The outbreak of COVID-19 in 2020 inhibited face-to-face education and constrained exam taking. In many countries worldwide, high-stakes exams happening at the end of the school year determine college admissions. This paper investigates the impact of using historical data of school and high-stakes exams results to train a model to predict high-stakes exams given the available data in the Spring. The most transparent and accurate model turns out to be a linear regression model with high school GPA as the main predictor. Further analysis of the predictions reflect how high-stakes exams relate to GPA in high school for different subgroups in the population. Predicted scores slightly advantage females and low SES individuals, who perform relatively worse in high-stakes exams than in high school. Our preferred model accounts for about 50% of the out-of-sample variation in the high-stakes exam. On average, the student rank using predicted scores differs from the actual rank by almost 17 percentiles. This suggests that either high-stakes exams capture individual skills that are not measured by high school grades or that high-stakes exams are a noisy measure of the same skill.

2020 年 COVID-19 的爆发抑制了面对面的教育并限制了考试。在全球许多国家，学年末的高风险考试决定了大学录取。本文研究了使用学校历史数据和高风险考试结果来训练模型以预测春季可用数据的高风险考试的影响。最透明和最准确的模型是一个以高中 GPA 为主要预测变量的线性回归模型。对预测的进一步分析反映了高风险考试与高中 GPA 的关系，对于不同的人群而言。预测分数略微有利于女性和低 SES 的人，她们在高风险考试中的表现比在高中时相对差。在高风险考试中，我们首选的模型占样本外变异的 50% 左右。平均而言，使用预测分数的学生排名与实际排名相差近 17 个百分点。这表明，要么高风险考试捕捉到了高中成绩无法衡量的个人技能，要么高风险考试是对同一技能的嘈杂衡量。