📚 Paternalism in Mathematical Decision Models: Utility, Constraints and Welfare | 数学决策模型中的家长主义：效用、约束与福利
In many A-Level mathematics questions, students meet optimisation, probability and utility-like functions. One interdisciplinary idea that links these tools is paternalism: a decision maker substitutes their own objective for an individual's choice. This article formalises paternalism using the mathematical language of Edexcel A-Level units such as Pure Mathematics, Statistics and Decision Mathematics, and shows how the same calculus, constraints and algorithms can be re-read as statements about whose preferences are being maximised.
在许多 A-Level 数学题中，学生都会遇到最优化、概率和类似效用的函数。连接这些工具的一个跨学科概念就是家长主义：决策者用自己的目标取代个体的选择。本文使用 Edexcel A-Level 单元（如纯数学、统计学和决策数学）的数学语言来形式化家长主义，并说明同样的微积分、约束与算法如何被重新解读为"究竟在最大化谁的偏好"这一问题。
1. Defining Paternalism in Model Terms | 用模型术语定义家长主义
Paternalism in mathematics means a planner restricts an agent's feasible set because the planner believes the agent' optimal. If an individual has a utility function U(x), paternalism replaces the individual's maximiser x* = arg max U(x) with a planner's preferred point xp.
在数学中，家长主义意味着计划者限制行为人的可行集，因为计划者认为行为人自己的选择不是最优的。如果个体的效用函数为 U(x)，家长主义将个体的最大化点 x* = arg max U(x) 替换为计划者偏好的点 xp。
It is useful to state three conditions. First, interference: the original feasible set F is replaced by a smaller set Fp ⊂ F, or the objective is replaced by V(x) ≠ U(x). Second, non-consent: the agent has not agreed to the substitution. Third, justification: the stated reason refers to the welfare of the very agent whose choice is restricted.
我们可以给出三个条件。第一，干预：原来的可行集 F 被替换为更小的集合 Fp ⊂ F，或者目标函数被替换为 V(x) ≠ U(x)。第二，未经同意：行为人并未同意这种替换。第三，理由：给出的理由指向的正是被限制选择的那个人自身的福利。
The key difference is not the mathematics itself, but whose objective enters the optimisation. The same calculus and constraint techniques from A-Level Pure Mathematics are used, but the objective function changes.
关键区别不在于数学本身，而在于谁的目标进入最优化。使用的仍然是与 A-Level 纯数学相同的微积分和约束技术，但目标函数发生了变化。
x* = arg max U(x) → xp, with xp ∈ Fp ⊂ F and V(xp) ≥ V(x) for all x ∈ Fp
2. Utility Functions and Individual Choice | 效用函数与个体选择
In decision models, an individual' imise U subject to a constraint such as x ≥ 0 or px ≤ m, where p is a price and m is available income.
在决策模型中，个体的偏好可以用效用函数 U(x) 表示，其中 x 可以是消费、学习时间或风险暴露。个体在约束（如 x ≥ 0 或 px ≤ m，其中 p 是价格、m 是可用收入）下选择 x 以最大化 U。
For example, if U(x) = 8x − x² with x ≥ 0, then U'(x) = 8 − 2x, so U'(x) = 0 gives x* = 4. Because U''(x) = −2 < 0, this stationary point is a maximum and U(4) = 32 − 16 = 16.
例如，若 U(x) = 8x − x²，x ≥ 0，则 U'(x) = 8 − 2x，令 U'(x) = 0 得 x* = 4。由于 U''(x) = −2 < 0，该驻点为最大值，且 U(4) = 32 − 16 = 16。
aller, the planner may use a different objective, say V(x) = 4x − x², whose maximiser is xp = 2. The gap |x* − xp| = 2 measures the behavioural change imposed by the paternalistic objective.
如果计划者认为 x 的真实收益更小，计划者可能使用不同的目标，例如 V(x) = 4x − x²，其最大化点为 xp = 2。差值 |x* − xp| = 2 度量了家长式目标所强加的行为改变量。
A subtle point for exam answers: U and V may both be concave quadratics, so both give well-defined interior maxima. The difference is entirely in the coefficients, and therefore entirely in the interpretation, not in the technique.
考试作答时的一个细节：U 与 V 都可以是凹二次函数，因此两者都能给出良定义的内点最大值。差别完全在系数上，因而也完全在解释上，而不在技术方法上。
3. Constraints, Feasible Sets and Lagrange Thinking | 约束、可行集与拉格朗日思路
oncave objective such as U(x) = 8x − x² the interior solution is found by differentiation instead.
单一预算约束下的可行集为 F = {x : x ≥ 0, px ≤ m}。目标为线性时，最优解总位于 F 的顶点；而目标为严格凹函数（如 U(x) = 8x − x²）时，内点解则通过求导得到。
ing is the Lagrangian L(x, λ) = U(x) + λ(m − px). The first-order conditions are ∂L/∂x = U'(x) − λp = 0 and ∂L/∂λ = m − px = 0.
用 A-Level 之上的视角看同一问题，即拉格朗日函数 L(x, λ) = U(x) + λ(m − px)。一阶条件为 ∂L/∂x = U'(x) − λp = 0 与 ∂L/∂λ = m − px = 0。
U'(x) = λp and px = m, so U'(x) ÷ p = λ
ither channel: it can change U, or it can change the constraint px ≤ m by adding a cap x ≤ xmax or a tax t that replaces p with p + t.
因此，每一英镑带来的边际效用在使用之间被拉平，λ 是预算放松一单位的影子价格。家长主义可以从两个渠道起作用：改变 U，或者改变约束 px ≤ m——例如加上上限 x ≤ xmax，或用税收 t 把 p 变成 p + t。
4. Social Welfare Functions and Aggregation | 社会福利函数与加总
nction W. A utilitarian planner uses W = Σ Uᵢ(xᵢ). A Rawlsian planner uses W = min{U₁, U₂, … d planner uses W = Σ wᵢUᵢ(xᵢ) with wᵢ > 0 expressing how much weight each person receives.
福利经济学把个体效用加总为社会福利函数 W。功利主义计划者使用 W = Σ Uᵢ(xᵢ)。罗尔斯式计划者使用 W = min{U₁, U₂, …}，只关心境况最差的行为人。加权计划者使用 W = Σ wᵢUᵢ(xᵢ)，其中 wᵢ > 0 表示每个人的权重。
Pure paternalism appears when the planner maximises a different utility from the agent' but the planner also counts an externality, the intervention is efficiency-based rather than paternalistic.
纯家长主义出现在计划者最大化的效用与行为人自身效用不同之时：对同一个人而言 V(x) ≠ U(x)。反之，若对每个行为人都有 V(x) = U(x)，只是计划者另外计入了外部性，则该干预属于效率型而非家长型。
raising the minimum. The agent 1 loses utility, and that loss is the paternalistic cost measured on the U₁ scale.
设干预前两个行为人各有 x₁ = x₂ = 4。若罗尔斯式计划者认为 U₂ 被错误设定，可能重新配置资源使 x₂ = 6、x₁ = 3，从而提高最小值。行为人 1 的效用下降，这一损失就是以 U₁ 尺度衡量的家长主义成本。
5. A Taxonomy of Paternalism | 家长主义的分类
nly removes options that the agent would reject if fully informed, so it replaces F with Fp but claims xp ≻ x* under corrected preferences. Hard paternalism imposes x = xp regardless of the agent' ng, which in a statistical model means changing the probability distribution over options.
不同程度的干预对应着模型上不同的数学操作。软性家长主义只移除行为人在完全知情下也会拒绝的选项，因此它把 F 换为 Fp，但主张在"修正后的偏好"下 xp ≻ x*。硬性家长主义不论行为人知情偏好如何，直接强制 x = xp。自由意志式家长主义保持 F 不变，只改变默认选项或呈现框架——在统计模型中即改变各选项的概率分布。
| Type | 类型 | Model feature | 模型特征 | A-Level link | A-Level 联系 |
| Soft | 软性 | F shrinks to Fp, claim about corrected preferences | 可行集缩小，主张修正偏好 | Inequalities, conditional probability | 不等式、条件概率 |
| Hard | 硬性 | x = xp imposed for all agents | 对所有行为人强制 x = xp | Equality constraints | 等式约束 |
| Libertarian | 自由意志式 | F unchanged, distribution over options shifts | 可行集不变，选项分布改变 | Probability, Statistics | 概率、统计学 |
| Efficiency | 效率型 | Planner adds an externality term | 计划者加入外部性项 | Rates of change, integration | 变化率、积分 |
Only the first three rows are genuinely paternalistic, because the justification refers to the restricted agent's own welfare. The fourth row is ordinary regulation, and mixing the two up is the most common conceptual error in extended answers.
只有前三行才是真正的家长主义，因为其理由指向被限制者自身的福利。第四行属于普通规制；把二者混淆是扩展题作答中最常见的概念错误。
6. Internalities, Present Bias and Discounting | 内部性、现时偏误与贴现
iscounting is time-consistent, meaning that the plan made today is still optimal tomorrow.
许多福利模型用每期因子 δ（0 < δ < 1）对未来效用贴现：W = Σ δᵗ U(cₜ)。指数贴现具有时间一致性，即今天做出的计划在明天依然最优。
o save, but at t = 1 they re-optimise with the same βδ and defect. The inconsistency is a mathematical property of the objective, not a moral failing.
行为模型把 δ 替换为 βδ，其中 β < 1，且 β 只作用于当期。于是行为人具有强烈的现时偏误：在 t = 0 时计划储蓄，但在 t = 1 时用相同的 βδ 重新最优化并发生偏离。这种不一致是目标函数的数学性质，而不是道德缺陷。
Internality means the agent' nefit(x) − c(x), where the true cost is c(x) + h(x) and h(x) > 0 is ignored. A planner who maximises U(x) − h(x) and imposes xp < x* is paternalistic in exactly the technical sense of Section 1.
"内部性"指行为人自己的决定产生了行为人在决策当下并未内化的成本：例如 U(x) = benefit(x) − c(x)，而真实成本是 c(x) + h(x)，其中 h(x) > 0 被忽略。若计划者最大化 U(x) − h(x) 并强制 xp < x*，那么它恰好符合第 1 节意义上的家长主义。
The soft-paternalist justification is consent at the ex-ante stage: the agent's own long-run self agrees with the restriction, even though the short-run self does not. This is why the argument depends on the discount factor, and why δ and β must be stated explicitly in any model answer.
软性家长主义的理由是事前阶段的同意：行为人的长期自我同意该限制，尽管短期自我不同意。这正是该论证依赖于贴现因子的原因，也是为什么在任何模型答案中都必须明确写出 δ 与 β。
7. Asymmetric Information and Principal-Agent Models | 信息不对称与委托-代理模型
principal-agent model the agent chooses hidden effort e, and output is y = f(e) + ε with ε ~ e. Because effort is costly to the agent, the privately optimal choice satisfies e* < esocial, the classic moral hazard result.
在委托-代理模型中，行为人选择不可观测的努力 e，产出为 y = f(e) + ε，其中 ε ~ N(0, σ²)。委托人观察到 y 但观察不到 e。由于努力对行为人是有成本的，私人最优选择满足 e* < esocial，即典型的道德风险结果。
A paternalistic contract imposes e ≥ emin directly, or prices insurance so heavily that the agent's feasible set no longer contains e*. The welfare effect depends on whether the principal's objective coincides with the agent' y: if it does, the restriction is efficiency-improving; if it does not, it is paternalism.
家长式合同直接施加 e ≥ emin，或者把保险定价得足够高，使行为人的可行集中不再包含 e*。福利效果取决于委托人的目标是否与行为人的真实效用一致：若一致，该限制提升效率；若不一致，它就是家长主义。
arison: with a linear contract w = a + by, the agent maximises a + bf(e) − c(e), giving bf'(e) = c'(e). Changing b changes e*, while imposing e ≥ emin two instruments are mathematically different even when they produce similar effort levels.
一个考试式的简洁对比：在线性合同 w = a + by 下，行为人最大化 a + bf(e) − c(e)，得到 bf'(e) = c'(e)。改变 b 会改变 e*，而施加 e ≥ emin 则改变可行集。即使两者产生相近的努力水平，它们在数学上仍是不同的工具。
8. Nudges versus Mandates: Comparative Statics | 助推与强制：比较静态
nchanged; only the optimum changes. A mandate instead changes F itself, for example by adding x ≤ xmax.
税收 t 把有效价格从 p 提高到 p + t，使预算线旋转
更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图