# Skill Formation, Child Labor, and the Schooling Consequences of the World War I Agricultural Boom

> Taylor Jaworski (University of Colorado Boulder and NBER), Carl T. Kitchens (Florida State University and NBER), Luke P. Rodgers (Florida State University)

## What is this paper about?

This paper studies how a large, temporary commodity price boom during World War I affected human capital accumulation in the United States. The central contribution is identifying two channels through which the boom reduced completed schooling: (1) an opportunity cost channel, where higher farm wages pulled teenagers out of school, and (2) a dynamic complementarity channel, where the interaction between early childhood resources and local child labor intensity determined whether younger children gained or lost schooling. The paper shows that the negative aggregate effect on younger children is driven entirely by child labor intensity, not by the resource shock itself.

## Important context

- The paper contributes to the literature on skill formation and dynamic complementarities (Cunha and Heckman 2007, Heckman 2007), extending these models to a historical agricultural setting where child labor mediates the effect of income shocks.
- It is distinct from papers that study commodity booms and schooling in developing countries (e.g., Bau 2020 on India). The US setting features a well-developed school system during the early high school movement, so the tradeoff is between attending high school (not just primary school) and farm work.
- The paper builds a new county-level panel of school enrollment and average daily attendance for 26 states, 1910 to 1930, drawn from state board of education reports. This dataset did not previously exist.
- Linked census data come from the Census Tree and the Census Linking Project, connecting children observed in 1910 or 1920 to their 1940 outcomes. Results are robust across both linking methods.
- The crop revenue index holds crop acreage fixed at 1910 shares and varies only prices, isolating price exposure from endogenous planting decisions. An IV strategy using crop suitability provides additional identification.

## Data and methods

**Data sources:**
- County-level enrollment and average daily attendance, 1910 to 1930 (newly constructed from state education reports)
- Census complete count data: 1910, 1920, and 1940 (linked via Census Tree and Census Linking Project)
- USDA crop price and production data for constructing the crop revenue index
- Controls for compulsory schooling laws, WWI draft exposure, boll weevil infestation, hookworm, malaria, Rosenwald schools, county health organizations, and 1918 influenza mortality

**Identification:**
- The crop revenue index measures county-level exposure to the wartime price boom based on pre-war crop composition (1910 shares multiplied by annual national prices)
- Difference-in-differences for county-level enrollment/attendance outcomes
- Individual-level regressions with county and birth-cohort fixed effects, parental controls, and interactions with child labor intensity for linked census outcomes
- IV using predicted crop shares from soil suitability data (following Kitchens 2023)

**Key specifications:**
- Opportunity cost channel: effect of cumulative exposure during ages 10 to 17 on years of completed schooling
- Dynamic complementarity channel: effect of exposure during ages 0 to 4, interacted with county-level child labor intensity (fraction of children 10 to 14 working in 1910)

## Key results

1. Enrollment and average daily attendance fell sharply at the peak of the boom (1919 to 1920), with larger declines in high-exposure counties.
2. Greater exposure during teenage years reduced completed schooling by 0.27 to 0.47 years, with effects concentrated at the high school margin.
3. Effects are larger for men than women, and present for both white and Black individuals.
4. For children exposed during early childhood (ages 0 to 4), the net effect depends on local child labor intensity. Once child labor is accounted for, the direct effect of early exposure on boys' schooling is approximately zero. In high child labor counties, the interaction drives the negative effect. For girls, the direct effect is weakly positive, consistent with lower opportunity costs in agricultural labor markets.
5. A calibration exercise yields implied discount factors between 0.95 and 0.99, consistent with forward-looking household optimization.

## Limitations and scope

- The analysis covers only the United States in the 1910 to 1930 period. The results may not generalize to other commodity booms, other countries, or settings without a well-developed secondary school system.
- The crop revenue index captures price exposure, not actual farm income. It does not account for changes in input costs, crop failures, or within-county variation in farm operations.
- County-level enrollment and attendance data are available for 26 states. States that did not publish county-level education statistics (including several in the Deep South) are excluded from the county panel analysis.
- Linked census samples are restricted to individuals who can be matched across census years. Linking rates differ by race and sex, and the Census Tree and Census Linking Project use different algorithms with different coverage.
- Child labor intensity is measured at the county level using the 1910 census. This is a proxy for the opportunity cost of children's time, not a direct measure of individual children's labor supply decisions.
- The paper does not identify the effect of the boom on schooling quality, teacher supply, or school infrastructure, only on enrollment, attendance, and years of completed schooling.
- The IV strategy relies on crop suitability as an instrument for price exposure. First-stage F-statistics are strong for white subsamples but weaker for Black subsamples, particularly when additional controls are included.

## Navigation guide

- **For the main argument:** Read Sections 1 (Introduction) and 2 (Investing in Human Capital) for the theoretical framework, then Section 5 (Empirical Analysis) for the core results.
- **For the data construction:** Section 4 describes the crop revenue index, linked census data, and county-level enrollment panel. Section 3 provides historical background on the WWI agricultural boom.
- **For robustness:** Appendix A contains alternative samples (farm owners vs. renters, same-state movers, off-farm residents), alternative linking methods (Census Linking Project), influenza controls, IV estimates, birth order heterogeneity, and the dynamic complementarity results by subgroup.
- **For the calibration exercise:** Appendix B derives implied discount factors following Bau (2020) and reports results under alternative assumptions.
- **Key tables:** Table 1 (main results, opportunity cost channel), Table 2 (young children, dynamic complementarities). Key figures: Figure 1 (agricultural prices and wages), Figure 3 (enrollment and attendance trends).