Record Details

Effect of Gliricidia sepium on resource use efficiency and maize grain yield: Statistical analysis

CGSpace

View Archive Info
 
 
Field Value
 
Title Effect of Gliricidia sepium on resource use efficiency and maize grain yield: Statistical analysis
 
Creator Graef, Hannah Arwen
 
Subject intensification
sustainable agriculture
 
Description Human nutrition is challenged by declining resources such as fertile soils and climate change effects. Thus, sustainable, resource use-efficient agricultural practices adapted to these challenges are essential to also allow food production in future – especially in vulnerable rural Tanzania in Africa. Agroforestry, intercropping with nitrogen-fixing trees (e. g. gliricidia (Gliricidia sepium (Jacq.))), has great potential to meet these challenges. Designed experiments within science play an important role in generating new knowledge, aiding in the development of sustainable agriculture. However, the structural components of an experiment need to be treated adequately in the analysis. This thesis studies the analysis of a randomised complete block split-split-plot design.
The hypotheses of this thesis were that maize (Zea mays L.) grain yield is (I) higher under ambient rainfall compared to drought, (II) increased by fertilization and (III) increased by intercropping with gliricidia. Also, (IV) gliricidia effects were expected to replace chemical fertilization effects.
The data analysis of different statistical approaches using R software is discussed critically.
An experiment was conducted in randomized complete block split-split-plot design to assess effects on maize grain yield in semi-arid Tanzania. Main-plot factor were maize intercropping treatments with gliricidia and/ or pigeonpea (Cajanus cajan (L.)), split-plot factor was fertilizer treatment (control and nitrogen and phosphorus fertilizers) and split-split-plot factor was shelter treatment (ambient and drought).
The original dataset was unbalanced containing a missing value. Two additional balanced datasets were created using multiple imputation by chained equations and either PMM (predictive mean matching) or additionally its more recommended extension of DADS (distance-aided donor selection). On these datasets, different methods of model calculation were used: Multi-stratum ANOVA (analysis of variance) versus LMM (linear mixed model), fixed versus random block effect and Satterthwaite versus Kenward-Roger approximation for degrees of freedom.
The multi-stratum ANOVA created with the unbalanced dataset was difficult to interpret. Using either balanced dataset, LMMs (linear mixed models) with fixed block effects with REML (restricted maximum likelihood) using Kenward-Roger or Satterthwaite approximation of freedom led to similar results as from multi-stratum ANOVA. External impact on one main-plot of the experiment caused problems due to supressing estimated negative block variances if LMMs were used. This led to partially different results than the multi-stratum ANOVA. Defining the block as fixed instead of random factor eliminated these issues leading to similar results as from multi-stratum ANOVA. If the estimated negative variance was suppressed by the model (random block effect) leading to a complicated covariance structure or the dataset was unbalanced, results with Kenward-Roger approximation were slightly different than with Satterthwaite approximation. In contrast, if the estimated negative variance was excluded (fixed block effect), both approximations led to the same result. The LMM formulation using ML (maximum likelihood) led to different results, probably also due to supressing an estimated negative variance, and was considered inappropriate.
Before model simplification, all model assumptions were met though difficult to interpret due to the low sample size. However, after model simplification of the balanced dataset from PMM, the assumption of normal distribution of the LMMs was rejected. Of the LMMs, only the residual residues could be extracted using R packages.
The results showed that fertilizer increased yield, confirming the IInd Hypothesis. Unexpectedly, drought increased yield rejecting the Ist hypothesis. Even though this might have been due to waterlogging, experimental errors cannot be ruled out. Both effects were agriculturally very relevant (increases by 100 % and 50 %, respectively). The IIIrd and IVth hypotheses were also rejected, as intercropping did not show any effect and there was no interaction with fertilizer treatment.
In conclusion, under these circumstances, gliricidia effects could not replace fertilizer effects, not improving resource use in terms of nutrients. Further research on agricultural methods improving resource use efficiency is needed. Unbalanced datasets with estimated zero block variance are recommended to be analysed in the unbalanced version using a LMM with REML and Kenward-Roger’s approximation and defining the block as fixed factor. Dataset occurrences such as estimated negative variances emphasize the need for interaction in science to publish statistically correct papers.
 
Date 2021-05
2023-04-21T13:49:01Z
2023-04-21T13:49:01Z
 
Type Thesis
 
Identifier Graef, H.A. 2021. Effect of Gliricidia sepium on resource use efficiency and maize grain yield: Statistical analysis. MSc thesis in Organic Agriculture. Kassel, Germany: University of Kassel.
https://hdl.handle.net/10568/130089
NATURAL RESOURCE MANAGEMENT
 
Language en
 
Rights Other
Open Access
 
Format 97 p.
application/pdf
 
Publisher University of Kassel