Forced labour risk is pervasive in the US land-based food supply

The paper addresses how much forced labour risk is embedded in a country’s land-based food supply, using the United States as a case study. It situates the research within the UN SDGs, emphasizing SDG 8.7 on eliminating forced labour, and notes that agriculture, fishing and forestry have high incidence of forced labour, often involving vulnerable migrant workers. Prior country-level social performance analyses have not linked social risks to specific food commodities at scale, limiting policy relevance. Detection is challenging due to complex global supply chains and the illicit nature of forced labour, yet evolving regulations (trade sanctions, human rights due diligence) increase the need for robust indicators. Building on earlier work that identified risks in fruits and vegetables consumed in the US, the study aims to expand the risk scoring method to include processing stages, quantify forced labour risk across diverse foods considering trade linkages, and identify risk hotspots within and across food categories.

The authors reference prior sustainability assessments that take country-level lenses but lack commodity-level resolution for social performance. They highlight the ILO definition of forced labour and evidence that agriculture, fishing and forestry have high incidence, especially among migrant workers. Previous work by Blackstone et al. (2021) quantified forced labour risk in US fruits and vegetables, but comprehensive inclusion of processing and diverse foods was missing. The paper also notes critiques of social risk indices that rank countries and may bias results, calling for standardized benchmarks and commodity/sector-specific evidence. The evolving regulatory landscape (e.g., import bans, human rights due diligence) underscores demand for better data and methods. Overall, the literature indicates gaps in high-resolution social risk assessment across full food supply chains and stages.

Scope: Land-based US food supply (excluding seafood); includes agriculture and up to three processing stages. Timeframe: 2015–2019 averages to smooth interannual variability. Products and origins: FAOSTAT Supply Utilization Accounts (SUA) used to estimate US supply; FAO Detailed Trade Matrix used to disaggregate imports by partner country. Commodity trees and extraction rates from FAO linked primary products to processed products, mapping multi-stage processing back to origins. Feed production upstream of animal products was excluded. Risk coding framework: A three-step qualitative coding approach assessed known occurrences of forced labour and government response for each country–commodity–stage combination. Step 1 used commodity-country-specific sources; Step 2 used sector-country-specific sources; Step 3 used country-level sources. Known occurrences contributed 85% of the qualitative risk level; government response (US Trafficking in Persons Report tiers) contributed 15%. Data sources included US Department of Labor’s List of Goods, US Department of State reports, Verité reports, investigative journalism (systematically screened in Nexis Uni; 2016–2019; n=709 articles double-coded), and hand-harvest indicators. Prison labour was coded as very high risk where applicable. When multiple sources conflicted, a mini-Delphi process was used to reach consensus. Quantification: Qualitative risk levels were converted to characterization factors (mrh-eq per worker hour) using a reference scale S-LCA approach adapted from the Social Hotspots Database (very high=10, high=5, medium=1, low=0.01, very low=0.001 mrh-eq). Labour intensity (worker hours per tonne) was estimated as price (US$ per tonne) times working hours per US$ of output (from SHDB/GTAP sector data). Price estimation used a hierarchy combining FAO producer prices and trade unit values, adjusted via GTAP-based correction factors; outliers were winsorized at the 5th/95th percentiles. Risk per unit output for each origin–product–stage was computed, then aggregated across stages using extraction rates and supply shares to obtain unweighted and weighted risks per tonne, and finally multiplied by per capita supply to obtain risk per capita (mrh-eq per capita). Equations (1)–(5) formalize these steps. Data quality: A pedigree matrix assessed reliability, temporal, geographical, and technical quality for risk coding, working hours, and prices (scores 1=high to 5=low). Overall data quality per observation averaged across indicators and data components. Conservative coding rules were applied (e.g., very high only for Step 1 occurrences; very low only for Step 3 country-level). Software: Data managed in Excel, Tableau Prep (v2022.3.1) and analyzed/visualized in Tableau Desktop (v2022.2.4).

• Dataset composition: 212 food products; 1,312 product–country combinations; 2,661 activity–country combinations scored (stages across origins). Of all activity–country combinations, 18% used Step 1 (commodity-specific), 49% Step 2 (sector-specific), and 33% Step 3 (country-level) data.
• Data availability: For agriculture, Step 1 data existed for 27% of combinations across 81 countries (11% government/NGO, 18% investigative journalism, 71% hand-harvest indicators). For processing stages, Step 1 data covered only 4% of combinations (most in US and Canada).
• Category hotspots (risk share): Meat, poultry and eggs (28%); Other products (discretionary foods like sweeteners, beverages, chocolate) (23%); Processed fruits and vegetables (18%). The latter two categories’ risk contributions exceeded their mass and economic value shares, indicating disproportionate risk.
• Domestic vs imported risk: 62% of total forced labour risk in the US land-based food supply was attributable to domestic (US) production or processing. Considering agriculture alone, 51% of risk came from US origins.
• Country contributions: China and Mexico were the second and third highest contributors at 13% and 8% of total risk, respectively. For China, 76% of imported risk was from apple juice concentrate; China supplied about 60% of US apple juice concentrate (≈766,830 tonnes/year, 2015–2019). For Mexico, 58% of imported risk was in unprocessed fruits and vegetables (notably avocados, tomatoes, chillies/peppers).
• Stage contributions: Across the entire supply, agriculture accounted for 85% of risk and processing 15%. However, processing contributions varied by product and reached high levels: maize starch (94%), frozen potatoes (66%), beer (52%), and shelled cashews (42%).
• Product-level hotspots and per-tonne risk (mrh-eq/t): • Fruits: avocados 1,159; lemons and limes 238; pineapples 225. Vegetables: chillies/peppers 434; tomatoes 215. • Processed produce: apple juice concentrate 7,779, with high risk due to high extraction (≈10 t apples per t concentrate) and reliance on Chinese imports. • Pulses/nuts: shelled cashews 15,741; 79% of US supply from Vietnam assessed as very high risk in both agriculture and processing. • Grains: rice 153 overall per tonne; risk varied by origin. • Animal-based: boneless beef 1,754; ≈90% US-sourced (agriculture very high; processing high). Dairy: skimmed dried milk 1,449; skimmed cheese 1,337; >99% US-sourced with very high (agriculture) and high (processing) risk; product yields (~10 t milk per t product) drive per-tonne risk. • Other products: cocoa powder/cake 20,999 (second highest per tonne in dataset) with agriculture-stage risk concentrated in Côte d’Ivoire and Ghana (84% of cocoa agriculture-stage risk); refined sugar 457 per tonne, with US processing (74% of end-product supply) assessed as very high and agriculture-stage risk concentrated in the Dominican Republic, Mexico, and the US (79% of sugar agriculture-stage risk).
• Data quality assessment: Observations spanned a continuum; many had medium quality working hours data. High-risk, high-quality combinations (e.g., Mexican avocados) flag priorities for action; low-quality combinations indicate targets for additional data collection and method refinement.

The study demonstrates that forced labour risk is widespread across the US land-based food supply and is not confined to a handful of well-publicized commodities or to imports from lower-income countries. Contrary to common assumptions, a majority of risk is domestic, reflecting the scale of US production/processing and systemic vulnerabilities in US food-sector labour (e.g., dependence on migrant labour and coercive features of tied-visa programs). Methodologically, integrating agriculture and processing stages and tracing risk through commodity trees and trade linkages reveals hotspots that traditional case studies and intermittent social audits often miss. Findings intersect with environmental and nutrition considerations: significant risks in red meat, juices, and refined sugars overlap with areas identified in environmental and health assessments, suggesting both synergies and tensions for multi-criteria policy design. The paper also engages with critiques of country-ranking risk tools, prioritizing commodity/sector-specific evidence and standardized benchmarks to reduce bias. Incorporating investigative journalism improved coverage in overlooked sectors and geographies but introduces potential reporting biases; the authors’ coding protocol and triangulation partly mitigate these. The results support a strategy that couples import controls (e.g., WROs) with robust domestic regulation, monitoring, and enforcement, and elevates worker-centered initiatives (e.g., Fair Food Program, Milk with Dignity) as models for achieving decent work. Improved, dynamic risk monitoring and better data on working hours can enhance decision relevance.

By extending a social risk assessment framework to include multiple processing stages and complex trade linkages, the study provides a high-resolution, quantitative map of forced labour risk embedded in the US land-based food supply. Major contributions include identification of category- and product-level hotspots (notably animal products, processed produce, and discretionary foods), quantification of the substantial domestic share of risk, and demonstration of how processing stages can dominate risk for certain products. The approach advances S-LCA practice beyond agriculture-only assessments and highlights the need to align policy, business due diligence, and worker-centered programs to reduce reliance on exploited labour. Future research should incorporate seafood, upstream feed production, transport, retail and waste stages; refine working-hours data and price estimation; expand commodity-specific evidence for processing; and develop dynamic monitoring frameworks that connect macro risk indicators with on-the-ground worker experiences.

• Scope exclusions: Seafood and upstream animal feed production were excluded; other supply chain stages (transport, retail, waste) were not assessed. The method is not a full cradle-to-grave S-LCA.
• Data gaps and resolution: Limited commodity-specific data for processing stages (only 4% Step 1); reliance on sector (Step 2) or country-level (Step 3) data where higher-resolution evidence was unavailable may under- or over-estimate risk, particularly in high-income countries.
• Data quality: Working hours and price data carry medium quality on average; unit value estimation required heuristics and GTAP-based adjustments; outliers handled via winsorization. Excessive overtime is itself an indicator of forced labour, complicating the use of working hours as a scaling factor.
• Source biases: Investigative journalism expands coverage but may reflect access and reporting biases; government indices (e.g., TIP, GSI) have been critiqued for political or methodological biases. The study period (2015–2019) and source selection may miss temporal changes (e.g., COVID-19 impacts).
• Modeling assumptions: Use of general FAO commodity trees and extraction rates (not country- or time-specific) and averaging of 2015–2019 values may mask variability; products with multiple processing stages and complex origin mixes introduce uncertainty.