Affirmative Action's Cumulative Fractional Assignments
summary
The gist
India’s reservation policy entitles designated groups to fixed fractions of faculty positions Positions are indivisible, arrive over time across departments…
In short
The paper investigates why India's reservation policy faced conflict: trying to maintain cumulative reserved counts and exact rounding at both department and university levels is mathematically impossible. This incompatibility forced errors that grew infinitely, explaining the litigation. A random-roster solution offers a practical compromise by preserving fairness in expectation while concentrating unavoidable aggregate errors at the university level.
Key concepts
- Monotonicity
- This means that as time passes and more positions are reserved, the total number of reserved spots for a group should never decrease. It ensures that the cumulative count of reserved seats only ever stays the same or increases over time, which is a basic requirement for any fair allocation system.
- Department Quota Compliance
- This constraint requires that each specific department's actual number of appointments must be very close to its assigned fraction of the total seats. The system must ensure that no single department is severely over- or under-represented compared to its entitlement.
- Random Roster Solution
- This is a proposed method where assignments are made randomly, but with careful engineering. It successfully meets departmental fairness and maintains the non-decreasing nature of reservations over time. It trades exact aggregate university compliance for expected proportionality, making it implementable in practice.
Terminology used across episodes
This episode discusses
The paper
Affirmative Action's Cumulative Fractional Assignments · Read on arXiv
Manshu Khanna, Haydar Evren
Peking University HSBC Business School · Akuna Capital
India's reservation policy entitles designated groups to fixed fractions of faculty positions. Positions are indivisible, arrive over time across departments, and appointments are irreversible. We show that no rule can keep cumulative reserved counts nondecreasing while rounding entitlements to adjacent integers at both the department and university levels; university-level compliance forces department-level errors that grow without bound. This impossibility explains why both solutions practiced in India provoked litigation and protests. A random-roster solution preserves irreversibility and department-level compliance, matches entitlements in expectation, and concentrates university-level errors. Dropping cumulative accounting restores exact rounding at both levels.
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: We're looking at the paper: "Affirmative Action's Cumulative Fractional Assignments".
Warren: Here's the gist of the paper. India’s reservation policy entitles designated groups to fixed fractions of faculty positions Positions are indivisible, arrive over time across departments, and appointments are irreversible We show that no rule can keep cumulative reserved counts nondecreasing while rounding entitlements to adjacent integers at both the department and university levels University-level compliance forces department-level errors that grow without bound This impossibility explains why both solutions practiced in India provoked litigation and protests A random-roster solution preserves irreversibility and department-level compliance, matches entitlements in expectation, and concentrates university-level errors.
Ruth: The Core Conflict of Constraints The paper investigates the incompatibility between three natural requirements for reservation systems: monotonicity (cumulative reserved counts never decrease), department quota compliance (each department’s cumulative allocation stays within one seat of its fractional entitlement), and university quota compliance (the same at the university level).
Warren: The central result demonstrates that these three requirements are mutually incompatible.
Ruth: Specifically, monotonicity together with exact university-level compliance forces department-level distortions that grow without bound This conflict is illustrated by comparing the Government’s solution, which pools vacancies at the university level, and the Court’s solution, which mandates department-level rosters.
Paper summary: Ruth: So, we're looking at this paper now called "Affirmative Action's Cumulative Fractional Assignments." The authors are pointing out a fundamental problem with how India handles faculty reservations. Essentially, they claim that the policy gives certain groups fixed fractions of positions to be filled, and those positions are indivisible units that come over time across different departments, and appointments can't be reversed.
Warren: That sounds like a classic allocation problem where you have these hard constraints on top of each other, trying to keep things fair while dealing with real-world limitations like whole seats. So what’s the core issue they’re highlighting in this paper?
Ruth: The core issue is that you can't satisfy three requirements simultaneously: keeping a running total of reserved spots nondecreasing, making sure each department stays very close to its exact fractional entitlement, and also making sure the overall university allocation matches the university level exactly. <ref:2111.11963#pg3>
Iris: That sounds like a constraint structure where monotonicity combined with exact university-level compliance forces department-level distortions that grow without bound, which is quite a strong result. But what does this mean for someone trying to design a system?
Warren: It means that if you try to enforce the rules at the university level, you inevitably get massive errors at the departmental level that keep getting bigger and bigger. That impossibility is what explains why both of the previous ways of solving this in India ended up causing legal trouble and protests.
Ruth: Exactly. The paper shows that this conflict isn't just a tricky implementation detail; it’s a structural impossibility under deterministic rules, which is why we see these kinds of real-world friction. It’s about how the accounting works over time across different levels of organization, not just the initial allocation.
Iris: So if you look at the math behind it, they prove that there isn't even one deterministic solution that can be both monotonic and aggregately proportional with a finite bias. <ref:2111.11963#pg3>
Paper summary: Warren: That’s a big statement because it rules out any simple, perfectly calculated roster you could just write down on paper. If you try to build a system where every single department gets its exact share and the university gets its exact share, but you have to track the cumulative count going up, it just breaks under continuous recruitment.
Ruth: The authors propose a principled second best response here by constructing what they call a random roster solution. They show that you can preserve monotonicity and make sure each department meets its quota exactly, but they trade exact aggregate compliance for something else—ex ante proportionality and bounds on violations. <ref:2111.11963#pg3>
Iris: But what does this random roster actually do? Does it solve the problem entirely, or is it just managing the fallout? What’s the mechanism behind this randomization they describe?
Warren: The construction involves breaking down that bihierarchical constraint structure, using tools from Budish et al. <ref:2111.11963#pg2> and Akbarpour and Nikzad two thousand twenty to engineer cross-department independence. That independence is what allows them to use concentration properties, specifically Chernoff bounds from Akbarpour and Nikzad two thousand twenty which ensures the university-level deviations concentrate in a manageable way <ref:2111.11963#pg1>.
Ruth: So the result is that because those departments draw their rosters independently based on this randomized distribution, the probability of any single category deviating from its entitlement at the university level stays within certain bounds. It’s a trade-off between perfect exactness and being practically implementable across all departments.
Iris: I wonder what this means for identification? If we use this randomized approach, what variation are we actually tracking that allows us to say it works? What would falsify their claim that the deviations concentrate?
Warren: The paper suggests that the concentration of university-level errors is the key mechanism here. It’s not about achieving perfect compliance everywhere, but managing where those inevitable small misses land, and proving that they don't spiral out of control across different departments.
Ruth: And this directly relates back to the real world situation in India, where both the government’s method—pooling vacancies at the university level—and what courts have mandated—a department-level roster—led to problems because they couldn't satisfy all these constraints. <ref:2111.11963#pg3>
Paper summary: Iris: So, if we take this result outside of the specific context of Indian faculty positions, what’s the general principle this paper is establishing about institutional design? What are the ingredients that lead to this impossibility?
Warren: The ingredients they identify are very specific: having entitlements stated as fractions, allocating indivisible units at two intersecting levels, and keeping a running account of past allocations. That running account is what forces the cumulative accounting that leads to the unbounded bias.
Ruth: It really highlights that the problem isn't just about how many people you hire; it’s about how you manage the counting and tracking of those hires across different organizational boundaries. The line separating these kinds of problems is defined by the accounting convention, not just by whether you use fractions or indivisible items.
Iris: So, what’s the big picture implication for policy makers looking at any system that uses fractional entitlements and discrete allocations across multiple tiers? What’s the practical advice here?
Warren: The paper suggests that in current practice, the random roster construction is a principled second best solution. It keeps things functional while managing expectations about aggregate compliance. But if you want to restore exact rounding at both levels, you have to drop that cumulative accounting altogether. <ref:2111.11963#pg3>
Ruth: And that brings us to the conclusion of this paper, "Affirmative Action's Cumulative Fractional Assignments." It shows that the struggle wasn't a failure of imagination, but a result of those mutually incompatible constraints we just talked about. The key is choosing between a deterministic system that fails or a randomized one that manages the errors predictably.
Iris: So for those listening who are interested in how this applies broadly, it seems like any institution with fractional quotas and two levels of proportionality needs to decide which constraint to relax: either you keep the running total count perfect, or you allow the departmental allocations to be slightly off while focusing on controlling the aggregate error.
Warren: That’s right. The paper lays out that there are clear dividing lines based on whether those entitlements are cumulative and never reset or if the account is adjusted with turnover each cycle. That distinction dictates which theoretical path you're allowed to take for implementation.
Conclusion: Ruth: So, we’re wrapping up our look at this paper on "Affirmative Action's Cumulative Fractional Assignments." It boils down to showing that the way India handles faculty reservations creates a structural impossibility when you try to enforce everything at once.
Warren: Exactly. The main point they nail is that you can't have a system where the running total of reserved spots always goes up, and each department stays perfectly close to its fraction, while also keeping the whole university allocation exactly on target.
Iris: I think what’s really striking is that this isn't just about finding a slightly better way to count things. It’s about proving that under those specific rules—fractional entitlements, indivisible units at two levels, and tracking past allocations—there simply isn't a deterministic roster solution that works perfectly.
Ruth: That’s the big implication for policy makers trying to design these systems in general. It suggests that the friction we see in real-world cases isn't just bad administration; it’s built into the mathematical structure of how you set up the quotas and how you track them over time.
Warren: I think they offer a path forward, though, by proposing this random roster solution. They say if you trade perfect exact aggregate compliance for something called ex ante proportionality—meaning the allocation is fair based on expectations before anything happens—you can keep monotonicity intact and control the resulting errors.
Iris: And that construction relies on using specific tools from constraint structures to ensure those potential errors don't just pile up randomly, but concentrate in a predictable way at the university level. It’s a very sophisticated way to manage uncertainty without sacrificing department-level accuracy entirely.
Ruth: So, the authors aren't just saying "this policy is hard"; they’re giving us a tool—the random roster—that works as a principled second best if you can live with managing those aggregate errors in a controlled manner. It shifts the focus from impossible perfection to manageable approximation.
Warren: Right. It moves the goalposts from trying to solve an impossible equation perfectly, to finding a strategy that maximizes fairness while minimizing the inevitable deviations across different organizational levels.
Iris: And what this means for identification is that if we look at any system with these features, we have a clear way to test whether a proposed solution is actually better than others by checking how those aggregate errors behave under variation.
Ruth: The next thing we need to dig into is exactly how this random construction works in practice, and what the specific mathematical bounds are on those concentration properties they mention.
More episodes
- Daily Summary for 2026-11-19
- 2608.21274-Recommendation Quality and the Concentration of Consumption: Experimental Evidence from Netflix
- 2408.02757-A nonparametric test for diurnal variation in spot correlation processes
- 2306.16393-High-Dimensional Canonical Correlation Analysis
- 2605.17299-Geometric Brownian motion with intermittent entries and exits
- 2607.04753-Fooling Yourself: how narratives shape beliefs
- 2408.14671-Inference on Treatment Effects with Repeatedly Mismeasured High-Dimensional Controls
- 2201.06898-Difference-in-Differences Estimators for Treatments Continuously Distributed at Every Period
- 2501.15307-Influence Functions: A Unified Differentiation-Based Framework
- The day in conversation for 2026-11-19