Quantifying Educational Competition: A Game-Theoretic Model with Policy Implications

Siyuan He

REVIEW 3 major objections 4 minor 11 references

Quantifying Educational Competition: A Game-Theoretic Model with Policy Implications

T0 review · 3 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A game-theoretic model of China's education system claims that burden-reduction policies fail because each family benefits from disobeying the rules, producing an unstable Nash equilibrium of escalating study time and falling welfare.

desk verdict The two-family game has no Nash equilibrium, but the underlying policy intuition is real; needs substantial rework. read the letter →

arxiv 2412.10974 v2 pith:JFVCGRL3 submitted 2024-12-14 econ.TH

Siyuan He This is my paper · ORCID

classification econ.TH MSC 91A1091A8091B15

keywords educationarmsraceburdenreductionpolicygametheoryNashequilibriumsignalingsocialwelfareeducationalequityChinacompetition

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that China's decades of 'burden reduction' education policies fail because they ignore the strategic interaction between families. In the model, a family that obeys a study-time cap while others disobey loses access to scarce high-quality educational resources, so each family has an individual incentive to defect. The result is an unstable Nash equilibrium in which all families invest more study time, scores and the cutoff rise, and everyone's utility falls. The paper extends the model with a job-market signaling framework to argue that biased perceptions of the wage premium for elite education, and not just scarce resources, drive the escalation. If the model is right, effective policy must weaken the link between academic performance and future rewards rather than simply cap study time.

What carries the argument

The engine of the model is a utility function $u_i = \log(2 + S_i - S_{\mathrm{cut}}) - P t_i$ for a family whose student's score is $S_i = \gamma_i t_i$, where $\gamma_i$ is aptitude, $t_i$ is study time, $P$ is the time-cost coefficient, and $S_{\mathrm{cut}} = \bar S + k\sigma_S$ is the cutoff for access to high-quality resources. Concavity of the log payoff gives a unique best response $t_i^* = 1/P + S_{\mathrm{cut}} - 2/\gamma_i$ in a large society, showing that a higher cutoff pushes up study time while higher aptitude reduces the effort needed. In the two-family special case the cutoff is the average score, and the best responses are strategic complements: when one family studies longer, the other's best response increases. The feedback loop $S_{\mathrm{cut}} \uparrow \Rightarrow t_i^* \uparrow \Rightarrow \bar t \uparrow \Rightarrow S_{\mathrm{cut}} \uparrow$ is the formal statement of the arms race, and the paper uses simulations to show that a larger score variance raises the cutoff and lowers utility.

What would settle it

Collect exam-score data for a large cohort along with measured study time and an aptitude proxy: if the residual variance in scores after controlling for study time and aptitude is large, the deterministic score equation fails. Alternatively, implement a randomized policy that exogenously lowers the cutoff (for example, lottery-based admission to high-quality schools) and observe whether families reduce study time; the model predicts a substantial drop, and its absence would falsify the feedback loop.

Watch

Extended reading notes

Core claim

The central claim is that the education arms race is a collective-action failure, not a simple oversupply of effort. Each family's study time is a best response to the distribution of others' scores; because success depends on clearing a cutoff that rises with everyone's effort, unilateral restraint is punished. The paper derives best-response functions and shows that in a two-family game the (disobey, disobey) outcome is an unstable Nash equilibrium, so burden-reduction rules collapse under defection. In the general model, the cutoff $S_{\mathrm{cut}}$ and average study time reinforce each other through the loop $S_{\mathrm{cut}} \uparrow \Rightarrow t_i^* \uparrow \Rightarrow \bar t \uparrow \Rightarrow S_{\mathrm{cut}} \uparrow$, reducing each family's maximized utility. A higher variance of student aptitude raises the cutoff and lowers social welfare, creating a trade-off between educational equity and utilitarian welfare, and the paper proposes that reducing the perceived premium of elite education can break the loop.

Load-bearing premise

The model assumes a student's exam score is exactly aptitude multiplied by study time, with no random error; if real scores are noisy, a family cannot know in advance whether extra study will clear the cutoff, which changes every best-response calculation and the arms-race conclusion.

Editorial extensions

If this is right

Any burden-reduction policy that only caps or recommends study time will be undermined because a single family can gain by exceeding the cap, and the paper's two-family game shows both families end up disobeying.
Policies that shrink the pool of competitors, such as the 50-50 vocational diversion, can lower the cutoff and reduce competition but do so by sacrificing the educational equity of students who are diverted.
If the perceived wage premium for elite education is inflated by cognitive bias, reducing that bias through diverse career signals and career support systems should lower equilibrium study time and raise social welfare.
Exam designs that make scores depend more on aptitude and less on accumulated study time weaken the link between effort and success and can reduce the escalation.
An increase in the variance of student aptitude or scores raises the cutoff and lowers the maximized utility of a typical family, so more heterogeneous competition is predicted to be more wasteful.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural extension is to add a random shock to the score equation; with risk-averse families the arms race may intensify because extra study acts as insurance against missing the cutoff, while risk-neutral families might reduce effort when success becomes a lottery.
The same best-response structure should apply to other contests with a rising cutoff, such as credential inflation in graduate admissions or competition for scarce housing; the model's quantitative predictions are testable by comparing equilibrium effort across settings with different score variances.
The paper's policy of weakening the link between academic performance and future wages could be tested by comparing regions or cohorts with different perceived wage premia; if the model is right, study time should respond to that perceived premium even when actual resources are unchanged.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The paper has a real policy question and a standard model, but the central equilibrium claim is internally inconsistent. The two-family game in §3.3 claims an unstable Nash equilibrium when both families disobey, but substituting the Table 1 best responses gives t1 = t1 + 2: the best-response lines are parallel, so no equilibrium exists, stable or unstable. That is a load-bearing flaw, because the arms-race narrative and the policy conclusion rest on that equilibrium.

What it does well: it takes a widely discussed policy failure—China's burden-reduction policies—and tries to formalize the strategic complementarity that makes individual defection rational. The setup is a rank-order tournament with a cutoff and a log utility, a structure familiar from contest theory. The qualitative story—competition escalates because each family's effort raises the cutoff for everyone—is plausible and is indeed the standard intuition. The paper also honestly states it uses no real data; the policy sections are explicitly illustrative.

Soft spots: besides the missing equilibrium, Eq. (13) and (20) have notational/formatting issues, but the algebra is actually consistent if you read them as intended fractions. The bigger problem is that the model defines the cutoff as a function of the average score, which builds in the arms race. The Spence extension introduces a cognitive bias parameter β >10 with no calibration; that is invented to produce the irrationality conclusion. The paper also does not cite the contest-theory literature that already has these results, so the novelty is thin.

For a reader: this is a work-in-progress that shows an interesting question and a plausible mechanism, but the formal analysis does not support the stated results. It is not publishable as-is. A serious referee could point out the flaw and the author could fix the model—for example by explicitly analyzing the no-equilibrium case as unbounded effort—but that would be a substantial revision.

Recommendation: send to a referee if you want to see whether the author can repair the equilibrium analysis; otherwise desk-reject. The paper is not ready for publication.

Referee Report

3 major / 4 minor

Summary. The paper proposes a game-theoretic account of China's education 'arms race.' It assumes a student's score is a deterministic product of aptitude and study time, defines a cutoff based on the average and variance of scores, and lets a family maximize log utility minus linear time cost. It analyzes a two-family game, claims an unstable Nash equilibrium in the both-disobey outcome, simulates the effects of score variance on the cutoff, and extends the model with a Spence signaling framework and a cognitive-bias parameter β. The policy section concludes that burden-reduction policies fail, that variance-reducing policies trade off equity against welfare, and that weakening the link between education and wages plus reducing β is the recommended remedy.

Significance. The topic is important and the paper is clearly written, with a transparent model and explicit limitation statements. However, the formal analysis contains load-bearing errors: the two-family game has no interior Nash equilibrium under the paper's own best responses, and key closed-form best-response equations are missing a division by γ_i. Because the policy conclusions in Sections 3.3 and 5 rest on those results, the central claims are not supported. The paper does not provide empirical data or robustness checks. Its potential contribution—linking competition escalation, equity/welfare trade-offs, and social cognition—would be interesting if the model were repaired, but in its current form the formal basis is absent.

major comments (3)

[§3.3, Table 1 and Table 2] The claimed 'unstable Nash equilibrium' in the both-disobey outcome does not exist. Solving the best responses t1* = (4t2+6)/5 and t2* = (5t1+4)/4 gives t1 = t1 + 2 (and t2 = t2 + 2.5), so no finite pair (t1,t2) satisfies both equations. Consequently, the (Disobey, Disobey) cell in Table 2 is not a well-defined payoff pair; it depends on t1 and t2 that are never simultaneously determined. The subsequent feedback loop in Eq. (21) presupposes an equilibrium, so Section 3.3's central claim about burden-reduction policy failure is unsupported.
[§3.2, Eq. (13); §3.4, Eq. (20)] Both closed-form best responses are algebraically wrong as displayed. From Eq. (11), setting ∂u_i/∂t_i = 0 yields t_i* = 1/P + γ_j t_j / γ_i - 4/γ_i, not t_i* = 1/P + γ_j t_j - 4/γ_i. Similarly, Eq. (20) should be t_i* = 1/P + (Scut - 2)/γ_i, not t_i* = 1/P + Scut - 2/γ_i. Table 1 uses the corrected two-family formula, so the text and the table are inconsistent; readers cannot reproduce the claimed best responses from the displayed equations.
[§5.1, Eq. (23)] The cognitive-bias parameter β > 10 is introduced without derivation, data, or calibration, and it directly produces the 'irrationality' conclusion. Since the paper's policy recommendation in Section 5.2 rests on reducing β, this is an ad hoc assumption rather than a result of the model. The paper itself acknowledges in the concluding limitations that uncertainty around achieving Scut is only partially addressed and that no real-world data are used; these concessions should be reflected in the strength of the policy claims.

minor comments (4)

[§3.3, Eq. (14)] The expression '10+8/2' should read '(10+8)/2'; as written it is ambiguous and would give 14 instead of 9.
[§3.3] The prose states that equalizing aptitudes moves the first-best from social utility -0.6 to -0.79, but the underlying social-welfare sums are not shown; Table 2's (Obey, Obey) cell sums to -0.9 and Table 3's (Obey, Obey) cell sums to -0.6, so the comparison would benefit from an explicit calculation.
[§5.2] The citation 'Zhu and Zhu (2018)' does not match the reference list, which contains Zhu and Zhu (2002).
[§2.2] There is a typo, 'arm race' should be 'arms race'.

Circularity Check

2 steps flagged · score 6.0 of 10

The arms-race and irrationality conclusions are already contained in the model's cutoff definition and in the free parameter β, rather than being independently derived; the claimed §3.3 Nash equilibrium is additionally unsupported by the paper's own best-response equations.

self definitional [Section 3.5, Equation (21), with Equations (2), (3), and (20)]
"Scut and individual best response t∗i constructs a mutually reinforcement structure: higher Scut will bring higher t∗i. And because everyone adjusts learning time to achieve new t∗i, a even higher Scut is again produced. And meanwhile, everyone’s u∗i decreases. Scut ↑ =⇒ t∗i ↑ =⇒ ¯t ↑ =⇒ Scut ↑ (21)"

The chain (21) is the composition of the paper's own definitions: Eq. (20) makes t_i* = 1/P + S_cut - 2/gamma_i, so t_i* increases mechanically with S_cut; Eq. (3) defines S_cut = mean + k*sigma_S, so S_cut increases mechanically with the mean score; and Eq. (2) defines the mean as the average of gamma_i*t_i. Thus the loop holds for any parameter values and contains no independent empirical content. The paper presents this formal identity as the discovery of an 'education arms race' and as the explanation for the failure of burden-reduction policy, so the central prediction reduces by construction to the relative-cutoff assumption in Eqs. (2)-(3).
other [Section 5.1, 'Understanding the irrationality' and Equation (23)]
"the coefficient β of Whigh reflects a cognitive bias, which is people may mentally magnify the extra benefit of Si > Scut by more than 10 times (e.g. β >10), Therefore, when making decisions, they believe that better academic rankings are everything sufficient of success, and do not give up studies no matter what."

The irrational persistence that the model claims to explain is installed as the free parameter beta > 10 in Eq. (23). This parameter is never estimated, derived, or externally anchored; the conclusion that families 'do not give up studies no matter what' is a paraphrase of the assumption beta > 10 rather than a result obtained from it. The subsequent policy proposal to 'reduce beta' simply reverses the inserted assumption, so the model's 'measurement' of irrationality reduces to the chosen magnitude of an unmeasured input.

full rationale

The paper is self-contained in the sense that it does not rely on self-citations or on an imported uniqueness theorem: the cited works (Spence 1973; Zhu and Zhu 2002; Xiang 2019, etc.) are background or inspiration, not load-bearing proofs by the same author. However, the two central qualitative conclusions are not independent of the model's inputs. The escalation loop in Eq. (21) follows directly from defining S_cut as a function of the average score (Eq. 3) and from the best-response formula (Eq. 20), so the 'education arms race' is a formal identity of the assumed relative-cutoff utility structure, not a separately established prediction. Similarly, the irrationality result is produced by the free cognitive-bias parameter beta > 10 in Eq. (23), and the policy advice to reduce beta is tautological. The paper itself concedes in Section 6 that it 'does not incorporate any real-world data at this stage,' so there is no external benchmark to give these derived identities independent falsifiable content. A separate, non-circular but serious formal problem is that the two-family equilibrium in Section 3.3 is not actually derived: the best responses in Table 1, t1* = (4t2+6)/5 and t2* = (5t1+4)/4, are parallel, so substituting one into the other gives t1 = t1 + 2 and no Nash equilibrium, stable or unstable, exists. Because this unsupported equilibrium is invoked as the basis for the burden-reduction-policy conclusions, the formal center of the paper is unreliable even apart from circularity. On the circularity scale, the main 'predictions' reduce by construction to the model's definitions and free parameters, giving a score of 6.

Assumptions & free parameters 6 free parameters · 4 assumptions · 1 invented entities

The model's conclusions are driven by hand-chosen parameters and the ad hoc β. No parameter is estimated from data, and the central predictions (escalation, policy ineffectiveness) are embedded in the modeling assumptions rather than independently tested. The free parameters are illustrative, but the paper presents the results as general.

free parameters (6)

P (time cost coefficient) = 0.5
Set by hand in all numerical examples; controls the slope of best responses and the magnitude of study time.
k (cutoff spread coefficient) = 1.645
Chosen to mimic a 5% cutoff percentile; used in Figure 2 simulations without independent justification.
γ1, γ2 (aptitudes) = 5 and 4
Chosen for the 2x2 example; the paper itself shows results change when γ2 is set to 5 (Table 3).
β (cognitive bias multiplier) = >10
Introduced in Section 5.1 without calibration or evidence; load-bearing for the irrationality conclusion.
W_high, W_low (wages) = 2000 and 1000 pounds
Illustrative wage values used in the Spence extension; not grounded in data.
σ_S (score variance settings) = 1 and 3
Hand-picked for Figure 2 to demonstrate the effect of variance; no empirical basis.

assumptions (4)

domain assumption Scores are deterministic: S_i = γ_i t_i with no random error
Equation (1); load-bearing for all best-response derivations and the cutoff-based utility.
domain assumption High-quality educational resources are scarce and access is determined by relative exam score
Section 2.1; defines the positional-good nature of the game.
domain assumption Most families expect to clear the cutoff (S_i ≥ S_cut) and continue investing
Section 2.2; justifies using only the winning branch of the utility function and the 'no give up' behavior.
ad hoc to paper β > 10 cognitive bias magnifies perceived wage premium
Section 5.1; no evidence or calibration, used to explain irrational persistence in education.

invented entities (1)

β (cognitive bias multiplier)
purpose: Models the claim that families mentally magnify the benefit of high exam scores by more than 10x, driving irrational persistence in the arms race.
The paper asserts β > 10 without data, calibration, or a falsifiable implication; it is introduced to fit the desired conclusion of irrational behavior.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Quantifying Educational Competition: A Game-Theoretic Model with Policy Implications." pith.science (2026). https://pith.science/paper/JFVCGRL3

@misc{pith2026241210974,
  author       = {Pith},
  title        = {Pith review of: Quantifying Educational Competition: A Game-Theoretic Model with Policy Implications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JFVCGRL3}},
  note         = {Machine review of arXiv:2412.10974}
}

read the original abstract

The competitive pressures in China's primary and secondary education system have persisted despite decades of policy interventions aimed at reducing academic burdens and alleviating parental anxiety. This paper develops a game-theoretic model to analyze the strategic interactions among families in this system, revealing how competition escalates into a socially irrational "education arms race." Through equilibrium analysis and simulations, the study demonstrates the inherent trade-offs between education equity and social welfare, alongside the policy failures arising from biased social cognition. The model is further extended using Spence's signaling framework to explore the inefficiencies of the current system and propose policy solutions that address these issues.

Figures

Figures reproduced from arXiv: 2412.10974 by the authors.

**Figure 1.** The difference in Scut affects same student’s best response. Parameters Definitions Value γi Student i’s Aptitude 3 P Time Cost Coefficient 0.5 Scut S 1 cut Set lowest threshold 0 S 2 cut Set the threshold to 3 3 [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗

**Figure 2.** The difference in σS affects same student’s utility. Policy 1: “CEE” Policy. CEE makes all students can go to the same examination room to fight for opportunities equally. However, this “participate by all people” feature makes it naturally a strong educational arms race. The high participant variance causes Scut continue to rise, and social welfare continue to decrease. 4.2 Lower variance brings the less education… view at source ↗

**Figure 3.** Difference in γi let families behave differently in education decisionmaking. Update the Model. Equation (23) is the updated payoff function for families to include wage’s impact on educational decision. The new added wage variables (Wlow and Whigh) are essentially reverse signals to job seekers (students). And it allow families to plan their learning time more rationally. If the wage Whigh premium for higher educa… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

11 extracted references · 11 canonical work pages

[1]

South China Morning Post

Anon (2023) 50-50 gamble for parents: A policy enacted six years ago to boost vocational education means that only half of junior high students in China can go on to get academic degrees. South China Morning Post

work page 2023
[2]

and Tirole, J

Dewatripont, M., Jewitt, I. and Tirole, J. (1999) The economics of career concerns, Part I: Comparing information structures. The Review of Eco- nomic Studies, 66(1), pp. 183–198

work page 1999
[3]

and Cai, Q

Jia, W., Deng, J. and Cai, Q. (2021) Game dilemmas and countermeasures for burden reduction of primary and secondary school students from the perspective of stakeholders. China Educational Technology, (09), pp. 51–58

work page 2021
[4]

and Sun, C

Li, H. and Sun, C. (2004) Survey and countermeasures on the burden of urban middle school students in Jin Nan and Jin Dong Nan regions. Shanxi Education, (17)

work page 2004
[5]

Double Reduction

Liu, F. and Dong, X. (2022) Key issues and contradictions in the imple- mentation of the “Double Reduction” policy. Journal of Xinjiang Normal 13 University (Philosophy and Social Sciences Edition), (01), pp. 91–97. DOI: 10.14100/j.cnki.65-1039/g4.20211109.003

work page doi:10.14100/j.cnki.65-1039/g4.20211109.003 2022
[6]

(1973) Job market signaling

Spence, M. (1973) Job market signaling. The Quarterly Journal of Eco- nomics, 87(3), pp. 355–374

work page 1973
[7]

and Liu, J

Wang, Y. and Liu, J. (2018) Analysis of changes and trends in primary and secondary school burden reduction policies over forty years of reform and opening up. Theory and Practice of Education, (31), pp. 17–23

work page 2018
[8]

burden reduc- tion

Xiang, X. (2019) A historical perspective on two rounds of “burden reduc- tion” education reforms in China over seventy years. Journal of East China Normal University (Educational Sciences Edition), (05), pp. 67–79. DOI: 10.16382/j.cnki.1000-5560.2019.05.006

work page arXiv 2019

Show all 11 references

[9]

burden reduction

Yang, L. and Zhang, X. (2019) Historical review and reflection on China’s “burden reduction” policies since the founding of the PRC. Educational Research, (02), pp. 13–21

2019
[10]

and Wan, L

Zhang, W. and Wan, L. (2018) Analysis of causes and countermeasures for deviations in the implementation of education burden reduction policies: From the perspective of stakeholders. Modern Education Science, (08), pp. 51–55, 61. DOI: 10.13980/j.cnki.xdjykx.2018.08.010

2018 doi
[11]

and Zhu, X

Zhu, J. and Zhu, X. (2002) Burden reduction for primary and secondary school students and the prisoner’s dilemma game. Educational Science, (04), pp. 11–13. 14

2002

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

South China Morning Post

Anon (2023) 50-50 gamble for parents: A policy enacted six years ago to boost vocational education means that only half of junior high students in China can go on to get academic degrees. South China Morning Post

work page 2023

[2] [2]

and Tirole, J

Dewatripont, M., Jewitt, I. and Tirole, J. (1999) The economics of career concerns, Part I: Comparing information structures. The Review of Eco- nomic Studies, 66(1), pp. 183–198

work page 1999

[3] [3]

and Cai, Q

Jia, W., Deng, J. and Cai, Q. (2021) Game dilemmas and countermeasures for burden reduction of primary and secondary school students from the perspective of stakeholders. China Educational Technology, (09), pp. 51–58

work page 2021

[4] [4]

and Sun, C

Li, H. and Sun, C. (2004) Survey and countermeasures on the burden of urban middle school students in Jin Nan and Jin Dong Nan regions. Shanxi Education, (17)

work page 2004

[5] [5]

Double Reduction

Liu, F. and Dong, X. (2022) Key issues and contradictions in the imple- mentation of the “Double Reduction” policy. Journal of Xinjiang Normal 13 University (Philosophy and Social Sciences Edition), (01), pp. 91–97. DOI: 10.14100/j.cnki.65-1039/g4.20211109.003

work page doi:10.14100/j.cnki.65-1039/g4.20211109.003 2022

[6] [6]

(1973) Job market signaling

Spence, M. (1973) Job market signaling. The Quarterly Journal of Eco- nomics, 87(3), pp. 355–374

work page 1973

[7] [7]

and Liu, J

Wang, Y. and Liu, J. (2018) Analysis of changes and trends in primary and secondary school burden reduction policies over forty years of reform and opening up. Theory and Practice of Education, (31), pp. 17–23

work page 2018

[8] [8]

burden reduc- tion

Xiang, X. (2019) A historical perspective on two rounds of “burden reduc- tion” education reforms in China over seventy years. Journal of East China Normal University (Educational Sciences Edition), (05), pp. 67–79. DOI: 10.16382/j.cnki.1000-5560.2019.05.006

work page arXiv 2019