## _wp0794 - 16. First Stage Results for Table 3, Column 2

## Source details

**Canonical URL:** [_wp0794 - 16. First Stage Results for Table 3, Column 2](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp0794.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp0794.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp0794.pdf.json)

---

### Introduction and scope
- Research question: Do short- and medium-term adjustment costs of trade liberalization influence children’s schooling and work decisions in the context of India’s 1991 trade reform?
- Context: India’s August 1991 IMF-supported program led to substantial, largely unanticipated tariff reductions set by the 1991 agreement and implemented through 1997.
- Empirical focus: How schooling and child labor changed across districts differing in prereform industrial composition and in the magnitude of industry-level tariff declines.

### Conceptual framework (mechanisms linking tariffs to schooling)
- Three primary channels through which tariff declines (dt) affect schooling probability:
  - Changes in family income (y0) → affects marginal utility of income and thus schooling decisions.
  - Changes in the child’s economic contribution to the household (w*) → higher w* reduces schooling.
  - Changes in returns to education (r) → lower r reduces schooling if α>0 (parents weight child’s future returns).
- Household decision rule: child attends school if utility from schooling u(yS, s, e) ≥ utility without schooling.
- Empirical focus abstracts from direct consumption-bundle effects in baseline and emphasizes variation in employment exposure to national tariff changes.

### Data and trade reform facts (preserve exact reported values)
- Microdata and sample:
  - Primary microdata: rural samples in the 43rd (Jul. 1987-Jun. 1988) and 55th (Jul. 1999-Jun. 2000) NSS rounds.
  - Sample analyzed: activities of more than 95,000 children age 10-14.
  - District panel: districts matched across rounds; districts treated as relevant local labor markets due to low permanent mobility.
- Aggregate child participation shares (Table 1, children ages 10-14):
  - Attend School: .485 (1983), .550 (1987/88), .667 (1993/94), .727 (1999/2000)
  - Work: .360, .250, .205, .142
  - Work Only: .355, .246, .202, .137
  - Market Work: .193, .138, .109, .076
  - Domestic work: .167, .113, .096, .066
- Trade reform magnitudes:
  - Average tariff declined from 83 percent in 1991 to 30percent in 1997.
  - Average district (employment-weighted) tariff in rural areas: .080 (1987) to .025 (2000) — decline of nearly 70 percent.
  - Average traded-goods (district) tariff in rural areas: .883 (1987) to .308 (2000).
  - Agricultural traded sectors only: .812 (1987) to .230 (2000).
  - Mining and Manufacturing only: .901 (1987) to .337 (2000).
  - Share of merchandise trade in GDP increased from about 10 percent in 1986/87 to about 19 percent in the late 1990s.
  - District variation in tariff change: decline in district tariffs varies between 0 to 59 percentage points.

### Empirical strategy and identification
- Key measure: district-level Tariff_dt constructed as prereform (1991) district-specific employment-weighted average of national industry ad-valorem tariffs at time t:
  - Tariff_dt = Σ_i ω_id * tariff_i,t, where ω_id are 1991 employment shares within district d.
- Traded-only measure (TrTariff_dt): same construction but weights use only employment in traded sectors.
- Main regression forms:
  - OLS base: y_jhdt = β_0 + β_1 Tariff_dt + π(age,gender) + H_ht + λ_d + τ_t + ε_jhdt
  - IV/main specification: Tariff_dt instrumented with TrTariff_dt; includes prereform district characteristics interacted with post-reform indicator.
- Robustness and controls:
  - District fixed effects λ_d and post-reform fixed effect τ_t.
  - Controls for household and child characteristics and age-gender polynomial.
  - Pre-treatment characteristics interacted with post indicator (share employed in agriculture, mining, manufacturing, trade, transport, services; share scheduled caste/tribe; share literate; state labor law indicators).
  - IV first stage: TrTariff_dt strongly correlated with Tariff_dt (first-stage results reported below).
  - Standard errors clustered at the state-year level.

### Main empirical findings (school attendance and magnitudes)
- General result: School attendance increased by less in districts that experienced larger tariff declines.
- Quantitative magnitudes:
  - In districts with no tariff change, regression-adjusted probability a child is in school increases by 17 percentage points between 1987 and 2000 (post-reform effect from OLS).
  - Average district tariff decline (.05) is associated with a 2 percentage point decline in schooling relative to the national baseline.
  - Thus, a district with the average tariff change experienced a 15 percentage point increase in schooling, 12 percent below the national trend (preserved as presented).
  - IV and OLS estimates both indicate larger tariff declines associated with lower schooling attendance relative to national trends; IV is main specification.

### Mechanisms and supporting evidence
- Dominant mechanism: poverty—schooling link operating through schooling costs and reduced household resources.
- Evidence for poverty channel:
  - Topalova (2005): higher exposure to trade liberalization associated with slower poverty reduction relative to national trend.
  - Table 11 results (replication of Topalova):
    - Headcount Ratio: Tariff −0.494** [0.239].
    - Poverty Gap: Tariff −0.195*** [0.069].
    - Log (Agricultural Wage): Tariff 1.051** [0.407].
  - Interpretation: for a district with average change in trade exposure, liberalization increases headcount rate by 2.7 percentage points (nearly 10 percent) relative to a district with no tariff change.
- Schooling-costs evidence:
  - Increased incidence of households taking loans for education and changes in education spending (Table 12):
    - Household Has Education Loan: Tariff −0.030*** [0.010].
    - Household Education Expenditure per capita: Tariff 16.581*** [4.580].
    - Log (1+Hh. Education Exp. per capita): Tariff 1.919*** [0.467].
    - Hh. Education Expenditure as Share of Total: Tariff 0.054*** [0.016].
    - Individual Education Expenditure: Tariff 28.113*** [7.763].
    - Log(1+Indiv. Education Expenditure): Tariff 1.901*** [0.498].
    - Individual Education Expenditure as Share of Total: Tariff 0.045 [0.029].
  - Heterogeneity by baseline schooling-cost mitigation (Table 13):
    - Panel A (mid-day meal): Tariff 0.905*** [0.221]; Tariff X mid-day meal −0.667** [0.299].
    - Panel B (scholarship): Tariff 0.716*** [0.196]; Tariff X scholarship −0.314 [2.995].
    - Panel C (free tuition): Tariff 2.872 [1.813]; Tariff X free tuition −2.223 [1.874].
  - Interpretation: smaller declines in schooling where free mid-day meals, scholarships, or free tuition were more prevalent prereform.
- Child labor demand and returns to education:
  - Little evidence that increased demand for child market wage work or declining returns to education are primary drivers.
  - Child activity patterns (Table 10):
    - All children: Tariff Attend School 0.362** [0.137]; Work only −0.117 [0.110]; Market Work −0.122 [0.111]; Domestic Work 0.05 [0.093]; Idle −0.167** [0.076].
    - Boys: Tariff Attend School 0.261* [0.147]; idle −0.052** [0.022].
    - Girls: Tariff Attend School 0.501** [0.206]; Market Work −0.172 [0.148]; Domestic Work 0.203** [0.099]; Idle −0.329* [0.165].
  - Findings: declines in schooling accompanied by increases in domestic work and increases in children reporting neither work nor school ("idle"); little increase in market wage work by children.
  - Returns to education (Table 9 and Table 8): evidence more consistent with increasing relative returns to education for literate adults with tariff declines (e.g., for literate men tariff −0.210* [0.116] participation; −2.399*** [0.764] days), making declining returns an unlikely explanation.

### Robustness and alternative channels
- Pre-trends:
  - Mean average tariff change −0.055; standard deviation 0.06.
  - Tests using NSS 38th (1983) and 43rd (1987-88) rounds assign prereform tariffs to 38th and post-reform tariffs to 43rd; prereform coefficient opposite in sign and much smaller than pre/post estimate.
  - Allowing prereform trend interacted with post-reform leaves tariff impact similar to main estimate.
- Controls for concurrent reforms:
  - District employment-weighted share of industries subject to industrial licensing and share open to FDI, and number of bank branches per capita included; little change in district tariff coefficient.
  - Including district employment-weighted industry exports likewise shows little sensitivity.
- School infrastructure:
  - No statistically significant correlations between schooling infrastructure measures and tariffs (Table 5).
  - Point estimates indicate larger tariff declines associated with increases in number of primary schools and declines in pupil-teacher ratios (if anything); controlling for number of primary schools per capita has little effect on tariff coefficient.
- Selective migration:
  - Regressions of district log population and male-to-female ratio on tariff and controls show no significant changes; little evidence that selective migration drives results.
- Consumption and input-price channels:
  - Consumption tariff constructed from NSS Schedule 1; consumption tariff estimates imprecise (Table 7).
    - Baseline employment-weighted tariff coefficient 0.362** [0.137] (Table 7, col.1).
    - Including consumption tariff alongside employment-weighted tariff: employment-weighted tariff 0.471* [0.269]; consumption tariff coefficients around −0.122 to −0.151 (imprecise).
  - Input tariff exposure instrumented with traded input tariffs:
    - Input tariff coefficient −0.322 [1.243] (imprecise); employment-weighted tariff conditional on input and consumption tariffs increases slightly but not statistically different from baseline.
  - Authors conclude basic results not driven by consumption or input-price channels, noting imprecision in consumption/input estimates and caution about instrumentation.

### Poverty elasticity of schooling and child labor (IV estimates)
- Instrumenting local poverty by traded tariff (Table 14, Panel A pooled):
  - Poverty (headcount ratio) instrumented by traded tariff:
    - School (attend): coefficient −0.794** [0.366] (column 1).
    - Work (any): 0.303 [0.246] (column 2).
    - Work only: 0.321 [0.246] (column 3).
    - Market work: −0.036 [0.193] (column 4).
    - Domestic work: 0.338 [0.223] (column 5).
    - Idle: 0.473** [0.220] (column 6).
  - Interpretation: a 1 percentage point fall in headcount poverty increases probability a child attends school by 0.7 percentage points (column 1). Girls' schooling and work are more sensitive to poverty than boys'.
  - Back-of-envelope: rural headcount poverty fell from 37 percent in 1987 to 24 percent in 1999; schooling increased from 55 percent to 73 percent for children 10-14. Using column 1 estimate, more than half of the increase in schooling can be explained by falling poverty; over one third of decline in children working without attending school explained by falling poverty.

### Instrumental-variable first-stage (Table 16)
- Dependent variable: district tariff
- Coefficient on District tariffs on traded goods (TrTariff): 0.341*** [0.068]
- F statistic for significance of instrument: 24.88
- R2: 0.92
- Number of observations: 95488
- Standard errors clustered at state-year level. Model controls include third order polynomial in child's age, gender indicator and age-gender interactions; household controls and initial conditions interacting with post-reform indicators as described in notes.

### Conclusion — quantitative summaries and policy implications
- Main empirical regularity:
  - Rural India experienced large increases in schooling and declines in child labor in the 1990s.
  - Districts with larger prereform employment shares in industries experiencing larger tariff declines showed smaller improvements in schooling and child labor relative to national trend.
- Key magnitudes:
  - A district without any change in final product protection experiences a 17 percentage point improvement in schooling rates for children 10-14 between 1987 and 2000.
  - A district with the mean change in protection experiences a 15 percentage point improvement (average tariff decline associated with a 2 percentage point smaller improvement).
  - Average tariff decline associated with a 2 percentage points decline in schooling attendance relative to districts with no tariff change, and a 2 percentage point relative decline in literacy.
- Lifetime implication illustrative calculation (authors' example, numbers preserved):
  - Rural families with a literate adult head have roughly 25 percent higher expenditures per person than families without a literate adult head (treated as an upper bound).
  - The two percentage point decline in literacy is associated with at most a 2.5 percent decline in per capita expenditures per year.
  - Assuming return to literacy constant over life cycle, permanence of decline, individuals become household heads at age 20 and continue to age 64, and a discount rate of 6 percent, lifetime per capita expenditures are at most 41 percent lower relative to an individual living in a community not facing these adjustment costs.
- Policy-relevant implication:
  - Evidence indicates the tariff–schooling link operates primarily through relative increases in poverty and the inability of households to cover schooling costs (credit constraints matter).
  - If poverty is the only channel, estimates imply half of the improvement in schooling and one third of the decline in child labor in rural India in the 1990s can be attributed to falling poverty.
  - Policy attention to human capital consequences of trade adjustment (targeted alleviation of schooling costs, credit access, or schooling subsidies in locally affected areas) is merited given asymmetric incidence of adjustment costs on affected cohorts.

*Content derived from the supplied IMF working paper excerpt (Appendix Table 16 and related sections of _wp0794).*

### References..............................................................................................................

### _wp0794 - References..............................................................................................................

### Figures
- 1. Average Nominal Tariffs .................................................................................................... 10
- 2. Tariffs by Industry Category............................................................................................... 10
- 3. Tariffs and Literacy............................................................................................................. 21

### Text Tables
- 1. Activities of Children in Rural India, 1983-2000 8
- 2. District Tariff Measures in Rural India............................................................................... 13
- 3. School Attendance and Tariffs in Rural India .................................................................... 17
- 4. School Attendance, Tariffs, and Other Reforms in Rural India ......................................... 19
- 5. Schooling Infrastructure and Tariffs in Rural Districts ...................................................... 19
- 6. Population and Tariffs by District, Rural Census Results .................................................. 22
- 7. Rural Schooling Attendance and Alternative District Tariffs............................................. 24
- 8. District Per Capita Consumption, Adult Literacy, and Tariffs in Rural India .................... 26
- 9. Adult Male Employment in Wage Work by Literacy and Tariffs in Rural India............... 27
- 10. Activities of Children by Gender and Tariffs in Rural India ............................................ 29
- 11. Poverty, Agricultural Wages and Tariffs in Rural India................................................... 31
- 12. Educational Expenditures and District Tariffs, Rural India.............................................. 32
- 13. School Attendance, Schooling Costs, and Tariffs in Rural India ..................................... 34
- 14. Activities of Children, Poverty, and Tariffs in Rural India .............................................. 36

### Appendix
- Appendix I. Data ..................................................................................................................... 38

### Appendix Tables
- 15. Descriptive Statistics......................................................................................................... 41

*Source: _wp0794 - References..............................................................................................................*

### 16. First Stage Results for Table 3, Column 2 ........................................................................ 4

### _wp0794 - 16. First Stage Results for Table 3, Column 2

### Introduction and scope
- Research question: Do short- and medium-term adjustment costs of trade liberalization influence children’s schooling and work decisions in the context of India’s 1991 trade reform?
- Context: India’s August 1991 IMF-supported program led to substantial, largely unanticipated tariff reductions set by the 1991 agreement and implemented through 1997.
- Empirical focus: How schooling and child labor changed across districts differing in prereform industrial composition and in the magnitude of industry-level tariff declines.

### Conceptual framework (mechanisms linking tariffs to schooling)
- Three primary channels through which tariff declines (dt) affect schooling probability:
  - Changes in family income (y0) → affects marginal utility of income and thus schooling decisions.
  - Changes in the child’s economic contribution to the household (w*) → higher w* reduces schooling.
  - Changes in returns to education (r) → lower r reduces schooling if α>0 (parents weight child’s future returns).
- Household decision rule summarized: child attends school if utility from schooling u(yS, s, e) ≥ utility without schooling.
- The probability of attending school depends on income, child contribution, returns to education, schooling costs (c), and prices; empirical strategy focuses on variation in employment exposure to national tariff changes (abstracting from direct consumption-bundle effects in baseline).

### Data and trade reform facts (preserve exact reported values)
- Data:
  - Primary microdata: rural samples in the 43rd (Jul. 1987-Jun. 1988) and 55th (Jul. 1999-Jun. 2000) rounds of India’s National Sample Survey (NSS).
  - Sample analyzed: activities of more than 95,000 children age 10-14.
  - District panel: districts matched across rounds; districts treated as relevant local labor markets due to low permanent mobility.
- Aggregate patterns in rural India (Table 1 participation shares for children ages 10-14):
  - Attend School: .485 (1983), .550 (1987/88), .667 (1993/94), .727 (1999/2000)
  - Work: .360, .250, .205, .142
  - Work Only: .355, .246, .202, .137
  - Market Work: .193, .138, .109, .076
  - Domestic work: .167, .113, .096, .066
- Trade reform magnitudes:
  - Average tariff declined from 83 percent in 1991 to 30percent in 1997.
  - Average district (employment-weighted) tariff in rural areas: .080 (1987) to .025 (2000) — decline of nearly 70 percent.
  - Average traded-goods (district) tariff in rural areas: .883 (1987) to .308 (2000).
  - Agricultural traded sectors only: .812 (1987) to .230 (2000).
  - Mining and Manufacturing only: .901 (1987) to .337 (2000).
  - Share of merchandise trade in GDP increased from about 10 percent in 1986/87 to about 19 percent in the late 1990s.
- District variation in tariff change: decline in district tariffs varies between 0 to 59 percentage points.

### Empirical strategy and identification
- Key measure: district-level “district tariff” Tariff_dt constructed as prereform (1991) district-specific employment-weighted average of national industry ad-valorem tariffs at time t:
  - Tariff_dt = Σ_i ω_id * tariff_i,t, where ω_id are 1991 employment shares within district d.
- Traded-only measure (TrTariff_dt): same construction but weights use only employment in traded sectors (nontraded sectors assigned zero tariffs).
- Identification: difference-in-differences exploiting within-district changes in Tariff_dt over time, controlling for district fixed effects λ_d and post-reform fixed effect τ_t, plus household and child controls and age-gender polynomial.
- Main regression forms:
  - OLS base specification: y_jhdt = β_0 + β_1 Tariff_dt + π(age,gender) + H_ht + λ_d + τ_t + ε_jhdt
  - IV/main specification: Tariff_dt instrumented with TrTariff_dt; includes prereform district characteristics interacted with post-reform indicator to allow differential trends.
- Robustness concerns addressed:
  - District fixed effects and post-reform interactions with prereform district attributes (share employed in agriculture, mining, manufacturing, trade, transport, services; share scheduled caste/tribe; share literate; state labor law indicators).
  - IV first stage: TrTariff_dt strongly correlated with Tariff_dt (first-stage results reported in Appendix Table 16).
  - Tests for prereform trends, cohort/placebo checks for older cohorts, region-level analysis, and controls for potential nontraded-sector measurement bias.

### Main empirical findings (school attendance and magnitudes)
- General result: In rural India in the 1990s, school attendance increased by less in districts that experienced larger tariff declines.
- Interpretation context and magnitudes:
  - Aggregate progress: in districts with no tariff change, the regression-adjusted probability a child is in school increases by 17 percentage points between 1987 and 2000 (post-reform effect from OLS reported).
  - Average district tariff decline (.05) is associated with a 2 percentage point decline in schooling relative to the national baseline.
  - Thus, a district with the average tariff change experienced a 15 percentage point increase in schooling, 12 percent below the national trend (calculation preserved as presented).
- Both OLS and IV estimates point to larger tariff declines associated with lower schooling attendance relative to national trends (IV is main specification; standard errors clustered at the state-year level).

### Mechanisms and supporting evidence
- Poverty-schooling link as dominant explanation:
  - Topalova (2005) documents higher exposure to trade liberalization associated with slower poverty reduction relative to national trend in rural India.
  - Evidence consistent with schooling costs channel:
    - Reduced schooling expenditures observed.
    - Increased reports of families taking loans for education.
    - Suggestive evidence: tariff impacts on school attendance more pronounced in areas with higher schooling costs.
  - Child work changes:
    - Children work more in districts with larger tariff declines, but additional work is largely in non-wage domestic activities.
    - Changes in schooling are much larger than the relative increase in work; significant rise in children reporting neither attending school nor working.
  - Little evidence that returns to education or market child labor demand strongly mediate the tariff–schooling relationship.
- Robustness checks:
  - No relationship between tariff declines and changes in school attendance for prereform cohorts (validation check).
  - No relationship between tariff declines and changes in literacy in older cohorts whose education should have been completed before liberalization.
  - Instrumental-variable strategy using TrTariff_dt to address nontraded-sector composition concerns.
  - Additional specifications that allow prereform trends and regional analyses corroborate main findings (details reported in body and appendices).

### Conclusion (empirical implications)
- Tariff declines during India’s 1991–97 liberalization are associated with attenuated gains in school attendance across districts more exposed via prereform employment composition.
- The principal channel consistent with the microevidence is a poverty—schooling connection operating through schooling costs (reduced household resources, increased borrowing for education, and greater incidence of neither school nor work).
- The findings highlight that adjustment costs from trade liberalization can have measurable adverse effects on human capital accumulation among children in the medium term, even amid large aggregate improvements in schooling.

*Source: _wp0794 - 16. First Stage Results for Table 3, Column 2.*

### 4.5 percentage points after the trade reforms (compared to the 17 percentage point rise observed

### _wp0794 - 4.5 percentage points after the trade reforms (compared to the 17 percentage point rise observed

### B. Robustness of Basic Findings
- Estimated attenuation: for almost all districts observed tariff changes are "not large enough to reverse the progress in schooling and child labor reduction in the 1990s in India."
- Tariff change statistics: mean average tariff change −0.055; standard deviation 0.06.
- Pre-trends test:
  - Used NSS 38th (1983) and 43rd (1987-88) rounds (both prereform) with region-level tariff variation (district identifiers not available in 38th round).
  - Assigned prereform tariffs (1987) to 38th round and post-reform tariffs (1997) to 43rd round.
  - Prereform coefficient (column 4) is opposite in sign and much smaller than the pre/post estimate (column 3), indicating pre-existing trends do not drive main result.
  - Allowing prereform trend interacted with post-reform indicator (column 5) leaves tariff impact magnitude and significance similar to main estimate (column 2).
- Other concurrent reforms:
  - Controlled for district employment-weighted share of industries subject to industrial licensing and share open to FDI, and number of bank branches per capita.
  - Including these controls in columns 2–4 of Table 4 does not materially change district tariff coefficient.
  - Including district employment-weighted industry exports (column 5) likewise shows little sensitivity.
- School infrastructure:
  - Data sources: primary schools per capita from 1991 census and 7th AIES; district schooling facilities from 6th (1993) and 7th AIES.
  - Table 5: no statistically significant correlations between schooling infrastructure measures and tariffs.
  - Point estimates: larger tariff declines associated with increases in number of primary schools and declines in pupil-teacher ratios (if anything).
  - Controlling for number of primary schools per capita has little effect on tariff coefficient (Table 4, columns 6–7).

### C. Literacy
- Method: regress district age-specific literacy rates (1991 and 2001 rural population census) on district tariff, post-reform indicator, district fixed effects, prereform district characteristics interacted with post, instrument tariff with traded tariffs.
- Key pattern:
  - Larger tariff declines associated with lower literacy rates for children ages 10-14 in 2001 relative to 10-14 in 1991.
  - Decline in literacy with tariff declines extends to 15-19 age group (they were 5-9 in 1991).
  - No significant association between tariffs and literacy for older cohorts (e.g., age 30 in 2001): falsification consistent.
- Magnitude: basic results in Table 3 compare children ages 10-14 in 1999 with 10-14 in 1987; Figure 3 shows similar magnitudes for literacy effects.

### D. Selective Migration
- Concern: families with higher propensity to educate move away from high-exposure districts.
- Tests:
  - Regressed log population counts by district on district tariff, post indicator, initial characteristics*post, instrumented by traded tariff (Table 6, columns 1–3).
  - No significant changes in population counts with tariff changes.
  - Regressed male-to-female ratio by district on same regressors (Table 6, columns 4–6): no evidence of significant gender composition changes.
  - Conclusion: little evidence of substantive selective migration driving results.

### E. Other Trade Channels (consumption and input price channels)
- Consumption tariff:
  - Constructed district consumption-weighted tariff.
  - Table 7:
    - Column 1: baseline employment-weighted tariff coefficient 0.362** [0.137].
    - Column 2: consumption tariff estimates imprecise; results consistent with increase in schooling of 17 percentage points or decline of 8 percentage points at mean consumption tariff change (caveats noted).
    - Column 3: including both employment-weighted and consumption tariffs yields employment-weighted tariff 0.471* [0.269]; consumption tariff coefficients around −0.122 to −0.151 (imprecise).
  - Authors do not instrument for consumption tariffs; caution about bias if consumption bundles vary with community wealth.
- Input tariffs:
  - Created district exposure to tariffs on inputs, instrumented with traded input tariffs.
  - Table 7, column 4: input tariff declines associated with higher levels of schooling (estimates imprecise: input tariff −0.322 [1.243]).
  - Column 5: employment-weighted tariff conditional on input and consumption tariffs increases slightly but not statistically different from baseline.
- Overall: basic results not driven by consumption or input-price channels.

### V. Mechanisms — Overview
- Candidate mechanisms: declines in returns to education, increases in child labor demand (w*), or increases in poverty/declines in living standards in more exposed districts.
- Main finding: little evidence for declining returns to education or rising child labor demand as primary mechanisms; evidence points to increases in poverty (relative to national trends) and schooling costs as key drivers.

### A. Returns to Education
- Indirect measures examined:
  - Relative per capita expenditure of literate vs. illiterate household heads (Table 8): tariff coefficient negative in most specifications (but standard errors large), consistent with increases in expenditures of literate relative to illiterate with tariff declines.
  - Adult male wage employment by literacy (Table 9):
    - Panel A (illiterate men): tariff 0.112 [0.296] participation; 0.472 [1.786] days — no significant rise in illiterate wage work.
    - Panel B (literate men): tariff −0.210* [0.116] participation; −2.399*** [0.764] days — tariff declines associated with increases in wage work for literate men (negative coefficient indicates relative movement favoring literate).
  - Interpretation: expenditure and adult employment data more supportive of increasing rather than decreasing returns to education with tariff declines; hence declining returns unlikely to explain schooling declines.

### B. Child Labor Demand
- Evidence from child activity categories (Table 10):
  - For all children (Panel A):
    - Tariff coefficient on Attend School 0.362** [0.137].
    - Work only −0.117 [0.110]; Market Work −0.122 [0.111]; Domestic Work 0.05 [0.093]; Idle −0.167** [0.076]; additional idle coefficient −0.240** [0.097] (table lists multiple columns).
  - Gender differences:
    - Boys (Panel B): tariff Attend School 0.261* [0.147]; idle −0.052** [0.022]; others not significant.
    - Girls (Panel C): tariff Attend School 0.501** [0.206]; Market Work −0.172 [0.148]; Domestic Work 0.203** [0.099]; Idle −0.329* [0.165]; other idle coefficient −0.328** [0.129].
  - Patterns:
    - Declines in schooling accompanied by increases in domestic work and increases in children reporting neither work nor school ("idle").
    - Little evidence that children withdraw from school to enter market wage work that generates cash income.
  - Interpretation:
    - Increase in domestic work and idleness could reflect avoiding schooling costs rather than increased market-earning opportunities for children.
    - Domestic work may indirectly increase household capacity to earn (complements), but authors find no evidence adults reduce domestic work with lower tariffs.

### C. Poverty
- Replication of Topalova (2005):
  - Table 11:
    - Headcount Ratio: Tariff −0.494** [0.239].
    - Poverty Gap: Tariff −0.195*** [0.069].
    - Log (Agricultural Wage): Tariff 1.051** [0.407].
  - Interpretation: for a district with average change in trade exposure, liberalization increases headcount rate by 2.7 percentage points (nearly 10 percent) relative to a district with no tariff change.
  - Declines in tariffs associated with declines in agricultural wages.
- Schooling cost evidence:
  - Table 12 (education finance):
    - Household Has Education Loan: Tariff −0.030*** [0.010] (higher incidence of loans in more affected districts).
    - Household Education Expenditure per capita: Tariff 16.581*** [4.580] (decline? table shows positive coefficient; context describes decline in amount spent — preserve raw numbers).
    - Log(1+Hh. Education Exp. per capita): Tariff 1.919*** [0.467].
    - Hh. Education Expenditure as Share of Total: Tariff 0.054*** [0.016].
    - Individual Education Expenditure: Tariff 28.113*** [7.763].
    - Log(1+Indiv. Education Expenditure): Tariff 1.901*** [0.498].
    - Individual Education Expenditure as Share of Total: Tariff 0.045 [0.029].
  - Authors' interpretation: in districts with larger tariff declines, relative increase in households taking out loans to finance education and decline in amount spent on education — consistent with schooling-costs mechanism (note: preserve numbers exactly as reported).
- Heterogeneity by baseline schooling costs (Table 13):
  - Interacted prereform measures of school costs (prevalence of free tuition, share obtaining free mid-day meals, share with scholarships) with district tariff in main specification.
  - Panel A (mid-day meal):
    - Tariff 0.905*** [0.221]; Tariff X mid-day meal −0.667** [0.299].
  - Panel B (scholarship):
    - Tariff 0.716*** [0.196]; Tariff X scholarship −0.314 [2.995].
  - Panel C (free tuition):
    - Tariff 2.872 [1.813]; Tariff X free tuition −2.223 [1.874].
  - Interpretation: declines in schooling relative to trend are smaller in districts with greater baseline prevalence of free midday meals, scholarships, or free tuition — consistent with schooling-costs channel.

### D. Poverty Elasticity of Schooling and Child Labor
- Instrumental variable approach: traded tariff used as instrument for local poverty to estimate poverty elasticity of schooling and child labor (Table 14).
- Pooled results (Panel A):
  - Poverty measure (headcount ratio instrumented by traded tariff):
    - School (attend): coefficient −0.794** [0.366] (column 1).
    - Work (any): coefficient 0.303 [0.246] (column 2).
    - Work only: 0.321 [0.246] (column 3).
    - Market work: −0.036 [0.193] (column 4).
    - Domestic work: 0.338 [0.223] (column 5).
    - Idle: 0.473** [0.220] (column 6).
  - Interpretation:
    - Column 1 implies a 1 percentage point fall in headcount poverty increases probability a child attends school by 0.7 percentage points.
    - Same 1 percentage point decline in poverty associated with 0.3 percentage points decline in probability a child works (imprecise).
    - Lower poverty elasticity of market work versus domestic/idle status, consistent with schooling-costs mechanism.
- Gender differences (Panels B and C):
  - Girls' schooling and work more sensitive to poverty than boys'; girl's responses nearly double boys' responses in many specifications.
- Back-of-envelope: rural headcount poverty fell from 37 percent in 1987 to 24 percent in 1999; schooling increased from 55 percent to 73 percent for children 10-14. Using column 1 estimate, more than half of the increase in schooling can be explained by falling poverty; over one third of decline in children working without attending school explained by falling poverty.

### VI. Conclusion (selected quantitative summaries and policy-relevant implications)
- Main empirical regularity:
  - Rural India saw large increases in schooling and declines in child labor in the 1990s.
  - Districts with larger pre-reform employment shares in industries that experienced larger tariff declines showed smaller improvements in schooling and declines in child labor relative to the national trend.
- Magnitudes:
  - A district without any change in final product protection experiences a 17 percentage point improvement in schooling rates for children 10-14 between 1987 and 2000.
  - A district with the mean change in protection experiences a 15 percentage point improvement in schooling (i.e., average tariff decline associated with a 2 percentage point smaller improvement).
  - Average tariff decline associated with a 2 percentage points decline in schooling attendance relative to districts with no tariff change, and a 2 percentage point relative decline in literacy (Figure 3).
- Lifetime implications illustrative calculation (authors' example preserving assumptions and numbers exactly):
  - Rural families with a literate adult head have roughly 25 percent higher expenditures per person than families without a literate adult head (treated as an upper bound).
  - The two percentage point decline in literacy is associated with at most a 2.5 percent decline in per capita expenditures per year.
  - Assuming return to literacy constant over life cycle, permanence of decline, individuals become household heads at age 20 and continue to age 64, and a discount rate of 6 percent, lifetime per capita expenditures are at most 41 percent lower relative to an individual living in a community not facing these adjustment costs.
- Mechanism and policy implication:
  - Evidence indicates the tariff—schooling link operates primarily through relative increases in poverty and the inability of households to cover direct and indirect schooling costs (credit constraints matter).
  - If poverty is the only channel, estimates imply half of the improvement in schooling and one third of the decline in child labor in rural India in the 1990s can be attributed to falling poverty.
  - Policy attention to the human capital consequences of trade adjustment (targeted alleviation of schooling costs, credit access, or schooling subsidies in locally affected areas) is merited given potentially substantial asymmetric incidence of adjustment costs on affected cohorts.

*Italicized source attribution: Content derived from the supplied IMF working paper excerpt (pages and tables as provided in source content).*

### Appendix I. Data

### Appendix I. Data

### Data sources and sample coverage
- NSS: Employment and Unemployment schedule (Schedule 10), Consumption and Expenditure Schedule (Schedule 1), Employment and Unemployment module, and “thick” rounds referenced throughout (43rd and 55th rounds; 42nd and 52nd rounds used for some education measures).
- Indian Census: 1991 and 2001 district-level data for literacy shares by age/age groups, population counts by age/age group and gender, and village abstracts (1991) for school counts.
- Annual Trade Database compiled by Tips Software Services Pvt. Ltd. (industry exports).
- Handbook of Industrial Statistics (lists of industries for FDI and licensing).
- Directory of Commercial Bank Offices in India (Volume 1), Reserve Bank of India, 2000 (number of bank branches).
- Besley and Burgess (2004) (state labor regulation indicators; 1991 information used).
- Topalova (2005) (headcount ratio and poverty gap computed from NSS).
- Evenson and McKinsey India Agriculture and Climate dataset (agricultural wages, updated to 1998; districts by 1961 boundaries; covers 271 districts across 13 Indian states; deflated by state-specific CPIAL, reference period Oct. 1973-Mar. 1974, from Ozler, Datt and Ravallion (1996)).
- Indian national input-output (IO) table for 1993 (input tariffs).
- All Indian Education Survey (AIES) 6th (1993) and 7th (2002) rounds (post-reform school counts, pupil-teacher ratios).

### Key variable constructions and definitions
- Schooling and Child Labor variables: see Section III.A for NSS details; literacy shares by age/age groups from 1991 and 2001 Census.
- Population counts: 1991 and 2001 rural Census by district, age/age group, and gender.
- Tariffs: see Section IV.A.
- Exports:
  - exp* exp dtidit i ortort ω= ∑  where , , , id id id i Emp Emp ω≡ ∑ is the employment of industry i in district d as a share total employment in district d.
  - Employment by industry and by district from the 1991 Indian Census.
  - Industry exports for 1987 and 1997 are used for the 43rd and 55th round, respectively.
  - Exports data from Annual Trade Database; exports expressed in millions real Rupees.
- FDI:
  - ,, * dtidit i FDIFDILib ω= ∑ , where , , , id id id i Emp Emp ω≡ ∑ is the employment of industry i in district d as a share total employment in district d.
  - FDI is an indicator equal to one if the industry is in the list of industries with automatic permission for foreign equity share up to 51 percent at time t.
  - Lists compiled from Handbook of Industrial Statistics; data for 1987 and 1997 used for the 43rd and 55th round, respectively.
- Industry Licensing:
  - ,, * dtidit i LicenseLicense ω= ∑ where , , , id id id i Emp Emp ω≡ ∑ is the employment of industry i in district d as a share total employment in district d.
  - License is an indicator equal to one if the industry is subject to licensing requirements at time t.
  - Policy details compiled from Handbook of Industrial Statistics; data for 1987 and 1997 used for the 43rd and 55th round, respectively.
- Number of Bank Branches:
  - Number of bank branches per capita = number of bank branches in the district (Directory of Commercial Bank Offices in India (Volume 1), RBI, 2000) divided by district population from the 1991 Indian Census.
  - Note: number of bank branches reported is the total number for the district; rural-only branch counts not available.
- Labor Regulation:
  - State labor regulation indicators from Besley and Burgess (2004) indicating pro employer, pro worker, or neutral based on amendments to the 1947 Industrial Disputes Act.
  - Smaller states not covered in Besley and Burgess (2004) coded as neutral.
  - Information for 1991 used.
- Poverty Measures:
  - Headcount ratio and poverty gap from Topalova (2005).
  - Computed from household expenditure information in "thick" NSS rounds.
  - Measures computed at district and NSS region levels using Deaton (2003a, 2003b) poverty lines and Deaton's methodology to adjust 1999/2000 NSS round for change in recall period.
- Agricultural Wages:
  - Average daily male agricultural wage in a district from Evenson and McKinsey dataset (1971-94 original dataset, updated to 1998).
  - Districts defined by 1961 boundaries; covers 271 districts across 13 Indian states.
  - Deflated by state-specific Consumer Price Index for Agricultural laborers (CPIAL) (reference period Oct. 1973-Mar. 1974) from Ozler, Datt and Ravallion (1996).
- Consumption Tariff:
  - From NSS Schedule 1: construct district-specific consumption weights.
  - Define , ,1987pd consshare as the share of total expenditures in district d in 1987 spent on good p.
  - Consumer perception of tariffs in district d at time t: ,, ,1987, * dtpdpt p ConsTariff consshare Tariff = ∑ .
  - Assumes homogeneous transmission of tariffs across districts within a given product.
- Input Tariff:
  - Use 1993 IO table, 1991 Census, and output tariffs.
  - For industry i, input tariff = weighted average of tariffs on inputs used in production for industry i with weights , ,1993ji sh (industry j's share of industry i's total input cost).
  - District input tariff constructed by weighting industry i's input tariff by i's employment share in the district in 1991:
    - , ,1991 , ,1993, ,1991 * id dtjijt ij d Emp InputTariff sh Tariff TotalEmp ⎛⎞ = ⎜⎟ ⎝⎠ ∑∑
- Educational Loans:
  - Indicator whether household has a loan for educational expense purposes from Employment and Unemployment schedule of the 43rd and 55th NSS rounds.
  - Question asked only to agricultural workers (excluding self-employed or employed elsewhere); covers on average 30 percent of households in a rural district.
- Ratio of per capita expenditure of literate to per capita expenditure of illiterate (PCE literate/PCE illiterate):
  - Computed two ways:
    - From Employment and Unemployment module (Schedule 10) that does not suffer from changes in recall period in 1999/2000 round.
    - From consumption module (Schedule 1).
- Educational Expenditure data:
  - Primary source: Schedule 1 of the 43rd and 55th NSS rounds. Question on educational expenditure changed between rounds from 30 day to 12 month recall period.
  - Expenditures include books and journals, newspapers, periodicals, library charges, stationery, tuition and other fees (school, college, etc.), private tutor/coaching centre (only in 55th round), other educational expenses.
  - Compute per capita household education expenditure (deflated by deflators proposed by Deaton 2003a, 2003b) and share of educational expenditures in household total expenditures.
  - Secondary source: NSS Schedule 25.2 from the 42nd (1986-87) and 52nd (1995-96) rounds (no recall-period change problem but fewer observations). Reports total expenditures on education per child; construct total educational expenditure as a share of total household expenditure and total educational expenditure for each child 10-14.
  - Using 42nd round, compute prevalence at district level of free tuition (free education), mid-day meals, and scholarships among children attending primary school.
- School Infrastructure:
  - Village abstracts in 1991 Census used to construct number of primary schools and total number of schools in rural districts.
  - Post-reform counts from 7th (2002) AIES.
  - Pupil-teacher ratios from 6th (1993) and 7th AIES.

### Descriptive statistics (Table 15: reported numbers are means)
- Child characteristics (NSS)
  - Female.458 .467
  - Age11.785 11.817
- Household characteristics (NSS)
  - Scheduled caste.183 .215
  - Scheduled tribe.099 .106
  - Hindu.843 .830
  - Islam.106 .121
  - Christian.019 .019
  - Sikh.021 .017
  - Head female.085 .086
  - Head age45.077 44.576
  - Head literate.463 .506
  - Head complete primary.139 .123
  - Head complete middle.083 .118
  - Head complete secondary.058 .062
  - Head complete higher than secondary.013 .051
- Household education expenditure information (43rd and 55th round) (NSS)
  - Household has loan for educational expense purposes .003 .002
  - Household educational expenditures per capita2.752 4.597
  - Log (1+hh educational expenditures per capita).647 1.184
  - Household educational expenditure as a share of total household budget.015 .022
- Education Expenditure information for an individual (42nd and 52nd round) (NSS)
  - Educational expenditures 8.660 14.267
  - Log (1+educational expenditures)  1.268 1.791
  - Educational expenditure per child as a share of total household budget.015 .015
- Baseline prevalence (from 42nd round, 1986/87) at district level
  - Baseline prevalence of mid-day meals in a district (1986/87) .192
  - Baseline prevalence of free tuition in a district (1986/87).903
  - Baseline prevalence of scholarships in a district (1986/87).021

- District-level variables (See data appendix)
  - Consumption tariff.567 .177
  - Input tariff.626 .199
  - Exports 25.79 7.5
  - Licensed industries.003 .001
  - FDI.000 .002
  - Number of primary schools per person.001 .001
  - Number of banks per 1000 people.064 .078

- Baseline district characteristics (1991 Census)
  - Emp. share services.072
  - Emp. share transport.013
  - Emp. share trade.033
  - Emp. share mining.005
  - Emp. share agriculture.806
  - Emp. share manufacturing.056
  - Share literate.373
  - Share scheduled caste/scheduled tribe .321

- Other district-level variables (NSS)
  - Headcount ratio.373 .242
  - Poverty gap.089 .048
  - PCE literate/PCE illiterate (schedule 10)1.324 1.248
  - PCE literate/PCE illiterate (schedule 1)1.309 1.247
  - PCE Primary/PCE non-primary (schedule 1)1.401 1.317

- District agricultural wages (log real wages) (Pande, Sharma)
  - 1987 1.715
  - 1998 1.919

- Note: The reported numbers are simple district means. PCE stands for per capita household expenditure.

### Instrumental variable first-stage (Table 16: First Stage Results for Table 3, Column 2)
- Dep. variable: district tariff
- Coefficient on District tariffs on traded goods (TrTariff): 0.341*** [0.068]
- Additional controls included:
  - Demographic characteristics: yes
  - Household characteristics: yes
  - District indicators: yes
  - Post reform indicator: yes
  - Initial district characteristics * post reform: yes
- F statistic for significance of instrument: 24.88
- R2: 0.92
- Number of observations: 95488
- Standard errors in parenthesis are clustered at the state-year level. ***, **, * denote significance at 1, 5, and 10 percentage levels.
- Model controls (as described in notes): a third order polynomial in the child's age, a gender indicator, and a third order polynomial in age interacting with the gender indicator; household controls include scheduled caste/tribe indicator, religion indicators (hindu, muslim, christian, sikh), controls for head of household gender, age, education, literacy; initial conditions interacting with post-reform indicators include percentage of workers in district employed in agriculture, mining, manufacturing, trade, transport, services (construction omitted), share of district’s population that is a scheduled caste/tribe, percentage literate, and state-labor laws indicators.
- Data: forty-third and fifty-fifth rounds of the NSS.

*Source: _wp0794 - Appendix I. Data (IMF working paper content provided).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp0794.pdf_