measures Si (1), Si (2), and two new measures Ri and Li as well as between environmental variance S2 xi and regression coefficient bi. The results suggest that Si (1), Si (2), Li, and Ri can be used as alternatives to Wi and the stability variance 2 i. This may be worthwhile, if certain statistical assumptions do not hold, particularly if significance testing is needed.
Article
Full-text available
In plant breeding, correlations between the statistics of stability and adaptability of popcorn cultivars are not yet well understood. Therefore, the objectives of the present experiment was to investigate the correlations between σdi2 and βi from Eberhart and Russell, ωi from Wricke, Si(1), Si(2) and Si(3) from Huehn, Pi from Lin and Binns and the rank-sum from Kang, and indicate the most reliable method for selecting popcorn cultivars. These statistics were estimated by data of crop yield from 19 Brazilian genotypes under 21 environments and popping expansion under 16 environments. The ωi, Si(1), Si(2), Si(3) and σdi2 were positively and significantly correlated indicating that just one in these five statistics is sufficient for selecting stable genotypes although they were not correlated with the means of crop yield and popping expansion. The βi was negatively and significantly correlated with Pi for crop yield indicating that the most adaptable genotypes tend to have the lowest estimates of Pi. Although Pi was not correlated with ωi, Si(1), Si(2), Si(3), or ωdi2 statistics, it displayed positive correlation with the Index 1 (crop yield and popping expansion + Si(1) rank) and Index 2 (crop yield and popping expansion + Wi) indicating that superior popcorn genotypes are also stable. Finally, both Pi and the rank-sum are useful statistics in breeding programmes where crop yield, popping expansion and stability are essential traits for selecting genotypes.
Article
Full-text available
Joint regression and Additive Main effects and Multiplicative Interaction (AMMI) models were compared for i) capacity of describing genotype-location (GL) and genotype-environment (GE) interaction effects (environments = location-season combinations), assessed in terms of estimated variance of heterogeneity of genotype regressions and of the sum of the variances of significant interaction principal component (PC) axes, and ii) repeatability between cropping seasons of measures of genotype stability across locations. These measures were Finlay and Wilkinson's regression coefficient for joint regression, and the Euclidean distance from the origin of significant interaction PC axes (D) and the absolute value of PC 1 score (| PC 1 |) for AMMI. Shukla's stability variance (σsup2;) was considered in addition. The study included three data sets for durum wheat, two for maize and one each for bread wheat and oat. Relationships between climatic variables and GL interaction occurrence were also assessed. AMMI proved distinctly more valuable in six data sets for description of GE effects and in four for description of GL effects over seasons. Its superiority was not crop-specific and tended to occur when more, distinct environmental constraints affected genotype responses. When both methods were appropriate, they provided a similar ordination of sites and genotypes for GL effects. The models that adequately described GL interaction over seasons generally provided also stability measures that were moderately repeatable between seasons. D was better repeatable than | PC 1 | and σ& 2; in a few cases. Ordination of locations on GL interaction PC 1 tended to be consistent both between wheat and between maize data sets having either no seasons or no genotypes in common.
Article
Full-text available
Multi-environment trial data are required, to obtain variety stability performance parameters as selection tools for effective cultivar evaluation. The interrelationship among seven stability parameters and their association with mean yield, along with the repeatability of these parameters across consecutive years was the objective of this study. Cottonseed yield data of 31 cotton cultivars, proprietary of Delta and Pine Land Co and other companies, evaluated in 20 locations over the 1999–2005year period in Greece, Spain and Turkey were used for combined analysis of variance in four datasets. Across locations in a single evaluation year (dataset A), across locations in each of two single consecutive evaluation year (dataset B), across locations and two consecutive years (dataset C) and across locations and three consecutive years (dataset D). For each dataset, cultivar phenotypic variance (sp2)(\sigma_{\rm p}^{2}) was appropriately partitioned in its components and the h2 and sge2\sigma_{\rm ge}^{2} component estimated. Furthermore, following the appropriate stability analysis bi,sdi2,\hbox{b}_{\rm i},\,\hbox{sd}_{\rm i}^{2}, si2,YSi\sigma_{\rm i}^{2},\,\hbox{YS}_{\rm i} and AMMI1 along with the GGE Biplot distance (GGED) and instability (GGEIN) parameters were obtained. The interrelationship among the parameters and their association with mean yield based on Spearman rank correlation was studied in each of the seven single evaluation years (dataset A). Rank correlation coefficients were also used as estimates of the repeatability of these stability parameters across consecutive year combinations (dataset B, C and D). The parameters GGED and YSi were consistently highly correlated with each other and mean yield in five out of seven single evaluation years. The data provided evidence that single year evaluation across locations might be sufficient to reliably rank cotton cultivars, based on mean yield along with GGED and YSi. Combined analysis across two consecutive years (dataset C) was more effective as compared to single year evaluation. GGED was relatively more repeatable than YSi and mean yield in single (dataset B) and 2-year comparisons (dataset C). Although GGED is an index depended and proportional to yield, provides a superior way to integrate mean performance and stability into a single measure, which can be assessed visually on biplots. Regarding the other stability parameters, the results were contradicting and of low repeatability across single years and two consecutive years. Cultivar evaluation combined across locations in 3years did not improve the repeatability of cultivar variance effects but resulted in very high repeatability of GGED, YSi and mean yield.
Article
Full-text available
Classically, selection for superior genotypes in cacao has been based on the successive harvest records across a number of years. Little information on the minimum duration of these harvest periods is available in the literature. The repeatability coefficient (ρ) was used to estimate this period. Twenty five cacao genotypes were assayed in a randomized block design with four replications and 16-plant plots. The following yield components were studied: number of healthy fruits per plant, number of collected fruits per plant, weight of humid seeds per plant and per fruit, and percentage of diseased fruits per plant, over 5 years (1986–90). Repeatability estimates were higher than 0.84 for all components, except percentage of diseased fruits per plant (^ρ - 0.41). With such estimates, it is possible to select genotypes on the basis of only two years of successive harvests, with a determination coefficient of 90%. The advantages of applying the repeatability coefficient to the cacao breeding program are discussed.
Chapter
This chapter has two objectives: first, to discuss implications of genotype by environment interaction (G×E) for breeding and second, to present major statistical models for assessing genotypic adaptation. We concentrate on widely used methodologies and on that which we believe will be extensively applied, but acknowledge the subjective element in such decisions.
Chapter
Die Entwicklung verteilungsfreier Rangtests zur Überwindung der Normalverteilungsannahme in den so häufig gebrauchten varianzanalytischen Modellen hat sich in zweierlei Weise vollzogen. Einmal werden sie als bedingte Tests, zum anderen als asymptotische Verfahren vorgestellt. Dabei zeigt es sich, daß die Methodik des bedingten Rangtests zwar in einer Vielzahl von speziellen Versuchsplänen anwendbar ist, wie das Buch von Puri und Sen (1971) angibt, daß aber diese Verfahrensweise nicht auf das Testen linearer Hypothesen in allgemeinen linearen Modellen zu erweitern ist. Asymptotische Untersuchungen haben hier hingegen in letzter Zeit zu befriedigenden Ergebnissen geführt. Dabei wurde von folgender koordinatengebundener Formulierung ausgegangen: In einem linearen Modell $$y = x_1b_1 + x_2b_2 + e$$ (1.1) betrachte man das Testproblem $$H:b_2 = 0 \; \text{gegen} \; K:b_2 \neq 0$$ (1.2)
Article
The Additive Main effects and Multiplicative Interaction (AMMI) model has been used extensively for analysis of multi-environment yield trials for two main purposes: understanding complex genotype x environment interactions and increasing accuracy. Nevertheless, AMMI analyses implementing best practices have been rare. Accordingly, this paper presents a simple protocol with four steps: (i) analysis of variance, (ii) model diagnosis, (iii) mega-environment delineation, and (iv) agricultural recommendations. This protocol is illustrated with an international bread wheat trial. This paper concerns a basic and common application of AMMI: yield-trial analysis without consideration of special structure or additional data for either genotypes or environments. Best practices involve using both treatment and experimental designs to gain accuracy and exploiting both broad and narrow adaptations to increase yields.
Article
Because of unpredictable conditions in Mediterranean environments, successful crop production requires improved adaptation and yield stability to mitigate major abiotic stresses such as drought and cold. In this study 380 durum wheat (Triticum turgidum L.) landraces with worldwide origins and four checks were evaluated in four rainfed research stations for 3 yr (2008–2011). The main objective was to investigate yield stability and adaptation patterns of the landraces to highland rainfed cold regions of Iran, where drought and cold are limiting factors. The experimental design was an unreplicated trial at each environment. Best linear unbiased predictions (BLUPs) data were used to analyze landrace × environment interactions using multivariate statistical methods. Considerable variability in yield, adaptation, and stability was observed for the landraces, which could be exploited for crop improvement. The landrace and environment groups allowed characterization of origins on the basis of similar performance of landrace within particular environments. Grouping of environments was not repeatable among the years. Many of the durum landraces exhibited a high combination of yield and stability for both drought and cold stresses, comparable to cold- and/or drought-tolerant checks. Most landraces originating from Europe and the United States could be considered as a genetic resource for specific adaptation, while the landraces from Asia may enhance genetic potential of yield stability. In conclusion, current durum wheat selection may lead to yield stability and specific adaptation, which provides opportunities for this collection to be useful for genetic improvement of both drought and cold tolerance in durum wheat.
Article
Parametric methods for estimating genotype-environment interactions and phenotypic stability (stability of genotypes over environments) are widely used in plant and animal breeding and production. Several nonparametric methods proposed by Huhn (1979, EDP in Medicine and Biology 10, 112-117) are based on the ranks of genotypes in each environment and use the idea of homeostasis as a measure of stability. In this study the statistical properties and tests of significance for two of these nonparametric measures of phenotypic stability (mean of the absolute rank differences of a genotype over the environments and variance among the ranks over the environments) are investigated. The purpose of this study is to develop approximate but easily applied statistical tests based on the normal distribution. Finally, application of the theoretical results is demonstrated using data on 20 genotypes (varieties) of winter wheat in 10 environments (locations) from the official German registration trials.
Article
The objective of this study- was to evaluate whether different statistical measures of phenotypic stability vary in their repeatability. Eight multi year and multi location variety tests of wheat, barley and oat were analysed separately for each year. Single year data of yield, of response parameters: environmental variance (S23 and coefficient of regression (b), and of stability parameters: deviation mean squares (S23), coefficient of determination (r2), ecovalence (W), and the nor, parametric measure variance of ranks (Si4), were correlated with multi year, multi location results.Repeatability of single year results was highest for yield, where rank correlation coefficients amounted to about 0.80. s2x and b showed medium values of nearly 0.55 The stability parameters s2d, r2, W and Si4 did not differ in their repeatability. Respective correlation coefficients possessed values of approximately 0.40 and were very variable from experiment to experiment. Reliability of single year results was especially low for experiments with high varieties × years interactions. Single year results of the examined variety tests could not serve as a basis to quantify phenotypic stability even if more than ten locations were involved.
Article
Assessment of cultivar performance in a cultivar x location x year experiment is often difficult because of the presence of a location x year interaction. Our objective is to demonstrate a method on separation of environment effects (location x year) into predictable and unpredictabel components. The analysis consists of two parts: (1) a regression analysis based on location effects (averaged over years), assuming that the location means represent predictable environmental variation; and (2) the estimation of stability (denoted type 4) based on the years within location mean squares, assuming that years within location represent unpredictable environmental variation. From the regression analysis in (1), a breeder can determine the optimum range of locations in which a cultivar is well suited, and from (2) he can choose the most stable cultivars. The advantage of type 4 stability is that it is independent of the other cultivars included in the test and of the regression coefficient estimated for predictable variation. Three sets of published data are used to illustrate the analysis. Type 4 stability is compared with type 3 stability (deviation mean square from regression on environmental index) for genetic consistency. The analyses suggest that type 4 stability is consistent and is therefore a potential genetic parameter, but type 3 stability is not.
Article
Key message: We present experimental data for wheat, barley, and triticale suggesting that hybrids manifest on average higher yield stability than inbred lines. Yield stability is assumed to be higher for hybrids than for inbred lines, but experimental data proving this hypothesis is scarce for autogamous cereals. We used multi-location grain yield trials and compared the yield stability of hybrids versus lines for wheat (Triticum aestivum L.), barley (Hordeum vulgare L.), and triticale (×Triticosecale Wittmack). Our study comprised three phenotypic data sets of 1,749 wheat, 96 barley, and 130 triticale genotypes, which were evaluated for grain yield in up to five contrasting locations. Yield stability of the group of hybrids was compared with that of the group of inbred lines estimating the stability variance. For all three crops we observed a significantly (P < 0.05) higher yield stability of hybrids compared to lines. The enhanced yield stability of hybrids as compared to lines represents a major step forward, facilitating coping with the increasing abiotic stress expected from the predicted climate change.
Article
The Additive Main effects and Multiplicative Interaction (AMMI) model, Genotype main effects and Genotype X Environment interaction (GGE) model, and Principal Components Analysis (PCA) are singular value decomposition (SVD) based statistical analyses often applied to yield-trial data. This paper presents a systematic comparison, using both statistical theory and empirical investigations, while considering both current practices and best practices. Agricultural researchers using these analyses face two inevitable choices. First is the choice of a model for visualizing data. AMMI is decidedly superior, not for statistical reasons, but rather for agricultural reasons. AMMI partitions the overall variation into genotype main effects, environment main effects, and genotype X environment interactions. These three sources of variation present agricultural researchers with different challenges and opportunities, so it is best to handle them separately, while still considering all three in an integrated manner. Second is the choice of a member of a given model family for gaining predictive accuracy. AMMI, GGE, and other SVD-based model families are essentially equivalent, but best practices require model diagnosis for each individual dataset to determine which member is most predictively accurate. Making these two choices well allows researchers to extract more usable information from their data, thereby increasing efficiency and accelerating progress.
Article
Producing specific varieties for each subregion of a target region, instead of widely adapted varieties, may exploit positive genotype X location (GL) interactions to increase crop yields. With reference to the Algerian durum wheat (Triticum durum Desf.) region, as defined in a Geographic Information System (GIS), this study aimed at (i) comparing additive main effects and multiplicative interaction (AMMI) vs.joint regression modeling of GL effects, (ii) verifying the reliability of a GIS-based definition of two subregions that extended the site classification on the basis of GL effects as a function of long-term winter mean temperature in the GIS, and (iii) comparing wide vs. specific adaptation in terms of observed and predicted yield gains (the latter for scenarios of defined and undefined selection locations). Twenty-four cultivars from international centers, Europe, and North Africa were evaluated across 3 yr in a total of 47 environments by randomized complete block designs with four replications per trial. The AMMI modeling was distinctly superior to joint regression. The AMNII + cluster analysis and pattern analysis classified test locations consistently and in good agreement with the GIS-based subregion definition. Under the hypothesis of six selection environments assigned to subregions in proportion to their size (three sites in each of two years) for late stage selection, specific adaptation provided 2 to 7% greater gains than wide adaptation over the region at similar costs. The advantage of specific adaptation was much larger (39% determined on the basis of observed gains) for the smaller, stressful inland subregion, where specific adaptation may also enhance food security. Performing earlier specific selection, or adopting distinct genetic bases for each subregion, may further increase the advantage of specific breeding.
Article
Stability statistics that are used to explain genotypic response to environments are not useful to plant breeders unless they are repeatable across sets of environments. The purpose of this study was to examine the repeatability of the stability estimators: regression coefficient, regression coefficient away from mean regression, mean squares for deviation from regression, Shukla's stability variance, variance of genotypic means, and genotypic coefficient of variation, in addition to coefficient of determination and mean yield (.)
Article
The presence of genotype × environment (GE) interactions in plant breeding experiments has led to the development of several stability parameters in the past few decades. The present study investigated the repeatability of these parameters for 16 chickpea (Cicer arietinum L.) genotypes by correlating their estimates obtained from extreme subsets of environments within a year and also over years. Based on the estimates of response and stability parameters within each trial, the ranking of genotypes in the low-yielding subset differed from that in the high-yielding subset. This indicates poor repeatability for response and stability parameters over the extreme environmental subsets. The estimates of mean yield and stability parameters represented by ecovalence, W2i, were consistent over years, whereas those of response parameters (bi, and S2i) showed poor repeatability. Our results suggest that single-year results for yield and stability can be used effectively for selecting cultivars with stable grain yield if tested in a wider range of environments.
Article
A suggestion is given of how to prove main and interaction effects in two-way layouts independent of each other even if the data are just ordinally scaled. Starting from HILDEBRAND'S (1980a, b) non-parametric approach which presupposes interval-scaled data, transformations of ranks are settled before analysis per analogy to the H-test takes place. That is, the same formula of an asymptotically X2-distributed test-statistic results but mean ranks are used instead of mean scores in order to partialize, for instance, main effects while testing interaction effect. Finally an allusion is given of how to handle ties as well as unequal sample sizes.
Article
Cultivar evaluation and mega-environment identification are among the most important objectives of multi-environment trials (MET). Although the measured yield is a combined result of effects of genotype (G), environment (E), and genotype x environment interaction (GE), only G and GE are relevant to cultivar evaluation and mega-environment identification. This paper presents a GGE (i.e., G + GE) biplot, which is constructed by the first two symmetrically scaled principal components (PC1 and PC2) derived from singular value decomposition of environment-centered MET data. The GGE biplot graphically displays G plus GE of a MET in a way that facilitates visual cultivar evaluation and mega-environment identification. When applied to yield data of the 1989 through 1998 Ontario winter wheat (Triticum aestivum L.) performance trials, the GGE biplots clearly identified yearly winning genotypes and their winning niches. Collective analysis of the yearly biplots suggests two winter wheat mega-environments in Ontario: a minor mega-environment (eastern Ontario) and a major one (southern and western Ontario), the latter being traditionally divided into three subareas. There were frequent crossover GE interactions within the major mega-environment but the location groupings were variable across years. It therefore could not be further divided into meaningful subareas. It was revealed that in most years PC1 represents a proportional cultivar response across locations, which leads to noncrossover GE interactions, while PC2 represents a disproportional cultivar response across locations, which is responsible for any crossover GE interactions. Consequently, genotypes with large PC1 scores tend to give higher average yield, and locations with large PC1 scores and near-zero PC2 scores facilitates identification of such genotypes.
Article
There are many statistics to assess cultivar yield stability: some deal with genotype x environment interaction (GEI) patterns (b1, and interaction principal component scores), some with GEI noise (s(di)/2), some with total GEI sum of squares (δ(i)/2), and others with rank changes [S(1)]. Our objective was to assess the repeatability in soybean [Glycine max (L.) Merr.] of different yield stability statistics. We evaluated nine statistics derived from additive main effects and multiplicative interaction (AMMI) analyses, regression coefficients from analyses using either yield-based or plant height-based environmental indices, as well as s(di)/2, δ(i)/2, S(1). Yield data were from Arkansas trials of maturity group V cultivars from 1987 to 1994. Stability analyses were warranted because GEl was significant, partitioning of GEI sum of squares by AMMI or regression techniques was generally significant, and all stability statistics found significant differences among the cultivars. The AMMI-derived stability statistics produced information on cultivar stability that was similar to that produced by s(di)/2 and δ(i)/2. The repeatability of cultivar rankings for s(di)/2, δ(i)/2, S(1), and most AMMI statistics, estimated in different single- or two-year periods were low, indicating that these statistics would not be useful to breeders or growers selecting for stability. The repeatability of two regression coefficient statistics and one AMMI-derived statistic were moderate, particularly when estimated over environments from 2 yr. Each of the three statistics relates to a different concept of stability and may be useful to growers and breeders attempting to select cultivars with predictable yield across environments.
Article
Cultivar performance trials are usually analyzed by analysis of variance techniques. In the field of practical applications, the required simplifying assumptions may often not be fulfilled. Nonparametric metbods based on ranks avoid this difficulty and provide a viable alternative to parametric procedures. The objective of this study was to compare four nonparametric procedures (Bredenkamp, Hildebrand, Kubinger, and van der Laan-de Kroon) and the parametric analysis of variance with each other for extensive data sets from German official registration trials (1985–1989) with winter oilseed rape (Brassica napus L.), fababean (Vicia faba L.), oat (Avena sativa L.), fodder beet (Beta vulgaris L.), and sugar beet (B. vulgaris). The van der Laan-de Kroon method is based on the crossover concept of interaction (different rank orders), while the other methods are based on the concept of deviations from additivity of main effects. The methods were compared using the numerical values of the test statistics. For the sources cultivars, locations, and interactions one obtains these relationships: ANOVA ≥ Hildebrand ≅ Kubinger > van der Laan-de Kroon > Bredenkamp. If the assumptions for parametric methods cannot be accepted, the van der Laan-de Kroon method is recommended for the crossover interaction concept, while the Hildebrand or Kubinger method (both of which are approximately equivalent) should be applied for the usual interaction concept. The Bredenkamp method cannot be recommended. Please view the pdf by using the Full Text (PDF) link under 'View' to the left. Copyright © . .
Article
To study the importance of the effects of genotype - Environment interactions on the yield of pigeonpea (Cajanus cajan L. Millsp.), 10 early-maturing genotypes were grown in a randomized complete block design with three replications in a total of seven environments spread over five regions of Kenya between 1987 and 1988. Results indicated the presence of a substantial genotype - Environment interaction effect on grain yield. The observed significant genotype - Environment interaction effect is discussed in relation to its importance in pigeonpea grain yield evaluation studies. It is noted that the best genotype in one environment is not always so in other environments. Results from regression analysis indicated that this method of analysis is appropriate for describing the response of pigeonpea genotypes grown in a number of locations. Analysis of variance showed significant additive and multiplicative genotype - Environment interaction effects. Only the first interaction principal component axis (IPCA) was found to be important in describing the multiplicative interaction effects. The additive main effects and multiplicative effects (AMMI) model allowed the partitioning of interaction variance into agronomically important sources (genotype groups), and the specific genotype × environment patterns that are the basis of these sources of variance were examined.
Article
High and stable yield is a very desirable attribute of soybean (Glycine max (L.) Merr.) cultivars. Stable yield of a cultivar means that its rank relative to other cultivars remains unchanged in a given set of environments. To characterize 12 soybean cultivars chosen from performance trials, data were obtained from 10 environments (five locations in 2 years). Six stability parameters from four statistical models were derived for each cultivar. Regression coefficients were significantly and positively correlated only with coefficients of variation; they are useful in characterizing whether cultivars responded well in favourable or poor environments. Nassar and Huhn's nonparametric measures, Si(1) and Si(2), were significantly and positively correlated with Eberhart and Russell's sdi2 and Wricke's ecovalence (Wi). The stability measures are useful in characterizing cultivars by showing their relative performance in various environments. Results revealed that high-yielding cultivars also can be stable cultivars. Correlations between stability parameters obtained from individual years over the same set of locations and cultivars were very low and nonsignificant, suggesting that single-year data are not reliable as basis for selection. To provide an additional guide for selection, Kang's rank-sum approach was applied, in which both yield (in rank) and measured nonparametric stability (in rank) were considered. In general, selection for yield only would sacrifice stability to some degree, and selection for stability only would sacrifice a certain amount of yield. The rank-sum approach reconciles the two and appeared to provide a useful means to characterize soybean cultivars.
Article
The stability and genotypic mean of four traits, grain yield, grain protein content, alveograph W and bread volume, were evaluated in three multi-location trials, each covering two years. The stability of each genotype was evaluated by environmental variance (s2 E), interaction variance (s2 W) and variance of the ranks of the phenotypic values corrected for the genotypic effect (s2 R). The bootstrap method was used to study correlations between the genotypic mean and the three stability statistics and to calculate their accuracy. The repeatability of the stability statistics was measured by correlations between the values obtained in each of the two years. In addition, theoretical smaller trials were generated by random sampling and the stability values calculated were correlated with those of the original trial. Environmental variance appears to be usable both for yield and for quality traits, but there is a risk of counter-selecting a high genotypic mean of W. Whatever the trait and statistic envisaged, stability is poorly repeatable and its evaluation requires several years and a large number of locations per year to minimise sampling and environmental effects.
Article
An experiment was undertaken to determine the stability of seed yield in 21 common bean genotypes representing three growth habits. Seven genotypes in each growth habit (determinate bush, indeterminate bush and indeterminate prostrate) were evaluated in replicated trials at three locations for three years under rain fed conditions in Ethiopia. A combined analysis of variance, stability statistics and rank correlations among stability statistics and yield-stability statistic were determined. The genotypes differed significantly for seed yield and genotype × environment (year by location) interaction (GE). The different stability statistics namely Type1, Type 2 and Type 3 measured the different aspects of stability. This was substantiated by rank correlation coefficient. There were strong rank correlations among Si 2d, Wi 2,σi 2 and Si 2, where as there was weak correlation between biand Ri 2, Si 2d, Wi 2, σi 2 and Si 2. R2 was significantly and negatively correlated with Wi 2, σi 2 and Si 2. σi 2 is significantly correlated with Wi 2.Yield is significantly correlated with bi and Ri 2.None of the statistics per se was useful for selecting high yielding and stable genotypes except the YS(yield-stability statistic). Most of the high yielding genotypes were relatively stable. Of the 21 genotypes, only 11genotypes were selected for their high yielding and stable performance. Genotypes with growth habit III and I (in determinate prostrate and determinate bush) were generally more stable than in determinate bush.
Article
For an estimation of phenotypic stability of genotypes grown in different environments three stability parameters have been proposed which are based upon the ranks of the genotypes in each environment: In a two-way table with K rows (genotypes) and N columns (environments) the original data xij (=phenotypic value of the i th genotype in the j th environment (i=1,2,...,K;j=1,2,...,N)) are transformed into ranks for each of the N environments separately. We denote: rij=rank of genotype i in environment j. Then, a genotype i may be considered to be stable over environments if its ranks are similar over environments (maximum stability = equal ranks over environments). Each statistic for the similarity of the ranks in each row = genotype may be used as a stability parameter. Three different measures are proposed and discussed.One of these nonparametric measures is defined as a ratio between variability of the rij's and mean of the rij's and, therefore, it represents a confounding and simultaneous consideration of stability and yield.Differences among genotypes have an effect on the stability measures and may lead to differences in stability among genotypes when in fact there is no genotype-environment interaction. To avoid this ambiguity one may correct the xij values for the genotypic effects and the nonparametric measures may be computed using the ranks based on the corrected values xij *=xij–(\-xi.–\-x..)where \-xi.=marginal mean of genotype i and \-x\2=overall mean.Finally, approximate tests of significance based on the normal distribution are discussed for the two nonparametric measures mean absolute rank difference and variance of the ranks for 1) testing the stability of a certain genotype and 2) comparing the stabilities of different genotypes.
Article
Twenty parametric and non-parametric measures derived from grain yield of 15 advanced durum genotypes evaluated across 12 variable environments during the 2004–2006 growing seasons were used to assess performance stability and adaptability of the genotypes and to study interrelationship among these measures. The combined ANOVA and the non-parametric tests of Genotype×environment interaction indicated the presence of significant crossover and non-crossover interactions, and of significant differences among genotypes. Principal component analysis based on the rank correlation matrix indicated that most non-parametric measures were significantly inter-correlated with parametric measures and therefore can be used as alternatives. The results also revealed that stability measures can be classified into three groups based on static and dynamic concepts of stability. The group related to the dynamic concept and strongly correlated with mean grain yield of stability included the parameters of TOP (proportion of environments in which a genotype ranked in the top third), superiority index (P i) and geometric adaptability index. The second group reflecting the concept of static stability included, Wricke’s ecovalence, the variance in regression deviation (S 2 di), AMMI stability value, the Huehn’s parameters [S i(1), S i(2)], Tennarasua’s parameter [NPi(1)], Kang’s parameter (RS) and yield reliability index (I i) which were not correlated with mean grain yield. The third group influenced simultaneously by grain yield and stability included the measures S i(3), S i(6), NPi(2), NPi(3), environmental variance (S 2 xi), coefficient of variability and coefficient of regression (b i). Based on the concept of dynamic stability, genotypes G6, G4, and G3 were found to be the most adapted to favorable environments, whereas genotypes G8, G9, and G12 were more stable and are related to the concept of static stability.
Article
Average commercial maize yield in the US has increased from about 1 Mg/ha in the 1930s to about 7 Mg/ha in the 1990s. Although the increase has been the result of both genetic and agronomic-management improvements, we contend that most of this improvement is the result of the genotype×management interaction. The genetic improvement in maize yield is associated neither with yield potential per se, nor with heterosis per se, but it is associated with increased stress tolerance, which is consistent with the improvement in the genotype×management interaction. The potential for future yield improvement through increased stress tolerance of maize in the US is large, as yield potential is approximately three times greater than current commercial maize yields. The mechanism by which maize breeders have improved stress tolerance is not known, but we speculate that increased stress tolerance may have resulted from the selection for yield stability. Stability analyses were performed on a number of high-yielding maize hybrids, including three hybrids that have been involved in some of the highest maize yields recorded in producers’ fields, to examine the relationship between yield and yield stability. Results of the stability analyses showed that high-yielding maize hybrids can differ in yield stability, but results do not support the contention that yield stability and high grain yield are mutually exclusive.
Article
Twenty-two different methods (parametric, nonparametric and multivariate) used for analysing genotype×environment (G×E) interaction were compared by applying them to two sets of experimental data (15 faba bean cultivars×12 environments and 11 pea cultivars×16 environments). A principal components analysis was performed on the rank correlation matrix arising from the application of each method. The 22 methods can be categorized, in both sets of experimental data, in three groups: (1) those which are mostly associated with yield level and show little or no correlation with stability parameters; (2) those in which both yield level and stability of performance are considered simultaneously to reduce the effect of G×E interaction; and (3) those methods which only measure stability. This analysis also separated those methods based on an agronomic concept of stability from those which are based on a biological one, as well as distinguishing between `dynamic' and `static' stability-based methods.