Elsevier

Field Crops Research

Volume 121, Issue 3, 3 April 2011, Pages 450-459
Field Crops Research

Stability analysis of farmer participatory trials for conservation agriculture using mixed models

https://doi.org/10.1016/j.fcr.2011.02.001Get rights and content

Abstract

Normally, the data generated from farmer participatory trials (FPT) are highly unbalanced due to variation in the number of replicates of different treatments, the use of different varieties, farmers’ management of the trials, and their preferences for testing different treatments. The incomplete nature of the data makes mixed models the preferred class of models for the analysis. When assessing the relative performances of technologies, stability over a range of environments is an important attribute to consider. Most of the common models for stability may be fitted in a mixed-model framework where environments are a random factor and treatments are fixed. Data from on-farm trials conducted in the Indo-Gangetic Plain (IGP) of South Asia under the umbrella of Rice–Wheat Consortium (RWC) were analyzed for grain yield stability using different stability models. The objective was to compare improved resource management technologies with farmers’ practice. The variance components of an appropriate mixed model serve as measures of stability. Stability models were compared allowing for (i) heterogeneity of error variances and (ii) heterogeneity of variances between environments for farmers-within-environment effects. Mean comparisons of the treatments were made on the basis of the best fitting stability model. Reduced-till (non-puddled) transplanted rice (RT-TPR) and reduced-till drill-seeded wheat using a power tiller – operated seeder with integrated crop and resource management RTDSW(PTOS)ICRM ranked first in terms of both adjusted mean yield and stability.

Research highlights

► Alternative resource conserving technologies were compared with the currents farmers practice of rice and wheat for grain yield on farmer's field in the Indo-Gangetic Plains of South Asia. ► The data generated from these farmers’ participatory trials were highly unbalanced. ► Assessing the relative performances of technologies over a range of environments, stability analysis for grain yield was carried out using SAS mixed model procedures. ► Stability models were compared allowing for (i) heterogeneity of error variances and (ii) heterogeneity of variances between environments for farmers-within-environment effects. ► Based on best fitting model, Reduced-till (non-puddled) transplanted rice (RT-TPR) and reduced-till/zero-till drill-seeded wheat were found best in terms of grain yield and stability.

Introduction

On-farm farmer participatory trials (FPT) involve active participation by the farmers in evaluating a technology under a wide range of farm conditions. The purpose and strength of farmer participatory testing lies in effectively assessing the effect of farmer resources and management on the technology (Petersen, 1994). FPTs are conducted across different locations and years to test different technologies within the target area. The data generated from these multi-environment trials are highly unbalanced due to variations in the number of replications of different treatments, the choice of different varieties, farmers’ management of the trials and their preferences for testing any subset of treatments. The variability inherent in on-farm and participatory work can produce irregularity in design and the need for more flexible statistical methods than are normally available to researchers (Stroup et al., 1993). The statistical methods used for the analysis of multi-environment varietal testing across years and sites can serve as useful tools in analyzing data from on-farm participatory trials (Riley and Alexander, 1997). These include the simple analysis of variance, regression approach, multivariate methodologies, the additive main effects and multiplicative interaction (AMMI) model as well as the powerful and flexible mixed-model which uses restricted maximum likelihood methodology (REML) algorithm.

Kidula et al. (2000) suggested that the traditional methods of analysis of variance (ANOVA) which emphasize hypothesis testing and significance levels cannot accommodate the complexity of FPTs data. This is largely because the main focus in FPTs is on prediction and taking action. Therefore, they recommended a combination of mixed-model procedures and stability analysis in order to arrive at meaningful conclusions. Likewise, Coe (2007) pointed out that the standard tools based on ANOVA are not appropriate. He demonstrated with examples the usefulness of mixed-model methodology for analysing FPT data with irregular design. Virk et al. (2009) successfully used restricted maximum likelihood (REML) to analyze quantitative traits of very highly unbalanced on-farm participatory varietal selection (PVS) trials. Parsad et al. (2009) emphasized the usefulness of mixed models to analyze the data generated from FPTs, considering the farmer or field effects as random and treatment effects as fixed.

When assessing the relative performance of various technologies/practices, stability of their performances is an important attribute to consider. Stability can be ascertained using various stability statistics (for review see Lin et al., 1986, Westcott, 1987, Becker and Leon, 1988, Piepho, 1998a). Traditional measures of stability include environmental variance (Lin et al., 1986), coefficient of variation (Francis and Kannenberg, 1978) and Shukla's stability variance (Shukla, 1972). Modified stability analysis as suggested by Hildebrand (1984) used the regression approach of Finlay and Wilkinson (1963) and Eberhart and Russell (1966) to assess the stability of treatments under different farmer management systems over a wide range of environmental conditions. There are three indicators of stability in regression analysis (i) coefficient of regression (b), (ii) variance of deviations from regression (sb2), and (iii) treatment mean. The regression coefficients which have a mean of unity indicate the rate at which the performance of a treatment varies relative to the changes in the environment. The variance of deviations indicates the reliability of the regression relationship. Denis et al. (1997) and Piepho (1999) suggested that most of the common stability measures may be embedded in a mixed-model framework where environments are a random and treatments are a fixed factor. Piepho (1999) showed how mixed model analyses of unbalanced data for the most common stability measures are readily available through the variance structures fitted using SAS procedure MIXED. Piepho and van Eeuwijk (2002) demonstrated with a realistic example the choice of an appropriate model and the interpretation of variance components as measures of stability. In their analysis the environments (locations, years) were considered as random factor and genotypes as fixed. An alternative approach to the regression analysis is the additive main effects and multiplicative interaction (AMMI) model (Kempton, 1984, Zobel et al., 1988, Gauch, 1992). The AMMI model was originally proposed as a fixed effects model. Assuming environments (or treatments) as random, the treatment × environment interaction can be analyzed in a mixed-model framework with a factor-analytic covariance structure to model the multiplicative terms (Piepho, 1997b).

The rice–wheat (RW) system is the lifeline of millions of food producers and consumers in the Indo-Gangetic Plains (IGP) of South Asia. The system has contributed to reducing poverty and hunger during the Green Revolution. Recently, however, widespread stagnation or decline of crop productivity and rising cost of cultivation have been reported. Since the demand of these two cereals has been projected to increase by more than 50% in 2020 and resources such as water and labor are going to be scarce, enhancing productivity and input-use efficiency are urgently needed. A number of improved land and crop management practices suitable for farmers in the region often termed as resource-conserving technologies (RCTs) have been developed and disseminated in the IGP under the umbrella of the Rice–Wheat Consortium (RWC) (for review see Gupta et al., 2002, Ladha et al., 2009b). The new RCTs have been integrated into the existing portfolio of technologies already being practiced by farmers in the framework of integrated crop and resource management (ICRM) (Ladha et al., 2009a). ICRM includes optimal land preparation, water management, crop establishment as well as nutrient, pest and weed management.

Researchers and extensionists evaluate rice–wheat production component technologies within the framework of ICRM in farmers’ fields in order to promote the successful ones at large to enhance sustainability and profitability of the farmers. On-farm farmer participatory trials are being conducted to compare various improved component technologies (RCTs) such as reduced, zero-tillage, drill-seeding either on flat or on raised beds, and nutrient management with that of typical farmer practice. The trials wherein the RCTs form the treatments are fully managed by the participating farmers in their fields under a range of conditions with a view to assess their overall performance and consistency. Often, the FPTs are conducted without a proper design at different locations/farmer fields and years. The data generated have large variability which complicates analysis.

The objectives of the paper were to (a) illustrate the yield stability analysis of a highly unbalanced data that originated from FPTs, incorporating heterogeneity of error variances and heterogeneity of variances between environments at farmer level, and (b) perform mean comparisons for the treatments using the best fitting stability model. The data used in this analysis came from 1985 on-farm FPTs conducted during 2005–2008 at various sites in Bangladesh, India and Nepal. Appendix A provides the list and brief description of technologies used.

Section snippets

Trial management

The study was conducted at 6 locations in the IGP of Bangladesh (Kushtia and Dinajpur), India (Modipuram, Karnal and Ballia) and Nepal (Bhairahawa) from 2005 to 2008. Table 1 provides the list of number of participating farmers and technologies at different sites. The trials were researcher designed and farmer managed. Farmers used a wide range of rice and wheat varieties. The most common rice varieties were Swarna (MTU = 7029) and Sarjoo – 52. Other rice cultivars included Sambha Mahsuri (BPT –

ANOVA

The conventional analysis of variance for grain yield for both rice and wheat showed highly significant (p < 0.0001) environment (ej), treatment × environment interaction [(te)ij] and farmer within environment effects (fkj). Treatment effects were highly significant (p < 0.0001) in case of wheat, whereas treatments were non-significant (p = 0.0615) at 5% level in case of rice. A plot of residual vs. predicted values for the model that accommodates heterogeneity of error variances across environments

Discussion

With on-farm/farmer participatory trials new technologies are tested for wide adoption. The difficulty in analyzing data from these trials is the great variation among the chosen farmers due to the quality and quantity of their resources and their methods of evaluation (Nair, 1993). Farmers are as heterogeneous as their environments (Crossa et al., 2002). Within these trials, information can be at various levels, i.e. sites, farms within sites or plot within farms, and variation in the

References (45)

  • V. Kumar et al.

    Role of herbicide-resistant rice in promoting resource conservation technologies in rice–wheat cropping systems of India: a review

    Crop Prot.

    (2008)
  • H. Pathak et al.

    Trends of climatic potential and on-farm yield of rice and wheat in the Indo-Gangetic Plains

    Field Crops Res.

    (2003)
  • H.C. Becker et al.

    Stability analysis in plant breeding

    Pl. Breed.

    (1988)
  • J. Burgueno et al.

    Using factor analytic models for joining environments and genotypes without crossover genotype × environment interaction

    Crop Sci.

    (2008)
  • R. Coe

    Analysing data from participatory on-farm trials

    Afr. Stat. J.

    (2007)
  • J. Crossa et al.

    A quantitative method for classifying farmers using socioeconomic variables

  • J.B. Denis et al.

    Modelling expectation and variance for genotype by environment data

    Heredity

    (1997)
  • S.A. Eberhart et al.

    Stability parameters for comparing varieties

    Crop Sci.

    (1966)
  • K.W. Finlay et al.

    The analysis of adaptation in a plant-breeding program

    Aust. J. Agric. Res.

    (1963)
  • M. Fikere et al.

    Genotype–environment interactions and stability parameters for grain yield of faba bean (Vacia faba L.) genotypes grown in South Eastern Ethiopia

    Int. J. Sustain. Crop Prod.

    (2008)
  • T.R. Francis et al.

    Yield stability studies in short-season maize. 1. A descriptive method for grouping genotypes

    Can. J. Plant Sci.

    (1978)
  • H.G. Gauch

    Statistical Analysis of Regional Yield Trials. AMMI Analysis of Factorial Designs

    (1992)
  • R.K. Gupta et al.

    Resource Conserving Technologies: Transforming the Rice–Wheat Systems of the Indo-Gangetic Plains

    (2002)
  • L.W. Harrington et al.

    The Rice–Wheat Consortium and the Asian Development Bank: a history

  • P.E. Hildebrand

    Modified stability analysis of farmer managed, on-farm trials

    Agron. J.

    (1984)
  • A.M. Kelly et al.

    The accuracy of varietal selection using factor analytic models for multi-environment plant breeding trials

    Crop Sci.

    (2007)
  • R.A. Kempton

    The use of biplots in interpreting variety by environment interactions

    J. Agric. Sci.

    (1984)
  • L.N. Kidula et al.

    Statistical analysis of on-farm trials in soil management project in South West Kenya. Participatory technology development for soil management by small holders in Kenya

  • Kumar, V., Ladha, J.K., 2011. Direct-seeding of rice: recent developments and future research needs. Adv. Agron. 111,...
  • J.K. Ladha et al.

    Integrating crop and resource management technologies for enhanced productivity, profitability and sustainability of the rice–wheat system in South Asia

  • R.C. Littell et al.

    SAS for Mixed Models

    (2006)
  • Cited by (0)

    View full text