本最后一章汇集了全书的线索——微观、宏观、制度和实证——来回答经济学中最重要的问题：为什么有些国家富裕而另一些贫穷，以及能做些什么？
发展经济学不是"应用增长理论"。它处理的是标准模型所抽象掉的协调失败、制度陷阱、人力资本缺口和政治经济学。它还展现了现代经济学中最引人注目的方法论革命：随机对照试验作为评估干预手段的兴起——以及最近寻求超越单个实验所能揭示的结构估计的反潮流。
本章综合了整本教科书。增长理论（第13章）提供框架。制度（第18章）提供深层决定因素。计量经济学（第10章）提供识别工具——工具变量、断点回归和因果推断的逻辑。行为洞见（第19章）为发展干预的设计提供信息。
前置知识：第10章（计量经济学基础——IV、回归），第13章（增长理论——Solow模型、稳态），第18章（制度经济学——AJR、掠夺性/包容性制度），第19章（行为经济学——助推、RCT）。
相关文献：Lewis（1954）；Rosenstein-Rodan（1943）；Murphy, Shleifer & Vishny（1989）；Acemoglu, Johnson & Robinson（2001）；Nunn（2008）；Mincer（1974）；Bleakley（2007）；Miguel & Kremer（2004）；Banerjee, Duflo & Kremer（2019年诺贝尔奖）；Todd & Wolpin（2006）；Attanasio, Meghir & Santiago（2012）；Deaton（2010）；Allcott（2015）；Lin（2012）；Rodrik（2004）。
最富裕的国家——挪威、瑞士、美国——人均GDP超过\$60,000（PPP）。最贫穷的国家——布隆迪、南苏丹、中非共和国——人均GDP低于\$500。最富和最穷之间相差超过100倍，而这一差距在两个世纪内急剧扩大。1800年，最富与最穷的比率约为5:1。到2000年，超过了100:1。这一"大分流"是发展经济学必须解释的核心事实。
Penn World Table揭示了若干模式。在19世纪初，分布大致单峰：几乎所有国家都很穷。工业革命创造了一个在20世纪加速的分流。到20世纪70-80年代，分布已明显变为双峰——"双峰"（Quah 1996）。2000年以来，中国和印度的快速增长部分填补了这一差距，但撒哈拉以南非洲基本仍处于较低的峰值。
| Kaldor事实（第13章） | 发展事实（本章） |
|---|---|
| 恒定的资本-产出比 | 工业化过程中上升的资本-产出比 |
| 恒定的劳动份额 | 农业劳动份额下降，工业先升后降，服务业上升 |
| 恒定的人均产出增长率 | 高度可变的增长；加速和停滞交替 |
| 平衡增长路径 | 结构转型；非平衡的、部门转移式增长 |
索洛模型（第13章）很好地捕捉了Kaldor事实。它没有捕捉到发展事实——它只有一个部门、一种劳动和平滑的收敛。发展经济学需要具有多个部门、异质劳动和陷阱可能性的模型。
图20.3.全球收入分布随时间变化（概式化）。滑动浏览各十年，查看从单峰（1800年）到双峰（20世纪70年代）再到部分收敛（2000年代）的演变。使用滑块或播放按钮。
现代部门使用资本和劳动力的Cobb-Douglas生产函数：
现代部门使用资本和劳动力的Cobb-Douglas生产函数：
现代部门在$MPL_M > \bar{w}$时雇用工人。在剩余劳动力阶段，现代部门面临以工资$\bar{w}$为基准的完全弹性劳动供给。利润（$\Pi_M = Y_M - \bar{w}L_M$）被再投资，创造良性循环：资本积累提高$MPL_M$，吸收更多工人，产生更多利润。
中国是最引人注目的现代例证。1980年至2010年间，中国将数亿工人从农村农业转移到城市制造业，实现了每年10%的增长率。经济学家争论中国是否在2010-2015年左右跨越了刘易斯转折点，证据是沿海制造业地区工资快速上涨。
图20.2.刘易斯二元经济模型。左：现代部门MPL曲线和维持生计工资。右：各部门产出。增加资本以吸收劳动力；注意刘易斯转折点。拖动滑块探索。
凯拉尼共和国有1000万工人。目前700万在维持生计部门工作，剩余劳动力为300万（$\bar{L} = 4$百万）。现代部门：$A_M = 2$，$K_M = 100$，$\alpha = 0.4$。
(a)当前现代部门产出（$L_M = 3$M）：$Y_M^{\text{before}} = 2 \times 100^{0.4} \times 3^{0.6} \approx 24.40$。重新分配100万工人后（$L_M = 4$M）：$Y_M^{\text{after}} = 2 \times 100^{0.4} \times 4^{0.6} \approx 28.99$。产出增益 = 4.59单位（增长18.8%），维持生计部门零损失，因为转移的工人是剩余劳动力。
(b)在转折点，$L_M = L - \bar{L} = 6$M。令$MPL_M = \bar{w} = 1$：$K_M^* \approx 3.80$——反映了剩余劳动力充裕和维持生计工资适中的低门槛。
标准索洛模型具有凹生产函数，保证唯一稳定的稳态。贫困陷阱需要S形（局部凸的）生产函数，在$sf(k)$和$(n+\delta)k$之间创造多个交叉点。
图20.1.贫困陷阱图。S形$sf(k)$曲线与$(n+\delta)k$线相交于最多三个点。拖动圆点查看收敛到低陷阱或高均衡。用滑块调整储蓄率和曲率。拖动初始条件圆点探索。
MSV模型产生两个Nash均衡：无工业化（贫困陷阱）和全面工业化（发达均衡）。政府可以充当协调机制——补贴跨部门的同步投资。
并非所有贫穷国家都陷入了陷阱。克雷和麦肯齐（2014）发现家庭层面贫困陷阱的证据有限。在国家层面，撒哈拉以南非洲的持续欠发达更符合陷阱动态，特别是当与制度失败和冲突结合时。
给定$f(k) = k^2/(1+k^2)$（S形），$s = 0.20$，$n+\delta = 0.10$。令$sf(k) = (n+\delta)k$并求解得$k = 0$和$k = 1$（重根——陷阱处于存在的边缘）。
更丰富的例子：$f(k) = k^{2.2}/(1+k^{2.2})$得到三个解：$k_L^* \approx 0$（贫困陷阱），$k_U \approx 0.72$（不稳定阈值），$k_H^* \approx 1.45$（高均衡）。在$k_U$处，生产函数局部凸，$g'(k_U) > 0$——不稳定。大推进需要注入每工人$\Delta k \approx 0.72$。
根本挑战是内生性：富裕国家能负担更好的制度。AJR（2001）提出了使用定居者死亡率的IV策略。第一阶段系数$\beta$为负且高度显著（F统计量 > 20）。2SLS估计$\hat{\delta} \approx 0.94$超过OLS（$\approx 0.52$）——与测量误差导致的衰减偏差一致。
自然实验强化了制度假说：朝鲜与韩国、东德与西德、改革前后的中国以及博茨瓦纳与其邻国都说明了制度分化如何驱动收入分化。
图20.4.制度与地理散点图。切换x轴变量以比较定居者死亡率、纬度和法治指数作为收入预测因子。使用下拉菜单切换视图。
结果：第一阶段F = 22.9，$\hat{\beta} = -0.61$，2SLS $\hat{\delta} = 0.94$（SE = 0.16），OLS = 0.52。(a)制度质量每增加一个单位导致人均GDP增加0.94个对数点。从第25百分位（得分5）到第75百分位（得分8）预测增加\$1 \times 0.94 = 2.82$个对数点——大约16.8倍。
(b)排除性限制的威胁：定居者死亡率可能代理当前疾病环境（直接降低生产率）；欧洲人可能在制度之外对基础设施进行了不同的投资。(c)IV > OLS可能由于衰减偏差：如果可靠性比率约为0.55，则\$1.52/0.55 \approx 0.94$。
| 收入组别 | 平均回报率（ρ̂） |
|---|---|
| 低收入国家 | 10.5% |
| 中低收入国家 | 8.7% |
| 中高收入国家 | 7.2% |
| 高收入国家 | 5.4% |
布莱克利（2007）利用钩虫感染流行率的地理变异表明，每标准差减少对应17%的收入增长。米格尔和克雷默（2004）发现驱虫将旷课率降低了25%，并具有大量溢出效应——每额外一年出勤约\$3.50，是已知最具成本效益的发展干预之一。
图20.5.明塞尔方程探索器。调整受教育年限和回报率，查看对数工资曲线如何移动。虚线显示额外4年教育的溢价。拖动滑块探索。
A国（低收入）：$\hat{\rho} = 0.10$，$\hat{\beta}_1 = 0.03$，$\hat{\beta}_2 = -0.0005$。B国（高收入）：$\hat{\rho} = 0.05$，$\hat{\beta}_1 = 0.05$，$\hat{\beta}_2 = -0.0008$。4年额外教育的溢价：A国 = $e^{0.40}-1 = 49.2\%$；B国 = $e^{0.20}-1 = 22.1\%$。
工资峰值经验年数$\text{Exp}^* = \beta_1 / (2|\beta_2|)$：A国为30年，B国为31.25年。回报率差异源于稀缺性、能力偏差、信贷约束、学校质量以及信号与人力资本效应。
巴纳吉、迪弗洛和克雷默因其减轻全球贫困的实验方法获得2019年诺贝尔奖。关键发现：现金转移有效且不减少努力；小额信贷不具变革性；驱虫具有极高的成本效益。RCT革命最大的贡献是用证据取代了先验信念。
| 干预措施 | 研究发现 | 研究 |
|---|---|---|
| 驱虫 | 缺勤率降低25%；显著的溢出效应 | Miguel & Kremer (2004) |
| 蚊帐 | 免费发放的采用率远高于费用分担 | Cohen & Dupas (2010) |
| 小额信贷 | 对商业收入的影响有限；未带来变革性的减贫效果 | Banerjee et al. (2015) |
| 现金转移（无条件） | 受益者进行了生产性投资；效果持续 | GiveDirectly (Haushofer & Shapiro 2016) |
| 现金转移（有条件，Progresa项目） | 入学率提高8个百分点，营养状况改善 | Schultz (2004) |
| 教师激励 | 激励薪酬提高考试成绩；设计细节至关重要 | Muralidharan & Sundararaman (2011) |
图20.6.RCT功效计算器。查看效应量、方差、显著性水平和集群化如何影响所需样本量。虚线标记80%功效。拖动滑块探索。
凯拉尼的部委预期每月\$30的收入效应（$\sigma = 120$）。在$\alpha = 0.05$、80%功效下：$N = 2 \times 120^2 \times (1.96+0.84)^2 / 30^2 \approx 251$每组。集群随机化（42个村庄，每村60户，ICC = 0.04）：设计效应 = 3.36，有效样本 = 750——远超251。
如果预算仅允许每组1,500：有效样本$\approx 446$。MDE $= \sqrt{2 \times 14400 \times 7.84 / 446} \approx \$22.50$/月——小于预期的\$30效应，因此研究仍有足够功效。
Zambian economist Dambisa Moyo's Dead Aid and her TED talk made the incendiary case: over \$1 trillion in aid to Africa hadn't just failed — it had "created dependency, fueled corruption, and killed African entrepreneurship." Bill Gates publicly called the book "evil." Jeffrey Sachs accused Moyo of advocating policies that would "lead to the deaths of millions." Moyo fired back that Sachs's own Millennium Villages Project was the real failure. The debate went nuclear. But who was actually right about the evidence?
高级托德和沃尔平（2006）将一个结构模型与Progresa RCT进行了验证，然后用它来模拟未经测试的反事实。阿塔纳西奥等（2012）表明CCT主要通过降低上学的机会成本而非放松预算约束发挥作用——一种基于机制的理解，使得可移植性成为可能。
解决方案不是结构对立简约形式——而是结构加上简约形式。RCT提供可信的因果估计；结构模型提供推广的框架。理想工作流程：用RCT识别参数，将其输入结构模型，对照实验数据验证，然后以诚实的不确定性界限进行外推。
图20.8.结构vs.简约形式比较。左面板显示原始RCT估计；右面板显示新站点的预测。随着情境差异增大，结构模型诚实地调整，而天真外推保持虚假的精确度。使用切换按钮切换情景。
米格尔和克雷默在肯尼亚发现旷课率降低25%；在印度的复制发现约3个百分点（不显著）。关键结构差异：蠕虫感染率75%（肯尼亚）vs. 20-30%（印度）；不同的学校质量和可及性；不同的童工机会成本；更小的溢出效应。
一个包含健康投入的上学结构模型，校准至肯尼亚，预测7个百分点。用印度参数重新校准：2-3个百分点——与复制结果一致。模型"知道自己不知道什么"：它调整预测并扩大置信区间，而不是错误地外推。
新结构经济学（林毅夫）认为政府应识别与潜在比较优势一致的产业。罗德里克将此扩展到绿色产业政策：清洁能源转型需要协调的公共投资，因为碳外部性被低估，干中学溢出效应未被内化。
有条件与无条件现金转移（UCT）之间的争论是当代政策的核心。GiveDirectly的项目表明UCT效果良好——接受者进行生产性投资且效果持续。当行为偏差阻碍最优投资时条件性可能重要（联系第19章），但当家庭本身就想投资于儿童的人力资本时可能不必要。
图20.7.现金转移RCT模拟器。调整转移金额、持续时间和条件性，查看处理效应如何因结果变量而异。当CI排除零时出现显著性星号。拖动滑块探索。
殖民时代（1945年前）奠定了制度基础。独立后时代（1945-1980年）以大推进思维为主。华盛顿共识（1980-2000年）推动市场化。RCT革命（2000-2019年）将焦点转向微观层面的证据。2015年后时代进行综合：大问题需要结构思维；具体政策问题需要实验证据。
You've now traversed the full arc: GDP measurement (Ch 7), capital accumulation (Ch 9), endogenous growth (Ch 13), institutions (Ch 18), and the empirical frontier (this chapter). This is the final stop — and the honest resolution is that no single theory wins.
The RCT revolution shows that specific interventions work: cash transfers increase income and welfare (GiveDirectly), deworming has large long-run returns (Miguel & Kremer), and information interventions change behavior. But the effect sizes are small relative to the income gap. A bed net that prevents malaria saves lives but doesn't explain the 50x difference in per capita income. Structural estimation (Buera, Kaboski & Shin 2011) quantifies the contribution of misallocation and market failures to aggregate productivity gaps — and finds that capital market distortions alone can explain a factor of 2-3x in TFP differences. The development economics toolkit now has two layers: RCTs identify local causal effects of specific interventions; structural models embed those effects in general equilibrium to ask about aggregate consequences.
Deaton's critique of RCTs: RCTs answer "did this intervention work in this context?" but not "will it work elsewhere?" or "why does it work?" Without theory, RCT results don't generalize. External validity (§20.7) is the binding constraint. Pritchett's critique: The interventions that RCTs study — bed nets, textbooks, deworming — are too small to explain the development gap. The big drivers are national institutions, industrial policy, and macroeconomic management. You can't randomize a country's institutions. China's challenge: The most dramatic poverty reduction in history (800 million people) happened through domestic policy reform, not through the interventions the aid community studies. China didn't need RCTs; it needed institutional change — and the specific institutional changes it made (dual-track liberalization, SEZs, export orientation) don't fit neatly into any theoretical framework.
The frontier is moving toward combining RCTs with structural models. RCTs identify local parameters; structural models embed them in general equilibrium. This is the "credibility revolution meets structural estimation" synthesis. Simultaneously, the revival of industrial policy (Lin, Rodrik) represents a return to big-picture thinking — but with better empirical discipline than the import-substitution era. The profession is also more honest about what it doesn't know: the historical contingency of development (why Botswana and not Zambia?) may involve path-dependent processes that resist simple causal explanation.
The honest answer to "why are some countries poor?" is: institutions and ideas are the fundamental causes, operating through multiple channels — property rights, human capital, technology adoption, political stability. RCTs help us understand specific mechanisms. Geography and culture interact with institutions rather than being alternatives to them. No single theory explains everything, and the question remains genuinely open. This is itself an important thing for the reader to understand: the biggest question in economics does not have a clean, consensus answer. What we know is that the proximate causes (capital, human capital, TFP) are well-measured, the deep causes (institutions, geography, culture) are genuinely debated, and the policy levers (specific interventions vs. institutional reform) operate at different scales with different evidence bases. The best development economists hold all of these in tension rather than committing to one story.
This is the final stop for BQ02, but the question is far from closed. Industrial policy is making a comeback — does state-led development work? China's growth miracle challenges the "inclusive institutions" story. Climate change threatens to reverse decades of convergence, with the poorest countries bearing costs they didn't cause. The AI revolution could accelerate or widen the gap depending on whether developing countries can adopt and adapt the technology. And the deepest puzzle endures: if we know what "good institutions" look like, why can't countries adopt them? The answer likely involves political economy — those who benefit from extractive institutions have the power to maintain them. The path from knowing what works to implementing it may be the hardest problem in all of economics.
Targeted health interventions work. Governance aid doesn't. The aggregate question is the wrong question.
高级800 million lifted from poverty without inclusive institutions. Exception or alternative model?
高级AJR's settler mortality instrument says institutions are the channel. But institutional persistence is more complex than a single IV.
高级You've seen the comparative advantage case (Ch 2), strategic trade under imperfect competition (Ch 6), and open-economy macro (Ch 17). Now the development perspective: East Asia's success involved strategic trade policy — but most countries that tried the same thing failed.
East Asian development involved export-oriented industrial policy: targeted protection of infant industries, export subsidies, and managed exchange rates — combined with strong human capital investment and macroeconomic discipline. Japan, South Korea, Taiwan, and China all deviated from free trade orthodoxy. This wasn't autarky — it was strategic engagement with global markets. The new structural economics (Lin) argues governments should identify industries consistent with latent comparative advantage and facilitate their development. Rodrik extends this to green industrial policy: the clean energy transition requires coordinated public investment because carbon externalities are underpriced and learning-by-doing spillovers are not internalized. The infant industry argument, dismissed for decades, has returned to mainstream respectability — with important caveats about implementation.
The selection problem: East Asia's success may have been despite industrial policy, not because of it. Countries that tried the same policies in Latin America and Africa failed — import substitution in Argentina, state-led industrialization in Tanzania and Ghana. The difference may be institutional quality, education levels, or cultural factors, not the trade policy itself. China's costs: China used industrial policy aggressively, but it also created massive overcapacity, zombie firms sustained by state banks, environmental destruction, and a real estate bubble. The costs of industrial policy are real and large. Government failure: Picking winners requires bureaucratic competence and insulation from rent-seeking. Most governments lack both. The theoretical conditions for beneficial strategic trade (Brander-Spencer) are knife-edge, and the practical conditions are even more demanding.
The development economics mainstream has softened on free trade absolutism. Rodrik's "industrial policy 2.0" argues for smart, accountable industrial policy with clear exit criteria — not the open-ended protection of the import-substitution era. The climate transition is creating a new rationale: green industrial policy (subsidies for renewables, EVs) is now mainstream in the US, EU, and China. The Stolper-Samuelson losers from trade still haven't been compensated in most countries, and the political backlash (Brexit, Trump tariffs) forced the profession to take distributional effects more seriously.
Pure free trade doctrine was too strong. Trade is beneficial, but the conditions under which strategic intervention works — strong institutions, bureaucratic accountability, hard budget constraints, export discipline — are demanding and uncommon. Most countries that tried industrial policy failed. The few that succeeded (Japan, Korea, Taiwan, China) did so under specific conditions that are hard to replicate. The honest answer: free trade is the right default for most countries most of the time; strategic intervention can work but usually doesn't; and the distributional effects of trade need to be addressed through domestic policy rather than ignored. The climate dimension adds a genuinely new element — carbon border adjustments, green subsidies, and supply chain reshoring are reshaping the trade landscape in ways the textbook framework needs to absorb.
This is the final stop for BQ05, but trade policy is evolving rapidly. Climate policy is reshaping trade: carbon border adjustments are being implemented in the EU, green subsidies are proliferating globally, and supply chain security concerns are driving reshoring decisions. The free trade framework needs to incorporate environmental externalities, geopolitical risk, and supply chain resilience — none of which the standard model handles well. The question "is free trade always good?" may be the wrong framing; the real question is "what combination of openness and strategic policy maximizes inclusive, sustainable development?" — and that question is wide open.
The development experience complicates the textbook answer. East Asia's strategic tariffs worked; Latin America's didn't.
入门China's trade policy defied free trade orthodoxy and produced the fastest growth in history. But the institutional preconditions were unique.
高级You've seen the efficiency-equity tradeoff (Ch 3), externality arguments for redistribution (Ch 4), mechanism design constraints (Ch 12), and optimal taxation (Ch 16). Now the global dimension: within-country inequality is dwarfed by between-country inequality, and the tools for addressing them are completely different.
Within-country inequality (Gini coefficients of 0.35–0.60) is dwarfed by between-country inequality (global Gini approximately 0.70). The richest decile in India earns less than the poorest decile in several OECD countries. Conditional cash transfers (Bolsa Familia, Progresa/Oportunidades) have reduced poverty and inequality in developing countries with modest efficiency costs. Human capital investment — education and health — is both efficiency-enhancing and equalizing: Mincer returns are higher in developing countries (10–14% vs. 5–7%), meaning the marginal year of schooling has larger returns precisely where inequality is greatest. Development economics provides a different set of tools from domestic tax-and-transfer: RCTs for evaluating specific interventions, structural policies for growth, and institutional reform for the deep determinants.
Growth vs. redistribution: In poor countries, growth is far more powerful than redistribution for reducing poverty. China lifted 800 million out of poverty through growth, not transfers. Redistributing a small pie does less than growing the pie. Focus on institutions and growth, not on dividing up what little there is. Against CCTs: Conditional transfers are paternalistic — why not unconditional? Targeting is costly and imperfect: administrative expenses consume resources, and the conditions assume governments know better than households what investments to make. Universal basic income may be simpler and more dignified. The migration question: If between-country inequality is the dominant dimension, the most powerful "redistribution" tool is allowing people to move from poor countries to rich ones. Open borders would do more for global equality than any tax system — but migration is politically unthinkable at the scale required.
The development community has moved toward a both/and position: growth and redistribution are complementary, not substitutes. Pro-poor growth — growth that disproportionately benefits the poor — is the goal. The GiveDirectly experiments on unconditional cash transfers have shown that recipients invest productively and effects persist, weakening the case for paternalistic conditionality. The global inequality literature (Branko Milanovic) has documented a "great convergence" since 2000: between-country inequality has fallen as China, India, and other emerging economies grew faster than rich countries. But within-country inequality has risen in many places, creating the "elephant curve" — global middle classes gained, the very rich gained, and the lower-middle classes of rich countries stagnated.
Inequality is a problem economics can partially solve — but the tools differ by scale. Within countries, optimal taxation and transfer design can reduce inequality with moderate efficiency costs (the Mirrlees-Diamond-Saez framework from Ch 16). Between countries, the answer is growth driven by institutions, human capital, and technology adoption. CCTs and development interventions help at the margin. The profession is more honest about this than it was a generation ago: the efficiency-equity tradeoff is real but smaller than many assumed, moderate redistribution has modest costs, and the biggest inequality is between countries, not within them. The uncomfortable truth is that the most powerful tools for reducing global inequality — institutional reform in poor countries, open migration, and technology transfer — are politically constrained in ways that economics alone cannot solve.
This is the final stop for BQ09, but the inequality frontier is shifting. Climate change is the next great inequality challenge — the poorest countries will bear the largest costs of a problem they didn't create. Climate adaptation finance, loss and damage compensation, and green technology transfer are where the equity question goes next. The AI revolution raises a parallel concern: will AI-driven productivity gains flow to countries and workers that already have the infrastructure to adopt it, or will they reach the global poor? And within rich countries, the political backlash against globalization has made inequality reduction harder, not easier — the distributional losers from trade and technology now vote for protectionism rather than redistribution. Economics can design better policies; whether those policies get implemented is a political question that the discipline is only beginning to engage with honestly.
GiveDirectly's results show unconditional cash works. But scaling from village experiments to national policy is the hard part.
中级Dan Riffle popularized the slogan in 2019. In a development context, within-country wealth concentration meets between-country poverty. The scale mismatch frames the problem differently.
中级凯拉尼实施CCT：每月\$50给2,500个随机选定的农村家庭，条件是80%以上的上学出勤率，为期18个月。对照组：2,500个家庭。功效计算（Eq. 20.10）：$\sigma = 120$时，MDE在80%功效下为每月\$27。预期的\$30-35效应远超这一阈值。
集群随机化（42个处理村 + 42个对照村，ICC = 0.04，集群规模60）产生设计效应 = 3.36。有效样本 = 每组744，超过309的最低要求。预注册结果指标：消费、入学率、膳食多样性、储蓄。
18个月后的结果：月消费+\$32（p < 0.01），学校入学率+8个百分点（p = 0.01），膳食多样性+0.4 SD（p < 0.01），储蓄+\$15（p = 0.02），成人劳动供给-2小时/周（p = 0.27，不显著）。服从率94%；劳动供给担忧被消除。\$50的转移产生\$32的消费增益，暗示存在本地支出乘数效应。
制度分析（第18章）：CCT建设国家能力——支付系统、监测基础设施、官僚问责制。上学出勤条件之所以有效，是因为凯拉尼在2005年改革期间投资了学校建设。没有学校，条件性毫无意义。
外部有效性（第20.7节）：塔拉尼共和国想要复制。简约形式：天真的外推忽略了塔拉尼更弱的制度和不同的人口结构。结构模型：预测入学率+5个百分点（vs. 凯拉尼的+8个百分点），消费+\$28（vs. \$32），入学率的90%区间为[+1个百分点, +9个百分点]。迪顿的批评适用：RCT回答"这里有效吗？"但不回答"那里会有效吗？"
教科书的线索在此汇聚：凯拉尼的发展取决于制度（第18章）、增长基本面（第13章）、宏观经济稳定（第14-16章）、行为洞见（第19章）和循证评估（本章）。
| 标签 | 公式 | 描述 |
|---|---|---|
| Eq. 20.1 | $Y_M = A_M K_M^\alpha L_M^{1-\alpha}$ | 现代部门Cobb-Douglas生产函数 |
| Eq. 20.2 | $Y_S = A_S \min(L_S, \bar{L})$ | 具有剩余劳动力的维持生计部门 |
| Eq. 20.3 | Lewis turning point: $MPL_S = \bar{w} \Rightarrow L_S^* = \bar{L}$ | 剩余劳动力耗尽阈值 |
| Eq. 20.4 | $\dot{k} = sf(k) - (n+\delta)k$, $f$ S-shaped | 具有贫困陷阱的资本积累 |
| Eq. 20.5 | $\pi_i = (1/\alpha - 1)(LF - 1)\alpha^{\alpha/(1-\alpha)}$ | MSV：工业化利润（协调） |
| Eq. 20.6 | $\text{Inst}_i = \alpha + \beta\ln(\text{settler mort}_i) + \mathbf{X}_i'\gamma + \varepsilon_i$ | AJR IV第一阶段 |
| Eq. 20.7 | $\ln w_i = \alpha + \rho S_i + \beta_1 \text{Exp}_i + \beta_2 \text{Exp}_i^2 + u_i$ | 明塞尔工资方程 |
| Eq. 20.8 | $Y = A(H)K^\alpha(hL)^{1-\alpha}$, $h = e^{\phi S + \psi\text{Health}}$ | 增广生产函数（健康+教育） |
| Eq. 20.9 | $\hat{\tau}_{ATE} = \bar{Y}_T - \bar{Y}_C$ | 随机化下的ATE估计量 |
| Eq. 20.10 | $N = 2\sigma^2(z_{\alpha/2}+z_\beta)^2 / \tau^2$ | 功效\$1-\beta$所需最小样本量 |