一句话：通过软件测试课程中906份自评/同伴评分与248份助教评分的对比，发现同伴评分可较好近似助教评分。
大规模班级的评分成本高，荷兰代尔夫特理工大学学生人数增长使问题更突出。作者在2018-2019学年软件测试课程中引入自评与同伴评分：学生两人一组，对三次作业进行自评和同伴评分。研究收集了906份自评/同伴评分提交，并与248份助教评分进行比较。结果显示：自评平均比同伴评分高8-10%；同伴评分似乎是助教评分的良好近似；当自评与同伴评分差异显著时，助教评分往往介于两者之间；学生性别和国籍似乎不影响自评与同伴评分。
Grading large classes has become a challenging and expensive task for many universities. The Delft University of Technology (TU Delft), located in the Netherlands, has observed a large increase in student numbers over the past few years. Given the large growth of the student population, grading all the submissions results in high costs. We made use of self and peer grading in the 2018-2019 edition of our software testing course. Students worked in teams of two, and self and peer graded three assignments in our course. We ended up with 906 self and peer graded submissions, which we compared to 248 submissions that were graded by our TAs. In this paper, we report on the differences we observed between self, peer, and TA grading. Our findings show that: (i) self grades tend to be 8-10% higher than peer grades on average, (ii) peer grades seem to be a good approximator of TA grades; in cases where self and peer grade differ significantly, the TA grade seems to lie in between, and (iii) the gender and the nationality of the student do not seem to affect self and peer grading.