The Cherry-Picking Trap
Listen
Radio episode about this paper
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: "The Cherry-Picking Trap".
Warren: Here's the gist of the paper. The model examines how a better-informed proposer's strategic selection of a policy from a "basket" of viable options can undermine trust with an uninformed responder, even when interests are generally aligned.
Ruth: The core mechanism revolves around the proposer's choice between two strategies: "cherry-picking" (selecting their favorite policy overall) or "moderation" (selecting among Pareto improvements).
Warren: When the responder is informed with probability lambda in (0, 1), they approve mutually beneficial policies and reject harmful ones.
Ruth: When uninformed, they reason about the proposer's selection process but cannot verify if a proposal is a "lemon" or a "peach." In equilibrium, the proposer often moderates, yet Pareto improvements are rationally rejected with high probability.
Title and authors: Iris: So, to unpack the title, "The Cherry-Picking Trap," it suggests a kind of logical bind for the proposer. It’s not just picking one good thing; it implies a trap where the very act of being well-informed leads to a situation where you are more likely to select something that looks good only to you, even when there are objectively better options available.
Ruth: That's right. The paper points out that when the proposer has access to a larger pool of viable policies, they gain this strategic flexibility, but that flexibility is also what increases the chance they’ll pick one specifically designed to favor their own interests over the responder's best interest.
Warren: It’s like if you have ten good investments and you pick one because it looks shiny, but the other nine are objectively better for your overall portfolio. The trap is that by picking that shiny one, you signal something negative to the person watching, even though you knew the whole set was strong.
Iris: The paper models how this inference—the responder inferring a selection bias—can undermine trust even when their underlying interests are actually broadly aligned with the proposer’s goals. This is key because it shows that misalignment of perception can be as damaging as actual disagreement on outcomes.
Ruth: Right, so it's not about whether the policy itself is good or bad in isolation, but about the mechanism of selection creating a suspicion that breeds distrust. The paper says this happens when flexibility breeds distrust.
Warren: And this leads into the first friction they identify: the proposer knows which policies are viable and can choose strategically, which increases their scope to cherry-pick one. That’s a huge cost to running the process, because it makes you more tempting for that single policy selection.
Iris: But then there's the second friction: if the responder can't evaluate the proposal well, they have to bet on the proposer selecting something Pareto improvement over others. If that trust becomes untenable because of their lack of expertise or opacity, that’s where you run into trouble >
The paper's summary: Ruth: So, summarizing what this paper does is that it moves beyond just saying "cherry-picking is bad." It actually models the dynamic between the proposer and the responder when we introduce a probability for how informed the responder is. They look at how much better things get if everyone knows each other’s level of knowledge.
Warren: The key finding here, which they highlight, is that both players benefit if the responder has a higher probability of being informed about policy quality. If the responder is more likely to see a Pareto improvement, they are more likely to approve it because they can verify it.
Iris: It seems like the model shows that in equilibrium, even when there are plenty of Pareto improvements available—which happens almost always in large baskets—the proposer still tends to propose those improvements with a probability close to one. But the responder is rationally rejecting those good options more often than they should be.
Ruth: That’s the productivity paradox they find interesting, right? The very act of trying to find better policies makes the proposer more focused on selecting something specific rather than just proposing whatever is available that improves things for everyone.
Warren: And this leads to a real consequence: greater productivity in identifying viable policies can actually backfire because it makes waiting for a more favorable policy seem like a bigger opportunity for the proposer, which then drives the responder to trust them less. It’s counterintuitive.
Iris: The paper identifies that this dynamic pushes behavior toward what they call zero-sum thinking in politics, where strategic selection causes the responder to treat a proposal with prejudice even when mutual benefits are clearly present >
The paper's improvements: Ruth: Now for the improvements they suggest. They point out that the current model assumes a fairly static setup, but it opens up possibilities for how we can make this dynamic work better, especially in terms of commitment strategies. They suggest that the proposer might be better off committing to proposing only those policies that are Pareto improvements, or perhaps giving up their ability to select anything at all.
Warren: That’s a commitment problem they highlight. The paper suggests the proposer would be logically happier if they were committed to only proposing improvements, or if they just stopped trying to pick things entirely. It makes sense because the flexibility is what creates the problem for trust.
Iris: They also look at how this works with different levels of information probability lambda. If we increase that probability that the responder is informed, both players actually benefit more from the system working correctly, but they still find conditions where trust becomes untenable >
Ruth: And looking at search and initiative, they found some interesting trade-offs. They suggest that greater scrutiny from a principal—the proposer—can actually attract more initiative from the proposer if they expect their proposals to be trusted.
Warren: That’s interesting because we often think more scrutiny just drives people away when they aren't sure the system is fair. Here, it’s framed as a way to reward initiative, but only if that initiative leads to proposals that gain trust rather than getting rejected outright >
Iris: They also discuss menu design. It suggests offering a menu of options partially delegates choice to someone who might be worse informed, which helps mitigate adverse selection because the proposer isn't picking secretly. However, they warn that if the menu is too narrow compared to all viable policies, it can backfire by lowering the acceptance rate >
Conclusion: Ruth: So wrapping up on "The Cherry-Picking Trap," the main implication for us is that strategic selection can cause the responder to treat a proposal with prejudice even when mutual benefits are likely. It’s a zero-sum game driven by how we select options from a larger pool, and this finding connects to some of the rational underpinnings we see in populist rhetoric today.
Warren: The core message is that despite abundant opportunities for mutual gain, the cherry-picking trap pushes behavior toward a zero-sum game because flexibility breeds distrust. It’s a friction point in how we make decisions about policies.
Iris: I think the most important thing is that this work shows we need to stop evaluating proposals in isolation and start reasoning from the fact that the proposer strategically selected that proposal from a whole set of feasible alternatives >
Ruth: Exactly. And for future work, they suggest exploring commitment strategies and how menu design can be optimized to balance offering mutually beneficial options against the risk of being perceived as cherry-picking too much.
Warren: It seems like the next step is figuring out exactly how much flexibility is optimal before that trust breaks down under pressure >
Iris: That’s what they are looking at next, exploring those commitment strategies and how to design menus that manage that trade-off better.
S. Nageeb Ali, Maximilian Mihm, Lucas Siga
Department of Economics, Penn State · Division of Social Science, NYU-Abu Dhabi · Department of Economics, University of Essex
econ.TH
Submitted: 2026-11-11
Updated: 2026-11-11
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 84/100
The gist: The model examines how a better-informed proposer's strategic selection of a policy from a "basket" of viable options can undermine trust with an uninformed responder…
Key concepts
- Cherry-Picking Trap
- This occurs when a proposer selects their single favorite policy from a large set of viable options instead of offering improvements. This strategy signals to an uninformed responder that the proposer is only interested in one specific outcome, leading the responder to reject proposals even if they are mutually beneficial.
- Informed Responder Probability ($\lambda$)
- This represents the chance that a policy responder knows whether a proposal is a 'lemon' (bad) or a 'peach' (good). When $\lambda$ is high, responders approve mutually beneficial policies and reject harmful ones, making the proposer more cautious about their selection.
- Commitment Problem
- This describes the proposer's dilemma: whether to commit to proposing only Pareto improvements (mutually better options) or commit to selecting a single policy regardless of quality. The trap suggests committing to cherry-picking is strategically worse for the proposer in the long run.
- Productivity Paradox
- This paradox shows that while being highly productive at finding good policies seems beneficial, it can backfire. If proposers focus too much on identifying viable options, they become more likely to engage in cherry-picking, which ironically makes the responder trust them less.
Terminology
Summary
The model examines how a better-informed proposer's strategic selection of a policy from a basket
of viable options can undermine trust with an uninformed responder, even when interests are generally aligned. The core mechanism revolves around the proposer's choice between two strategies: cherry-picking
(selecting their favorite policy overall) or moderation
(selecting among Pareto improvements).
When the responder is informed with probability lambda in (0, 1), they approve mutually beneficial policies and reject harmful ones. p. 3 If the responder commits, then as in Armstrong and
Vickers (2010), she will reject some mutually beneficial policies so as to motivate the proposer
to select policies that are even better for her. But her set of approved policies might depart from
her favorite ones to curb the proposer’s temptation to cherry-pick.Where the paper says it
lemon
or a peach.
In equilibrium, the proposer often moderates, yet Pareto improvements are rationally rejected with high probability.
The analysis reveals that the cherry-picking trap is fundamentally a commitment problem for the proposer. p. 3 In essence, the cherry-picking trap is a commitment problem. The proposer would often be
better off if he could commit to proposing only Pareto improvements, or if he could relinquish
his ability to select proposals altogether. In essence, the cherry-picking trap is a commitment problem. The proposer would often be
better off if he could commit to proposing only Pareto improvements, or if he could relinquish
his ability to select proposals altogether. In that sense, his ability to cherry-pick ends up
being self-defeating. Both players can also be better off if the responder is informed with
higher probability.Where the paper says it
Where the paper says it
Where the paper says it
In large baskets of policies, the proposer nearly always proposes Pareto improvements with a probability converging to one, yet these are often rejected by the responder. p. 3 While the cherry-picking trap emerges even when the basket has only a few policies, it is
particularly potent in “large” baskets. Here, the proposer nearly always moderates, putting
forward Pareto improvements with probability converging to one, and yet the responder rejects
those proposals with significant probability. In an optimal stopping variant of our problem, we also
document a productivity paradox : the parties may be worse off when the proposer is better at
identifying viable policies because the responder then fears cherry-picking to a greater degree. 2
In turn, her randomization steers the proposer toward the equilibrium mix of proposals but does not make
him indifferent; for generic baskets, he has a strict best response.Where the paper says it
Where the paper says it
The study also identifies implications for search and initiative. p. 3 If the proposer commits, he
offers a lottery over policies that guarantees that the uninformed responder accepts; his optimal
commitment corresponds to a concavified payoff function. We also evaluate the implications of requiring the proposer to offer a menu of options, not
just a single one, to the responder. Were she to always reject, the proposer would always
moderate (because cherry-picking would offer no reward), in which case she would do better
to approve. Thus, in equilibrium, she must mix, and the proposer’s distribution of proposals
1An informed responder faces no dilemma, as she can judge a proposal on its merits. We show that by delegating some of the choice to the
uninformed responder in this way—a form of inexpertise-based delegation—both parties can
benefit if the menu is sufficiently broad relative to the basket. However, a narrow menu can
backfire, leading to lower acceptance odds than the single-proposal benchmark.Where the paper says it
Where the paper says it
Where the paper says it
The model demonstrates zero-sum thinking in politics: strategic selection can cause the responder to treat a proposal with prejudice and reject it even when mutual benefits are likely. This finding connects to rational underpinnings of populist rhetoric, where voters may be wary of policies others support. The overall conclusion is that despite abundant opportunities for mutual gain, the cherry-picking trap pushes behavior toward a zero-sum game.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Precision in Policy Evaluation: The AI system can be improved to evaluate policy proposals by reasoning from a broader set of viable alternatives rather than assessing them in isolation, specifically by asking
why it was chosen
from a larger basket of options. -
Trust-Aware Decision Making: The system can be designed to model the responder's rational rejection based on the fear that a proposal has been cherry-picked, leading to better alignment when interests are broadly aligned but not perfectly so.
-
Commitment Strategy Optimization: The AI can learn optimal commitment strategies for both players, determining whether to commit to proposing only Pareto improvements or selecting a single policy uniformly at random based on the probability of being informed.
-
Menu Design and Delegation: The system can dynamically adjust the proposed options from a menu to maximize acceptance rates by balancing the benefit of offering mutually beneficial options against the risk of being perceived as cherry-picking too many policies.
-
Effort Allocation in Search: When identifying viable policies is costly, the AI can optimize effort allocation across multiple divisions by concentrating attention on those where productivity yields higher returns, based on whether the responder is informed or uninformed.
-
Risk Aversion Modeling: The system can incorporate risk attitudes into its decision-making process, learning that a more risk loving proposer will be tempered by a more risk averse responder, leading to better strategic outcomes in certain equilibrium states.
Abstract
When a better-informed proposer puts forward a policy proposal, the responder should not take the proposal at face value but ask why it was chosen. We model how this inference can undermine trust even when interests are broadly aligned. A proposer privately observes a "basket" of viable policies and proposes one; a responder decides whether to approve the proposal but may lack the expertise to evaluate it. A larger basket makes a Pareto improvement more likely to be available but also gives the proposer greater scope to cherry-pick a policy that favors him at the responder's expense. In the unique equilibrium, Pareto improvements are often proposed and yet rationally rejected. Flexibility that should enable agreement instead breeds distrust.