From Experimental Network to Meta-Analysis: Methods and Applications with R for Agronomic and Environmental Sciences

ISBN-13: 9789402416954 / Angielski / Twarda / 2019 / 155 str.

David Makowski; Francois Piraux; Francois Brun

Recenzje

"I would recommend the Meta-analysis in medical research book, or the R meta-analysis tutorials that compare different R packages." (Ramzi El Feghali, ISCB News, iscb.info, Issue 69, July, 2020)

Spis treści

Chapter 1. Introduction and examples
Objectives of the analysis of experimental networks and meta-analysis
Data
The type of data
The data collection
Data validation
Analysis
Main steps
Presentation of the tested hypotheses
Collection of data
Data validation
Data analysis
Validation of the analysis
Communication of results
Objective of the book
A simple example of a mixed model
Definition
Data
Model definition
Estimate
Comparison with the model without random effect
References

Part I. Analysis of experimental networks

Chapter 2. Basic Concepts
Agronomic experimentation
Experimental network
Definition
Example of experiment network
Environmental concept
Objectives of network of experiments
Concept of population of environments
Interaction concept
References

Chapter 3. Analysis of network of experiments in blocks of complete randomness as a studied factor
Objective of the chapter
Example "wheat"
Modelization
Model with a random experiment effect
Model with a fixed experimental effect
Example
How to choose between a model with a fixed experimental effect and a model with a random experiment effect?
Model evaluation
Normality
Homoscedasticity
Independence
Suspicious data
Average comparisons
Hypothesis tests: equality tests
Confidence intervals
Hypothesis tests: equivalence tests
Example
Example "wheat": R script and commented analysis
References

Chapter 4. Advanced Methods for Network Analysis

Analysis of average data
Step 1: Analysis of individual experiments to estimate treatment averages
Step 2: Analysis of the average data
Example
A variant: analysis of average data with a fixed model
Estimation of the interaction variance treatment-experimentation
R script
Experiments with heterogeneous variances
Introduction
Example "wheat"
For further
Missing data
Origin of missing data
Adjusted averages
The factors place and year
Goal
Example "wheat_pluri"
Model for analyzing average data
Variance estimation of the treatment-year-place interaction
Variance of the difference between two treatments
Analysis of the example "wheat_pluri" and script R
References

Chapter 5. Planning an Experimental Network
Goal
Comparison of two treatments
Case of a multilocal network
Case of a multi-local and multi-year network
Other contrasts
Average comparison of several witnesses
Comparison to the overall average
References

Part II. The meta-analysis

Chapter 6. Basics for meta-analysis
Definition, origin and main stages of the meta-analysis
Estimated average effect size
Goal
Systematic search of studies, selection of references and data extraction
Estimation of the average effect size with a model without random effect
Estimation of the average effect size with a random effects model
Meta-regression
Goal
Example
Regression models with and without random effect
Example (continued)
Critical analysis of results
References

Chapter 7. Specific statistical problems for the meta-analysis
Setting the effect size
Correction of the bias related to the use of ratios
Difference between observation means
Effect sizes for binary data
Correlation coefficient
Effect sizes based on variance
Generalized linear models for discrete data analysis
Binomial logit model with random effects to analyze the effect of a treatment
Example
Mixed nonlinear models
Interest and definition
Example
Bayesian models
Definition
Example: meta-analysis with MCMCglmm
References

Annex. R resources to implement the methods of analysis networks and meta-analysis
KenSyn package: R code and datasets of the examples presented in the different chapters
Installation
Content and use
Implement the mixed model under R
Adjust a mixed model
Manipulate the results of mixed models under R
The metafor package, dedicated to performing meta-analyzes under R
Bayesian approach with the mixed model
MCMCglmm package

Coda package
References

Blurb

Data analysis plays an increasing role in research, scientific expertise and prospective studies. Multiple data sources are often available to estimate a key parameter or to test a hypothesis of scientific or societal interest. These data, obtained under different environmental conditions or based on different experimental protocols, are generally heterogeneous. Sometimes they are not even directly accessible and should be extracted from scientific articles or reports. However, a comprehensive analysis of the available data is essential to increase the accuracy of estimates, assess the validity of research conclusions and understand the origin of the variability of the experimental results. A quantitative synthesis of the data set available allows for a better understanding of the effects of explanatory factors and for evidence-based recommendations.

Designed as a methodological guide, this book shows the interests and limitations of different statistical methods to analyze data from experimental networks and to perform meta-analyses. It is intended for engineers, students and researchers involved in data analysis in agronomy and environmental science. Our objective is to present the main statistical methods to analyze data from experimental networks and scientific publications. Each chapter exposes one or more methods and illustrates them with examples processed with the R software. Data and R codes are provided and commented in order to facilitate their adaptation to other situations. The codes can be reused from the KenSyn R package associated with this book.