Coding Smart in Academia: Evidence-Based Software Engineering Approaches for High-Quality Research Software Projects

研究软件描述了在为出版物生成结果的过程中编写的任何代码或使用的软件——例如，用于处理原始数据或进行统计分析的脚本——通常由没有软件工程背景或丰富经验的研究人员编写。研究软件的不准确或效率低下会影响依赖于它的研究。虽然用于研究的其他工具可能会被校准、测试或详细评估，但软件通常会被忽视，并且可能无法识别问题。该项目将从三个重点领域（软件工程技术、软件质量和研究软件项目属性）调查研究软件，以更好地理解用于开发软件的软件工程技术，如何最好地评估研究软件的质量，以及哪些研究软件项目因素对开发高质量产出有影响。它将考虑在传统软件开发中广泛使用的见解或技术是否可以在研究软件开发中使用。该项目还将从选定的研究软件项目中收集经验数据，为研究软件工程社区确定基于证据的建议，以帮助构建更高质量的研究软件。为了实现这些目标，本项目将评估评估研究软件质量的方法，并考虑开发研究软件的项目的背景和属性，确定这些可能如何影响生成的软件。大多数先前的研究和文献都涉及“科学软件”。该项目将涉及更广泛的“研究软件”领域，这是一种更具包容性和跨学科的方法，也可以从用于艺术和人文学科研究的软件中获取实践。之前的深入研究发生在研究软件项目的广泛机构支持和“研究软件工程师”角色出现之前。这项研究将更新关于这些景观变化是否表明实践变化的文献。许多现有的关于研究软件的研究提供了关于最佳实践的建议，然而，很少有经验证据（特别是定性实验数据）来证明这些建议是有效的，而它们可以有效应用于的研究软件项目的范围并没有清楚地显示出来。为了解决这一证据差距，本项目将确定潜在的改进和更有效的软件工程方法或技术，目前在研究软件项目中未充分使用。这些建议的干预措施将在一组有代表性的研究软件项目中测试其影响。紧密匹配项目的“控制”组（例如：相似的软件工程技术使用，基线研究软件质量，以及项目背景，如开发团队规模）将不接受干预。关于软件开发实践和项目背景的数据将在研究期间通过调查和访谈从参与项目的两组中收集。这些选定项目中的研究软件质量也将在整个实验期间进行测量，以评估这些干预对软件质量的影响。建议的干预措施将在此实验测试的基础上进行完善，为一系列研究软件项目遇到的当前研究软件挑战提供可靠的循证和经过验证的解决方案。这项研究的结果将使研究人员能够确定哪些软件工程技术和方法是最有效的，并将适合他们的项目（列出已知的好处或缺点），并演示如何评估和提高他们的研究软件的质量，所有的建议都有数据的有力支持。

Research software describes any code written or software used in the process of generating results for a publication - for example, a script used to process raw data or carry out a statistical analysis - and is commonly written by researchers without a software engineering background or extensive experience. Inaccuracies or inefficiencies in research software impact the research that relies on it. While other tools used for research might be calibrated, tested, or evaluated in detail, software can often be overlooked, and problems might not be identified. This project will investigate research software from three key areas of focus (software engineering techniques, software quality, and research software project properties), to gain a better understanding of the software engineering techniques used to develop it, how to best assess the quality of research software and which research software project factors are influential in developing high quality outputs. It will consider whether insights or techniques used widely within traditional software development could be of use within research software development. The project will also gather empirical data from selected research software projects to identify evidence-based recommendations for the research software engineering community to help build better-quality research software. To achieve these aims, this project will assess methods for evaluating research software quality and consider the contexts and properties of projects developing research software, identifying how these might impact the software generated. Most previous research and literature have dealt with 'scientific software'. This project will address the wider area of 'research software' - a more inclusive and interdisciplinary approach which also captures practices from software used for research within the arts and humanities. Previous in-depth studies took place before widespread institutional support of research software projects and the emergence of a 'research software engineer' role. This study will update the literature on whether these landscape shifts indicate changes in practice. Many existing studies covering research software offer recommendations regarding best practices, however, little empirical evidence (particularly qualitative experimental data) is given to demonstrate that these recommendations are effective, while the range of research software projects they can be validly applied to is not clearly shown. To address this evidence gap, this project will identify potential improvements and more effective software engineering methods or techniques currently under-used within research software projects. These suggested interventions will be tested for their impacts within a representative set of research software projects. A 'control' group of closely matched projects (for example: similar software engineering technique usage, baseline research software quality, and project contexts such as development team sizes) will not receive an intervention. Data about software development practices and project contexts will be gathered from both groups of participating projects via surveys and interviews over the study period. Research software quality within these selected projects will also be measured throughout the experimental period to assess the impacts of these interventions on software quality. The recommended interventions will be refined based on this experimental testing, generating robust evidence-based and verified solutions to current research software challenges encountered by a range of research software projects. The outcomes of this study will allow researchers to identify which software engineering techniques and methods are most effective and would suit their project (listing known benefits or drawbacks) and demonstrate how to assess and improve the quality of their research software, with all suggestions strongly supported by data.