Centralized Admissions for Engineering Colleges in India
Proceedings of the 2019 ACM Conference on Economics and Computation(2019)
摘要
The centrally funded technical institutes (CFTIs) for undergraduate studies in India consist of the prestigious Indian Institutes of Technology (IITs) as well as several non-IITs (the National Institutes of Technology, the Indian Institutes of Information Technology, and others). The IITs use the candidate rankings obtained from performance on the JEE Advanced exam for their admissions, whereas the non-IITs use a different set of rankings (obtained from the JEE Main exam) for their admissions. Until 2014, the IITs and the non-IITs used two separate processes to allocate seats to candidates. Every year, several individual candidates would get two seats, one from each of these processes. As a result, at least one of those two seats would go vacant. This would especially be a problem for the IITs as their allocation process would complete before that of the non-IITs. In 2015, we designed and implemented a new joint (centralized) seat allocation process for undergraduate admissions to over 500 programs spread across 80 CFTIs in India, bringing the IIT and the non-IIT processes together under a single umbrella. Our process asks each candidate to submit a single preference list over all these programs, and allots each candidate at most one seat. Our process is based on the well known Deferred Acceptance algorithm, which produces a fair allocation that respects the candidate rankings. However, complex affirmative action seat reservations led us to make a number of algorithmic innovations, including (i) a carefully constructed heuristic for incorporating non-nested common quotas that span multiple programs, (ii) a method to utilize unfilled reserved seats with no modifications to the core software, and (iii) a robust approach to reduce variability in the number of reserved category candidates admitted, while retaining fairness. Because candidates also have outside options (e.g., attending a privately funded institute, or retaking the entrance exams the following year), and we were not allowed, in anticipation of attrition, to speculatively admit more candidates than the capacity permitted, we resorted to a multi-round seat allocation process. After every round, we ask the candidates to accept their allotted seat (if any) by paying a seat acceptance fee by a deadline. Those who don't vacate their seats, and these seats are subsequently allotted to other candidates in future rounds. We implemented a number of process innovations to make the multi-round process smoother and more efficient. These innovations are described in our full paper \citejournalPaper Our new seat allocation process went live in 2015, and has remained in successful use since, with continuing improvements. The vacancies at the IITs have reduced by nearly 70% (e.g., there were only 198 vacancies in 10,988 seats in 2017, relative to 587 vacancies in 9,784 seats in 2014 under the legacy process). Meanwhile, vacancies in the non-IITs have reduced only slightly, and concern us, though reducing these was outside our initial mandate. In 2017, the non-IITs had 6,510 vacancies in their 25,220 seats, when classes began. Data indicates that almost all these vacancies are avoidable in the sense that there were eligible candidates who want these seats. The key causes of the vacancies at the non-IITs are revealed by our analysis of the admissions data: (i) Candidates who accept a seat are allowed to surrender them until the penultimate round (resulting in a refund of their seat acceptance fee), and over 70% of fresh allocations in late rounds at the non-IITs are rejected by candidates. Since most withdrawals occur just before the deadline, only a small fraction of those seats are successfully filled. (ii)The number of withdrawals is very large (there were 5,525 withdrawals in 2017). Seemingly, candidates are listing programs and accepting seats they don't really want (knowing that they can withdraw later and recover their seat acceptance fee, and face no penalty).
更多查看译文
关键词
affirmative action, algorithm, college admission, deferred acceptance, implementation, market design, stable matching
AI 理解论文
溯源树
样例
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要