A comparison between Joint Regression Analysis and the Additive Main and Multiplicative Interaction model: the robustness with increasing amounts of missing data

Abstract

This paper joins the main properties of joint regression analysis (JRA), a model based on the FinlayWilkinson regression to analyse multi-environment trials, and of the additive main effects and multiplicative interaction (AMMI) model. The study compares JRA and AMMI with particular focus on robustness with increasing amounts of randomly selected missing data. The application is made using a data set from a breeding program of durum wheat (Triticum turgidum L., Durum Group) conducted in Portugal. The results of the two models result in similar dominant cultivars (JRA) and winner of mega-environments (AMMI) for the same environments. However, JRA had more stable results with the increase in the incidence rates of missing values.

Missing value imputation in multi-environment trials: Reconsidering the Krzanowski method

“…For instance, plants might be destroyed by animals, by floods, or during the harvest, while yield measurements may be erroneously carried out or incorrectly entered into the data base (Rodrigues et al 2011). …”

An alternative methodology for imputing missing data in trials with genotype-by-environment interaction: some new aspects

“…Multi-environment trials usually give rise to incomplete data sets (Rodrigues et al, 2011;Bergamo et al, 2008). Possible ways of analysing such trials are: (i) extracting a balanced subset of data by deleting those genotypes or environments that contain missing values (Yan et al, 2011);(ii) filling the missing cells with environmental means; or (iii) filling the missing cells with estimated values obtained from fitted multiplicative or mixed linear models (Kumar et al, 2012;Arciniegas-Alarcón et al, 2011).…”

Imputação múltipla livre de distribuição em tabelas incompletas de dupla entrada

“…Embora os experimentos com interação GxE sejam planejados para serem balanceados, é comum a ocorrência de valores ausentes por diversos motivos, como a retirada de genótipos de baixo desempenho, a consideração de novos genótipos, erros humanos e causas naturais (Rodrigues et al, 2011). Assim, experimentos desbalanceados são usualmente obtidos e não podem ser analisados diretamente por metodologias clássicas eficientes, como a do modelo de efeitos principais aditivos e interação multiplicativa (AMMI) ou da análise biplot GGE (Yan et al, 2007;Yang et al, 2009;Gauch Junior, 2013).…”