Who Is Prioritized? Current Need versus Predicted Gains in China's Dibao Allocation: Evidence from a Revealed-Allocation Model

arXiv:2609.08411 · econ.EM · Submitted 2026-09-08 · Read on arXiv

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: "Who Is Prioritized? Current Need versus Predicted Gains in China's Dibao Allocation".

Warren: Here's the gist of the paper. When social assistance is scarce, should it prioritize households in greatest need or those expected to benefit most?

Ruth: <ref:2609.08411#pg2> The paper distinguishes allocation principles by combining predicted policy gains with entry into China’s Minimum Living Standard Guarantee (Dibao).

Warren: <ref:2609.08411#pg2> Predicted gains explain little of allocation: a Shapley decomposition attributes 96.9% of the improvement in allocation fit to household priorities and 3.1% to predicted gains.

Ruth: <ref:2609.08411#pg4> Lower income, lower education of the household head, and elderly presence consistently predict higher priority across supported outcome-value specifications.

Title and authors: Ruth: So, this paper is looking at the China Minimum Living Standard Guarantee program, the Dibao, and asking a really fundamental question about who gets help when there isn't enough money for everyone: should they prioritize those in greatest need or those who seem like they'll benefit most from it?

Warren: Exactly. This study tries to untangle whether the actual allocation logic is driven by what households are currently deprived of, or by what the model predicts they will gain if they receive assistance.

Iris: It seems like the core tension here is between targeting deprivation and targeting expected impact, which we've seen before when thinking about how scarce transfers should be distributed <ref:2609.08411#pg2>.

Ruth: Right. This paper tackles that tension head-on by combining predicted policy gains with entry into the Dibao program to see how they align with observed allocation principles, which is what this paper, "Who Is Prioritized? Current Need versus Predicted Gains in China's Dibao Allocation: Evidence from a Revealed-Allocation Model," does.

Warren: The main finding they show is that those predicted gains explain very little of the actual allocation pattern; it's almost entirely driven by household priorities <ref:2609.08411#pg4>.

Iris: That’s what caught my eye. They use a Shapley decomposition, and they find that ninety-six point nine percent of the improvement in how well the allocation fits household needs comes from those household priorities, while only three point one percent comes from those predicted policy gains <ref:2609.08411#pg4>.

Ruth: That’s a big number for prioritizing need over prediction, isn't it? What does this mean for how we think about social assistance systems generally?

Warren: It means that when you look at the observed allocation, the structure is much more closely tied to household circumstances related to poverty and vulnerability than it is to the predicted consumption or education gains <ref:2609.08411#pg4>.

Iris: And they find very stable patterns for what drives that priority: poorer households, households with less-educated heads, and those with elderly members consistently show higher conditional allocation priority across all the outcome-value specifications they tested <ref:2609.08411#pg19>.

Ruth: So the paper suggests that when you're making allocation decisions based on this data, focusing on need—like income level or education of the head—is a much better fit for what actually happens than just focusing on how much benefit someone will get from the program.

Title and authors: Warren: Right. And they show that this difference is pretty stark in practice with their counterfactual results; if you held the local capacity fixed and made household priorities equal to one, you only got about ten point nine percent overlap with actual recipients, meaning eighty-nine point one percent of people would churn under the baseline specification <ref:2609.08411#pg21>.

Iris: That churn figure is telling us that changing the targeting objective doesn't just tweak a formula; it selects a substantially different group of beneficiaries even if the number of slots stays exactly the same <ref:2609.08411#pg23>.

Ruth: So to wrap up, this paper argues that there's a real difference between targeting households according to their need versus targeting them according to predicted impact, and that treating predicted treatment effects as the primary way to implement existing objectives isn't actually a simple technical fix for the problem <ref:2609.08411#pg24>.

Warren: The conclusion they draw is that conditional on the outcomes and information they observed, entry into the Dibao program is much more related to lower income, lower education of the household head, and the presence of elderly members than to predicted consumption and education gains <ref:2609.08411#pg26>.

Iris: And a bit more on that structure for those priorities, they found that these central patterns hold up across different welfare specifications; for instance, cross-period validation shows household priorities accounting for eighty-one point eight percent of the improvement in allocation fit in the CFPS dataset <ref:2609.08411#pg26>.

Ruth: This suggests that what we're seeing here isn't just a statistical artifact; it points to a fundamental distinction between what a household is actually facing and what they are expected to gain from the program <ref:2609.08411#pg24>.

Warren: So if you were designing an allocation system for something like this, the paper implies that you should be thinking about how need affects priority rather than just what the predicted treatment effect is, because they are two separate allocation questions <ref:2609.08411#pg26>.

Iris: And one limitation they point out is that these estimated household priorities should be interpreted as a revealed conditional allocation priority rather than something you can directly convert into a government welfare weight, since they might reflect information unavailable in the local administrative processes <ref:2609.08411#pg25>.

Ruth: That’s an important caveat because it tells us that even if we get these priorities right from the data, applying them directly to a formula might miss some of the real-world complexities and information flows involved <ref:2609.08411#pg25>.

Title and authors: Warren: So, to sum up this paper on "Who Is Prioritized? Current Need versus Predicted Gains in China's Dibao Allocation: Evidence from a Revealed-Allocation Model," it shows that household priorities, driven by things like lower income and elderly presence, dominate the allocation fit far more than predicted policy gains do <ref:2609.08411#pg4>.

Iris: And the counterfactual analysis with that eighty-nine point one percent churn is a strong demonstration of how much different those two approaches—need versus gain—actually are in terms of who gets selected <ref:2609.08411#pg21>.

Ruth: It really highlights that moving toward impact-oriented targeting isn't just about tweaking the math; it means picking a completely different set of people to be targeted, which is a significant policy choice <ref:2609.08411#pg22>.

Warren: So, for anyone working on allocation systems, this paper suggests separating the prediction of who will benefit most from the actual revealed priorities of who should receive priority based on need <ref:2609.08411#pg26>.

Iris: And their finding that these core priority patterns are robust across various welfare specifications gives us a solid base to expect those need-based indicators to be stable even when we change some of the other model parameters <ref:2609.08411#pg26>.

Ruth: It’s interesting how they show that their results hold up across cross-period and cross-dataset validations, with household priorities accounting for ninety-six point nine percent of the improvement in allocation fit in CHFS, eighty-five point nine percent in CHARLS, and eighty-one point eight percent in CFPS <ref:2609.08411#pg26>.

Warren: That consistency across different datasets reinforces the idea that these underlying patterns of need-based priority are quite structural within the observed allocation logic <ref:2609.08411#pg7>.

Iris: And while they found this structure, they did acknowledge that their best-fitting outcome value parameters, like for consumption and education gains, were relatively flat and didn't establish a very robust cardinal ordering between those two types of gains <ref:2609.08411#pg20>.

Ruth: That’s a good point because it contrasts with the strong stability they found in the priority directions themselves, showing that while the *magnitude* of gains might be fuzzy, the *order* in which households are prioritized remains very clear <ref:2609.08411#pg26>.

Warren: So, to finish up on this paper about "Who Is Prioritized? Current Need versus Predicted Gains in China's Dibao Allocation: Evidence from a Revealed-Allocation Model," the main message is that need-based targeting is distinct from impact-based targeting, and they provide strong evidence that poverty and vulnerability affect priority independently of predicted gains <ref:2609.08411#pg24>.

Title and authors: Iris: And the implication for policy is clear: you can't just use predicted treatment effects as a simple way to make the existing objective more accurate; it fundamentally changes the distributive criterion used to select recipients <ref:2609.08411#pg24>.

Ruth: They suggest that while modeling these household priorities is useful for describing what's observed, we shouldn't directly convert them into a targeting formula because they might capture information not present in the standard administrative processes <ref:2609.08411#pg25>.

Warren: So, this paper gives us a way to see the tension between those two ideas—need versus gain—and it shows that for this specific program, need seems to be the stronger driver of who actually gets allocated resources <ref:2609.08411#pg4>.

Iris: And they leave us with a clear picture of what drives that selection process, even when we try to account for what households are expected to gain from the assistance <ref:2609.08411#pg3>.

Ruth: We've talked about how the paper distinguishes between those two components, and it seems like this separation is key to understanding why the allocation looks the way it does <ref:2609.08411#pg7>.

Warren: It’s a good framework for thinking about how we might design better systems, by making sure we're not just optimizing for a predicted outcome but are also respecting the actual priorities revealed by who actually receives help <ref:2609.08411#pg8>.

Iris: And remembering that the counterfactual results show how much difference exists when you shift from one ranking system to another, because that churn of eighty-nine point one percent is substantial even with a fixed number of slots <ref:2609.08411#pg21>.

Ruth: So, this study on "Who Is Prioritized? Current Need versus Predicted Gains in China's Dibao Allocation: Evidence from a Revealed-Allocation Model" concludes that the observed allocation is much more strongly related to household need than to predicted policy gains <ref:2609.08411#pg26>.

Warren: It’s a nuanced finding that tells us we need to be careful not to replace existing rules just because they are based on predictions <ref:2609.08411#pg24>.

Iris: And the stability across different specifications suggests that these underlying structures of household priority aren't easily broken by minor changes in how we model the welfare itself <ref:2609.08411#pg26>.

Ruth: It’s a solid piece of evidence showing that we need to prioritize understanding what drives the observed allocation structure rather than just focusing on what the model *thinks* households want <ref:2609.08411#pg25>.

Warren: That separation between predicted gains and revealed priorities is something that needs to be kept in mind when thinking about how we model these kinds of social assistance programs <ref:2609.08411#pg9>.

The paper's summary: Ruth: So, to quickly recap, this paper is looking at how allocation decisions are made in China's Dibao program and comparing what households actually need versus what the program predicts they will gain from it.

Warren: Yeah, and the main takeaway is that when you look at actual who gets help in those local markets, the factors driving priority are way more about current need than about predicted policy gains.

Iris: Exactly. The Shapley decomposition shows that ninety-six point nine percent of how much the allocation fits household needs comes from household priorities, while only three point one percent comes from predicted gains.

Ruth: That's a huge split there. It means that when you're trying to figure out who gets help, focusing on things like lower income or less education for the head is way more important than just looking at what the model predicts they will consume or learn.

Warren: Right. And I think that separation of predicted gains from revealed priorities is a key thing for anyone modeling these systems, because it shows that you can't just use predicted treatment effects as a simple way to fix an allocation rule and expect it to work.

Iris: I agree with Warren there. The paper suggests that if you're targeting based on what the model predicts they will gain, you are essentially choosing a different set of beneficiaries than if you target them based on observed poverty and vulnerability indicators.

Ruth: That's the core argument, isn't it? It implies that two different targeting objectives lead to substantially different people actually receiving the assistance.

Warren: And that difference is really quantified in their counterfactual analysis. They set household priorities equal to one, holding local capacity fixed, and you only get about ten point nine percent overlap with actual recipients, which means eighty-nine point one percent of people would churn under the baseline rules.

Iris: That churn figure shows that moving from a gain-oriented recipient ranking to a need-oriented one is not just a technical adjustment; it changes the selection basis completely.

Ruth: It's interesting because they also found very stable patterns for those priorities across all their tests, even when they changed some of the other welfare parameters in the model.

Warren: That stability means that those core drivers—poorer households, less-educated heads, elderly members—are robust and don't disappear if you change how you measure consumption or education gains.

Iris: But one thing they flag is that these estimated household priorities should be read as a revealed conditional priority, not something you can just plug directly into a government targeting formula because it might capture information unavailable in the standard administrative processes.

Ruth: That's a fair point, Iris. It means the model helps us describe what’s observed, but it doesn't perfectly map onto how the actual bureaucracy works.

Warren: So the paper sets up this distinction between targeting predicted treatment effects and prioritizing poverty and vulnerability as two separate allocation questions that need to be answered distinctly.

Iris: And they suggest that for policy implementation, you can't just use predicted treatment effects as a proxy for getting the existing objective right because it fundamentally changes who is selected.

Ruth: It really makes you think about how we design these systems, knowing that there's this deeper structure of need driving the allocation than we might initially expect.

Warren: And if you want to figure out what actually drives that need-based priority in more detail, maybe we should look at how those household priorities are being revealed in the data itself.

The paper's improvements: Ruth: So, to wrap up this part, the authors aren't just describing what they found; they're suggesting ways to use this kind of modeling better for real policy decisions in social assistance systems.

Warren: Right. They look at how you can improve allocation systems by making that distinction between distributional targeting and impact targeting more rigorous through modeling household priorities conditional on predicted policy gains.

Iris: They are proposing a way to improve allocation methods by using a Shapley decomposition to separate those two components, showing ninety-six point nine percent of the fit comes from household priorities versus only three point one percent from predicted gains.

Ruth: That’s like they're saying we should use that decomposition framework so we can clearly see which part of the allocation puzzle is driven by what people actually need and which part is just theoretical prediction.

Warren: Yeah, and this separation lets us change how allocation rules are made based on whether we are targeting predicted treatment effects or prioritizing poverty and vulnerability.

Iris: The implication here is that if we want to improve existing systems, we need to focus on separating the predicted gains from the revealed priorities so allocation rules can be changed based on that distinction.

Ruth: So it's about having a model structure where you can analyze both sides—the prediction and the observed need—to see what truly drives who gets resources.

Warren: And this modeling approach is important because it shows that treating predicted treatment effects as just a technical way to implement an existing objective isn't actually a simple fix for the problem, because it changes the selection basis.

Iris: Exactly. The authors suggest that instead of just tweaking numbers in a formula, we need to acknowledge that moving toward impact-oriented targeting means selecting a substantially different set of beneficiaries.

Ruth: That’s the shift from simply predicting who will benefit most to actually prioritizing who is in greatest need, and that's a big policy decision.

Warren: And they caution that these estimated household priorities should be interpreted as revealed conditional priorities, not just weights you can directly convert into a targeting formula because they might reflect information unavailable in the local administrative processes.

Iris: That’s the caveat; so we use the model to describe what's observed, but we have to be careful not to assume it perfectly reflects every local process.

Ruth: So, this paper is essentially saying that improving allocation systems requires a deeper look at how need and prediction interact in a way that respects both.

Warren: And they want us to keep that separation between predicted gains and revealed priorities in mind whenever we think about designing any kind of social assistance mechanism.

Iris: They are also showing robustness here, meaning these findings hold up across different welfare specifications, which gives us confidence that the core priority patterns are stable even when we change some of the other model parameters.

Ruth: It’s good to know that the structure itself is solid, even if the precise measurements might get a little fuzzy depending on what you choose to measure.

Warren: And this leads us right into where things get tricky—how do we actually operationalize these revealed priorities in a way that doesn't just rely on assumptions about what households are thinking?

Conclusion: Ruth: So we've looked at how this paper, "Who Is Prioritized? Current Need versus Predicted Gains in China's Dibao Allocation: Evidence from a Revealed-Allocation Model," suggests that observed allocation is much more closely related to household need than it is to predicted policy gains.

Warren: Yeah, the big idea is that when you’re designing these systems, you can’t just rely on what the model predicts people will gain; you have to account for what they actually need and how they are revealed in the allocation process.

Iris: I just want to stress that this separation between predicted gains and revealed priorities is important because it shows that if we target based on predictions, we’re effectively choosing a different group of beneficiaries than if we target based on observed poverty indicators.

Ruth: Right, so the implication for policy is that you can't use predicted treatment effects as just a technical way to implement an existing objective; it fundamentally changes the distributive criterion used to select recipients.

Warren: And they show that this difference manifests in real churn numbers, like eighty-nine point one percent of people would churn if you shifted the targeting objective from gain-oriented to need-oriented under their baseline specification.

Iris: That churn figure is a strong demonstration of how much different those two approaches are in terms of who gets selected even when the number of program slots stays fixed.

Ruth: It's interesting because they also found that these core priority patterns, like lower income and elderly presence, are quite stable across all the different welfare specifications they tested.

Warren: That stability means that those underlying drivers don't disappear if you change how you measure consumption or education gains; the structure of need is pretty hard to break.

Iris: But one thing they flag is that these estimated household priorities should be read as a revealed conditional priority, not just something you can directly plug into a government targeting formula because it might reflect information unavailable in the local administrative processes.

Ruth: That’s the caveat; so the model helps us describe what's observed, but we have to be careful not to assume it perfectly reflects every local process or every piece of information.

Warren: So, this paper sets up this distinction between targeting predicted treatment effects and prioritizing poverty and vulnerability as two separate allocation questions that need to be answered distinctly.

Iris: And they suggest that for policy implementation, you can't just use predicted treatment effects as a proxy for getting the existing objective right because it fundamentally changes who is selected.

Ruth: It really makes you think about how we design these systems, knowing that there's this deeper structure of need driving the allocation than we might initially expect.

Warren: And if you want to figure out what actually drives that need-based priority in more detail, maybe we should look at how those household priorities are being revealed in the data itself.

Department of Economics, University of California, Riverside

econ.EM

Submitted: 2026-09-08

Updated: 2026-11-13

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: When social assistance is scarce, should it prioritize households in greatest need or those expected to benefit most? <ref:2609.08411#pg2> The paper distinguishes allocation principles by combining…

Key concepts

Allocation Components
The paper distinguishes between two types of allocation: the welfare gain households are predicted to get from the program and their relative priority assigned based on those gains. This separation is key to understanding how Dibao recipients are selected.
Shapley Decomposition
This statistical tool was used to measure which factor explains observed allocation fit. It revealed that household priorities account for 96.9% of the improvement in fit, while predicted policy gains only account for a small 3.1%, indicating priorities are much more influential.
Household Priority Patterns
The analysis identified stable patterns in who receives priority: poorer households, those with less-educated heads, and elderly members consistently receive higher conditional priority across most models. These factors reflect vulnerability rather than just predicted economic outcomes.

Terminology

Summary

When social assistance is scarce, should it prioritize households in greatest need or those expected to benefit most? p. 0

Where the paper says it

Policy Gains or Household Need? The Allocation Logic of China’s Dibao Program Haojie Liu, Jiyuan Ling, and Zihan Lin1 Abstract When social assistance is scarce, should it prioritize households in greatest need or those expected to benefit most?

<ref:2609.08411#pg2> The paper distinguishes allocation principles by combining predicted policy gains with entry into China’s Minimum Living Standard Guarantee (Dibao). p. 0
Where the paper says it

The Allocation Logic of China’s Dibao Program Haojie Liu, Jiyuan Ling, and Zihan Lin1 Abstract When social assistance is scarce, should it prioritize households in greatest need or those expected to benefit most? Using panel data from the China Household Finance Survey, we distinguish allocation principles by combining predicted policy gains with entry into China’s Minimum Living Standard Guarantee (Dibao).

<ref:2609.08411#pg2> Predicted gains explain little of allocation: a Shapley decomposition attributes 96. p. 0
Where the paper says it

We estimate heterogeneous predicted gains in consumption and education and then recover the conditional priorities revealed by recipient selection. Predicted gains explain little of allocation: a Shapley decomposition attributes 96.

9% of the improvement in allocation fit to household priorities and 3. p. 0
Where the paper says it

Predicted gains explain little of allocation: a Shapley decomposition attributes 96. 9% of the improvement in allocation fit to household priorities and 3.

1% to predicted gains. p. 0
Where the paper says it

The Allocation Logic of China’s Dibao Program Haojie Liu, Jiyuan Ling, and Zihan Lin1 Abstract When social assistance is scarce, should it prioritize households in greatest need or those expected to benefit most? Using panel data from the China Household Finance Survey, we distinguish allocation principles by combining predicted policy gains with entry into China’s Minimum Living Standard Guarantee (Dibao).

p. 0
Where the paper says it

The Allocation Logic of China’s Dibao Program Haojie Liu, Jiyuan Ling, and Zihan Lin1 Abstract When social assistance is scarce, should it prioritize households in greatest need or those expected to benefit most? Using panel data from the China Household Finance Survey, we distinguish allocation principles by combining predicted policy gains with entry into China’s Minimum Living Standard Guarantee (Dibao).

<ref:2609.08411#pg4> Lower income, lower education of the household head, and elderly presence consistently predict higher priority across supported outcome-value specifications. p. 0
Where the paper says it

1% to predicted gains. Lower income, lower education of the household head, and elderly presence consistently predict higher priority across supported outcome-value specifications.

<ref:2609.08411#pg4> These findings show that Dibao allocation is more closely aligned with household circumstances associated with poverty and vulnerability than with policy gains predictable from the outcomes and information observed in our data. p. 0
Where the paper says it

1% recipient churn. These findings show that Dibao allocation is more closely aligned with household circumstances associated with poverty and vulnerability than with policy gains predictable from the outcomes and information observed in our data, underscoring the distinction between distributional and impact targeting in social assistance.

<ref:2609.08411#pg4>

Conceptual Framework of Allocation Components

The analysis distinguishes between two conceptually different components of observed allocation: the gains households are predicted to obtain from the program and the relative priority assigned to households conditional on those gains. <ref:2609.08411#pg7> This distinction is central to our analysis. p. 5

Where the paper says it

Conceptual Framework Predicted Policy Gains and Allocation Priorities Following Björkegren, Blumenstock, and Knight (2026), we distinguish two conceptually different components of observed allocation: the gains households are predicted to obtain from the program and the relative priority assigned to households conditional on those gains. This distinction is central to our analysis.

<ref:2609.08411#pg8> The model-implied welfare gain from allocating Dibao to household i is the difference between its contribution to welfare under receipt and non-receipt: <ref:2609. p. 6
Where the paper says it

ui(Ti) = ∑ j bijvij (Ti) + aiTi , where vij (Ti) denotes household i’s value of welfare-relevant outcome j under treatment status Ti , bij captures the value attached to that outcome, and ai captures components of program value not represented by the modeled outcomes. The model-implied welfare gain from allocating Dibao to household i is the difference between its contribution to welfare under receipt and non-receipt: ∆Si = Si(1) − Si(0).

08411#pg8> Substituting the household welfare representation gives <ref:2609. p. 6
Where the paper says it

The model-implied welfare gain from allocating Dibao to household i is the difference between its contribution to welfare under receipt and non-receipt: ∆Si = Si(1) − Si(0). Substituting the household welfare representation gives ∆Si = w(xi)[ui(1) − ui(0)] From the definition of ui(Ti), ui(1) − ui(0) = ai + ∑ j bij [vij (1) − vij (0)] Defining the treatment-induced change in outcome j as ∆vij = vij (1) − vij (0), the model-implied gain from allocating Dibao to household i can therefore be written as 7

08411#pg8> The model-implied gain from allocating Dibao to household i can therefore be written as <ref:2609. p. 6
Where the paper says it

The model-implied welfare gain from allocating Dibao to household i is the difference between its contribution to welfare under receipt and non-receipt: ∆Si = Si(1) − Si(0). Substituting the household welfare representation gives ∆Si = w(xi)[ui(1) − ui(0)] From the definition of ui(Ti), ui(1) − ui(0) = ai + ∑ j bij [vij (1) − vij (0)] Defining the treatment-induced change in outcome j as ∆vij = vij (1) − vij (0), the model-implied gain from allocating Dibao to household i can therefore be written as 7

08411#pg8> This expression separates two margins of allocation. p. 7
Where the paper says it

∆Si = w(xi)(ai + ∑ j bij∆vij ) This expression separates two margins of allocation. The term in parentheses captures the welfare gain generated by assigning Dibao to household i, while w(xi) captures the household’s relative priority conditional on that gain.

<ref:2609.08411#pg9>

Decomposition of Allocation Fit

The empirical decomposition evaluates how much each component contributes to reproducing the observed ranking of Dibao entrants within local allocation markets. <ref:2609.08411#pg9> The first component asks how much of observed allocation can be explained by heterogeneous predicted policy gains. p. 7

Where the paper says it

This distinction provides the basis for our empirical decomposition below. The first component asks how much of observed allocation can be explained by heterogeneous predicted policy gains.

<ref:2609.08411#pg9> The second asks which household characteristics are associated with greater allocation priority after conditioning on those gains. p. 7
Where the paper says it

The first component asks how much of observed allocation can be explained by heterogeneous predicted policy gains. The second asks which household characteristics are associated with greater allocation priority after conditioning on those gains.

<ref:2609.08411#pg9> A Shapley decomposition similarly attributes 96. p. 2
Where the paper says it

667 in the full model. A Shapley decomposition similarly attributes 96.

9% of the improvement in allocation fit to household priorities and only 3. p. 0
Where the paper says it

Predicted gains explain little of allocation: a Shapley decomposition attributes 96. 9% of the improvement in allocation fit to household priorities and 3.

1% to predicted gains. <ref:2609.08411#pg4> Household priorities account for 38. p. 0
Where the paper says it

Using panel data from the China Household Finance Survey, we distinguish allocation principles by combining predicted policy gains with entry into China’s Minimum Living Standard Guarantee (Dibao). We estimate heterogeneous predicted gains in consumption and education and then recover the conditional priorities revealed by recipient selection.

51, or 96.9%, of the improvement in allocation fit, whereas predicted gains account for 1. p. 0
Where the paper says it

Predicted gains explain little of allocation: a Shapley decomposition attributes 96. 9% of the improvement in allocation fit to household priorities and 3.

23, or 3.1%. <ref:2609.08411#pg7>

Identification of Household Priority Patterns

The recovered priorities exhibit three particularly stable patterns: poorer households, households with less-educated heads, and households with elderly members receive higher conditional allocation priority. <ref:2609.08411#pg18> Among the 27 outcome-value specifications satisfying the allocation-compatibility criterion, lower income and lower household-head education are negatively associated with priority in every case, while elderly presence predicts higher priority in 26 of 27. p. 0

Where the paper says it

Lower income, lower education of the household head, and elderly presence consistently predict higher priority across supported outcome-value specifications. Holding local program capacity fixed and removing household-priority differences while retaining the full model’s estimated outcome values, the resulting ranking overlaps with recipients by only 10.

<ref:2609.08411#pg19> Income measures current economic resources, household-head education is associated with longer-run socioeconomic circumstances, and elderly presence captures a distinct dimension of household vulnerability. p. 0
Where the paper says it

1% to predicted gains. Lower income, lower education of the household head, and elderly presence consistently predict higher priority across supported outcome-value specifications.

<ref:2609.08411#pg19>

Counterfactual Allocation Results

The counterfactual allocation illustrates that changing the targeting objective can alter beneficiary incidence substantially even without changing the number or geographic distribution of program slots. <ref:2609.08411#pg21> Holding local capacity fixed, set household priorities equal to one, and retain the outcome-value parameters estimated in the full model, the resulting ranking overlaps with actual recipients by only 10. p. 3

Where the paper says it

We construct a counterfactual that holds the number of Dibao slots fixed within each local market, sets household priorities equal to one, and retains the outcome-value parameters estimated in the full model. The resulting gain-oriented ranking overlaps with actual recipients by only 10.

9%, implying 89. p. 0
Where the paper says it

Holding local program capacity fixed and removing household-priority differences while retaining the full model’s estimated outcome values, the resulting ranking overlaps with recipients by only 10. 9%, implying 89.

1% recipient churn under the baseline specification. p. 3
Where the paper says it

9%, implying 89. 1% recipient churn under the baseline specification.

<ref:2609.08411#pg21> The counterfactual quantifies how different a gain-oriented recipient ranking is from observed Dibao allocation, and this 89. p. 3
Where the paper says it

We construct a counterfactual that holds the number of Dibao slots fixed within each local market, sets household priorities equal to one, and retains the outcome-value parameters estimated in the full model. The resulting gain-oriented ranking overlaps with actual recipients by only 10.

1% churn is not an efficiency or welfare loss but a measure of difference between two recipient rankings. p. 3
Where the paper says it

1% recipient churn under the baseline specification. This exercise does not estimate a welfare or efficiency loss.

<ref:2609.08411#pg21>

Conclusion on Targeting Principles

Our results point to a basic difference between targeting households according to need and targeting them according to predicted impact. <ref:2609.08411#pg24> Under need-based targeting, poverty and vulnerability may affect a household’s priority even when its predicted gains are small. p. 2

Where the paper says it

This distinction is important because targeting on predicted treatment effects is not simply a more accurate version of targeting on deprivation. If poverty and vulnerability affect selection independently of predicted gains, replacing the existing allocation rule with a gain-based rule would change the distributive criterion used to select recipients.

<ref:2609.08411#pg24> Allocating assistance according to predicted treatment effects is not simply a technical way to implement the existing objective more accurately. p. 3
Where the paper says it

The paper contributes to the targeting literature by separating predicted policy gains from the household priorities revealed by recipient selection. It also clarifies the relationship between deprivation targeting and impact targeting: allocating on predicted treatment effects may change who receives social assistance, rather than merely improve the accuracy of an existing targeting rule.

<ref:2609.08411#pg24> The empirical conclusion is narrower: conditional on the outcomes and information observed in our data, Dibao entry is much more closely related to lower income, lower education of the household head, and the presence of elderly members than to predicted consumption and education gains. p. 0
Where the paper says it

1% to predicted gains. Lower income, lower education of the household head, and elderly presence consistently predict higher priority across supported outcome-value specifications.

<ref:2609.08411#pg26> Predicting who is likely to benefit most and deciding who should receive priority are therefore two separate allocation questions. p. 7
Where the paper says it

Two households with similar policy gains may therefore receive different allocation priority because their household characteristics imply different relative weights. Conversely, among households with similar priority characteristics, allocation may favor those predicted to benefit more from Dibao.

<ref:2609.08411#pg26>

Robustness Across Specifications

The main findings are robust to alternative specifications. Adding health to the welfare vector yields 28. p. 9

Where the paper says it

Education expenditure is constructed separately from the CHFS education module as the log of one plus real household spending on education and training, so that educational investment remains distinct from food consumption. Health is not included in the baseline welfare vector because its measurement is less consistent across survey waves.

3% overlap with actual recipients, while a reduced priority specification yields 29. p. 0
Where the paper says it

Lower income, lower education of the household head, and elderly presence consistently predict higher priority across supported outcome-value specifications. Holding local program capacity fixed and removing household-priority differences while retaining the full model’s estimated outcome values, the resulting ranking overlaps with recipients by only 10.

4%, compared with 29.1% in the baseline model. p. 2
Where the paper says it

Our central question is whether systematic differences in household priority remain after accounting for policy gains predictable from the researcher’s information set. We distinguish among households’ baseline circumstances, their predicted responses to assistance, and their allocation priority conditional on those responses.

<ref:2609.08411#pg23> Priority directions are also stable across supported specifications, showing that the central priority patterns do not depend on the baseline outcome set or a particular outcome-value specification. p. 0
Where the paper says it

1% to predicted gains. Lower income, lower education of the household head, and elderly presence consistently predict higher priority across supported outcome-value specifications.

<ref:2609.08411#pg26> Cross-period and cross-dataset validation confirms this stability, with household priorities accounting for 96. p. 8
Where the paper says it

For cross-period validation, W1 uses 2015 characteristics, identifies entry between 2015 and 2017, and measures outcomes in 2019. This structure separates baseline characteristics used to predict gains and allocation priorities from subsequent program entry and postentry outcomes.

9%, 85.9%, and 81.8% of the improvement in allocation fit in CHFS, CHARLS, and CFPS, respectively, compared with predicted gains at 3. p. 0
Where the paper says it

Predicted gains explain little of allocation: a Shapley decomposition attributes 96. 9% of the improvement in allocation fit to household priorities and 3.

1%, 14.1%, and 18.2%.

Implications for Policy Implementation

The findings suggest that moving from observed Dibao allocation toward impact-oriented targeting would represent more than a change in predictive methodology; it would select a substantially different set of beneficiaries. <ref:2609.08411#pg22> This contrast indicates that the existing system gives separate priority to poverty and vulnerability, and replacing that priority with predicted gains changes the basis on which recipients are selected. p. 2

Where the paper says it

This distinction is important because targeting on predicted treatment effects is not simply a more accurate version of targeting on deprivation. If poverty and vulnerability affect selection independently of predicted gains, replacing the existing allocation rule with a gain-based rule would change the distributive criterion used to select recipients.

<ref:2609.08411#pg24> The estimated household priorities should be interpreted as a revealed conditional allocation priority rather than a government welfare weight, as they may reflect information unavailable in the CHFS or local administrative processes. p. 2
Where the paper says it

The interpretation is necessarily limited by the setting: Dibao receipt is observational and binary, implementation is decentralized, and local administrators may observe information unavailable in the CHFS. We therefore refer to the estimated component as a revealed conditional allocation priority, rather than as a normative government welfare weight.

<ref:2609.08411#pg25> This component is useful for describing observed allocation but should not be directly converted into a targeting formula. p. 2
Where the paper says it

We therefore refer to the estimated component as a revealed conditional allocation priority, rather than as a normative government welfare weight. The results show that household-priority information is far more informative about observed allocation than predicted gains.

<ref:2609.08411#pg25>

Weak Identification of Cardinal Values

The best-fitting outcome-value parameters are approximately βC = 2.08, βE = 0.78, but the objective profiles are relatively flat, particularly for consumption. p. 0

Where the paper says it

Using panel data from the China Household Finance Survey, we distinguish allocation principles by combining predicted policy gains with entry into China’s Minimum Living Standard Guarantee (Dibao). We estimate heterogeneous predicted gains in consumption and education and then recover the conditional priorities revealed by recipient selection.

<ref:2609.08411#pg19> Under the 5% objective-gap criterion, support for βC extends to the upper search boundary of 20, while support for βE extends to approximately 5. p. 2
Where the paper says it

We adapt the revealed-policy-preference framework developed by Björkegren, Blumenstock, and Knight (2026) to study these components. Using panel data from the China Household Finance Survey (CHFS), we first estimate heterogeneous predicted gains in consumption and education under selection on observables and common support.

25; zero remains supported for both. p. 9
Where the paper says it

Food consumption is annual food expenditure divided by household size, deflated by the consumer price index, and transformed to logs; first-stage estimates use four-year changes in this measure. Education expenditure is constructed separately from the CHFS education module as the log of one plus real household spending on education and training, so that educational investment remains distinct from food consumption.

<ref:2609.08411#pg20> This indicates that the point estimates do not establish a robust cardinal ordering between consumption and education gains, contrasting with the stability of the household-priority patterns across supported outcome-value specifications. p. 0
Where the paper says it

1% to predicted gains. Lower income, lower education of the household head, and elderly presence consistently predict higher priority across supported outcome-value specifications.

<ref:2609.08411#pg26> The more robust findings concern the structure of observed allocation rather than a single counterfactual assignment for individual households. p. 0
Where the paper says it

1% recipient churn. These findings show that Dibao allocation is more closely aligned with household circumstances associated with poverty and vulnerability than with policy gains predictable from the outcomes and information observed in our data, underscoring the distinction between distributional and impact targeting in social assistance.

<ref:2609.

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

  1. Improve allocation systems by distinguishing between distributional and impact targeting by modeling household priorities conditional on predicted policy gains using a Shapley decomposition which attributes 96.9% of allocation fit to household priorities versus only 3.1% to predicted gains.

  2. Allow AI systems to generate counterfactual allocation rankings by holding local program capacity fixed and setting household priorities equal to one, showing that such a gain-oriented ranking overlaps with actual recipients by only 10.9%, implying 89.1% recipient churn under the baseline specification.

  3. Enable AI systems to identify stable household priority patterns across different welfare specifications, specifically noting that lower income and lower household-head education imply higher priority in all 27 outcome-value specifications, suggesting a robust structure for need-based targeting.

  4. Develop models that separate predicted policy gains from the household priorities revealed by recipient selection so that allocation rules can be changed based on whether they target predicted treatment effects or prioritize poverty and vulnerability.

Abstract

Using the 2017--2021 China Household Finance Survey panel and a sample of 3,102 households that did not receive Dibao in 2017, this paper applies the BBK revealed-allocation model to examine how household economic circumstances and predicted policy gains explain observed Dibao receipt in China. We find that household income is the allocation criterion most difficult to substitute with other factors. When all gain-adjustment terms are positive, constraining the income-rank coefficient to be nonnegative reduces the model's original improvement in fit by approximately 33.6%; by contrast, setting both predicted-gain coefficients for food consumption and education spending to zero still preserves approximately 95.6% of that improvement. These results indicate that observed Dibao allocation places greater emphasis on prioritizing households facing current livelihood hardship than on pursuing predictable short-run improvements in expenditure outcomes. This allocation orientation reflects the trade-off in social assistance between basic-livelihood protection and expected policy gains and offers a new perspective on the potential long-run social-welfare value of basic-livelihood protection.

Related papers