The advantage of regression and covariate utilisation over ANOVA in engineering education research

Despite the availability and potential usefulness of demographic and contextual data in many quantitative studies within engineering education, the preference for ANOVA over regression models remains prevalent, often without clear justification. A mapping review of literature from the EJEE and JEE spanning 2012–2022 identified 98 studies using ANOVA or t-tests. Further investigation suggested that 62% and 78% of these studies, in the two flagship journals, respectively, potentially could have benefitted from employing regression analyses. This article argues for the broader adoption of regression analysis when appropriate, and its potential superiority over ANOVA in numerous contexts based on statistical and practical considerations. This argument is supported through both conceptual discussions and statistical evidence, alongside simulation studies utilising three widely used databases. The findings of simulation studies showed that regression analyses not only use demographic and contextual data more effectively but also offer enhanced analytical power, accuracy, and robustness against Type I errors, particularly in identifying disparities in educational outcomes and exploring intersectionality. Thus, we advocate for a more proactive consideration of adopting regression methods, especially integrating meaningful covariates, for research design and method choice.尽管人口统计学与情境数据在工程教育领域的诸多定量研究中已可获取且具备潜在应用价值，但学界仍普遍更倾向于使用方差分析（ANOVA）而非回归模型，且此类选择往往缺乏明确依据。本研究针对2012—2022年刊载于《EJEE》与《JEE》的文献开展范围综述，共筛选出98项使用方差分析（ANOVA）或t检验的研究。进一步分析显示，这两份旗舰期刊中分别有62%与78%的此类研究，本可通过采用回归分析获得更优研究结果。本文主张在适宜场景下更广泛地应用回归分析，并基于统计学与实践层面的考量，论证回归分析在诸多场景中相较方差分析的潜在优势。本文通过概念性论述与统计学实证，结合基于三类通用数据库开展的模拟研究，为上述主张提供支撑。模拟研究结果表明，回归分析不仅能更高效地利用人口统计学与情境数据，还可提升分析效力、准确性以及对一类错误（Type I error）的鲁棒性，尤其在识别教育结果差异与探索交叉性议题时优势更为显著。因此，本文倡议在研究设计与方法选择环节，应更主动地考量采用回归分析方法，尤其是纳入具有研究价值的协变量（covariates）。