Comprehensive evaluation of model uncertainty in qualitative network analyses
Comprehensive evaluation of model uncertainty in qualitative network analyses
复制标题
定性网络分析中模型不确定性的综合评估
DOI：
10.1890/12-0207.1
复制
发表时间：
2012
影响因子：
6.1
通讯作者：
A. Constable
中科院分区：
文献类型：
--
作者：
J. Melbourne‐Thomas;S. Wotherspoon;B. Raymond;A. Constable
定性网络分析提供了广泛的优势，制定的想法和测试的生态系统功能的理解，探索反馈动力学，并在数据有限的情况下进行定性预测。它们已被应用于广泛的生态问题，包括探索基本系统结构的不确定性的影响。然而，我们认为，在定性网络分析模型的不确定性的问题已被探索不足，有必要为一个连贯的框架来评估不确定性。为了解决这个问题，我们已经开发了一个贝叶斯框架来解释不确定性，可以在比较和评估替代模型配方的特性和行为时应用。具体来说，我们认识到，从以前开发的模拟方法定性建模的结果可以被解释为边际似然转换为贝叶斯因子模型比较。然后，我们测试和扩展我们的定性模型结果的贝叶斯解释，以解决替代模型之间和内部的比较。使用的例子，我们展示了我们的贝叶斯解释框架可以提高定性建模的应用，解决生态网络的结构和功能的不确定性。
Qualitative network analyses provide a broad range of advantages for formulating ideas and testing understanding of ecosystem function, for exploring feedback dynamics, and for making qualitative predictions in cases where data are limited. They have been applied to a wide range of ecological questions, including exploration of the implications of uncertainty about fundamental system structure. However, we argue that questions regarding model uncertainty in qualitative network analyses have been under-explored, and that there is a need for a coherent framework for evaluating uncertainty. To address this issue, we have developed a Bayesian framework for interpreting uncertainty that can be applied when comparing and evaluating the characteristics and behavior of alternative model formulations. Specifically, we recognize that results from previously developed simulation approaches to qualitative modeling can be interpreted as marginal likelihoods that translate to Bayes factors for model comparison. We then test and extend our Bayesian interpretation of qualitative model results to address comparisons both between and within alternative models. With the use of examples, we demonstrate how our Bayesian framework for interpretation can improve the application of qualitative modeling for addressing uncertainty about the structure and function of ecological networks.