COLLABORATIVE RESEARCH: META-ANALYSIS: EVALUATION AND IMPROVEMENT OF AN IMPORTANT SYNTHETIC TOOL

在环境生物学中，对同一主题的不同研究可能会得出非常不同的结论。有时，这些分歧是因为不同的方法或回答的问题略有不同，但其他时候，它们是由于被研究的年份、地点或物种之间的真正差异。重要的是，科学家能够区分这些可能性并自信地得出结论。例如，一些研究表明，一起种植不同的作物会给农民带来更高的整体产量，而另一些研究则表明，种植单一类型的作物会产生最高的产量。这些相互矛盾的结果是否反映了本质上的随机性、科学家在这两种类型的研究中使用的不同方法，还是其他原因？一种叫做荟萃分析的统计学方法已经被用来解决这个问题。荟萃分析帮助了许多学科，因为它允许科学家将许多研究的结果结合在一起，考虑到它们的不同之处，并提供否则会被隐藏的见解。这个项目将使用先进的技术来改进元分析的完成方式。研究人员将专注于将该技术应用于生态学领域。从事该项目的早期职业科学家将接受现代生态学方法的培训，包括荟萃分析。该项目还将开发可公开使用的在线材料，从而帮助教育全国各地的生态学家使用元分析方法，并改进元分析在生态问题中的应用。为了提高元分析在生态学中的应用，该项目将：1)系统地回顾最近的生态元分析，描述它们通常是如何进行的(即统计模型和计算影响大小的方式)和数据集的特征(例如，研究的数量、每项研究的样本量、研究间和研究内差异的大小)；2)在生态数据集中可能遇到的一系列条件下(例如，使用样本大小和在生态学中观察到的不同变异源的大小)，使用模拟来评估现有荟萃分析模型和建议的替代方案的性能(例如，加权方案不同，使用贝叶斯方法与频率方法，或使用非独立性的不同调整)。这些结果将有助于提高国家汇集和准确解释复杂的生态研究结果的能力。

In environmental biology, different studies on the same topic can reach very different conclusions. Sometimes those disagreements are because of different methods or slightly different questions being answered, but other times they're due to true differences between years, places or species being studied. It is important that scientists be able to distinguish between these possibilities and confidently reach conclusions. For example, some studies suggest that growing different crops together results in higher overall yields for farmers, whereas other studies suggest that growing a single type of crop results in the highest yields. Do these conflicting results reflect some randomness in nature, different methods used by scientists in the two types of studies, or some other reason? A statistical approach called meta-analysis has been developed to solve this problem. Meta-analysis has helped many disciplines because it allows scientists to combine results from many studies, taking into account how they differ and providing insights that would otherwise remain hidden. This project will use advanced techniques to improve how meta-analyses are done. Researchers will focus on using the technique in the field of ecology. Early career scientists engaged in this project will be trained in modern ecological methods, including meta-analysis. This project also will develop online materials that will be available publicly, thus helping educate ecologists across the nation in meta-analytic methods and improving the application of meta-analysis to ecological problems.To improve the application of meta-analysis in ecology, this project will: 1) systematically review recent ecological meta-analyses to describe how they are typically performed (i.e., the statistical model and the way the size of an effect is calculated) and the characteristics of the dataset (e.g., the number of studies, the sample sizes of each study, and the magnitude of among-study and within-study variation); 2) use simulations to evaluate the performance of existing meta-analysis models and proposed alternatives (e.g., differing in weighting schemes, using Bayesian vs. frequentist approaches, or using different adjustments for non-independence), under a range of conditions likely to be encountered in ecological datasets (e.g., using sample sizes and the magnitudes of different sources of variation observed in ecology). The results will be useful for improving the nation's ability to bring together and accurately interpret the complicated results of ecological studies.