Green Revolution Statistics and the Question of Causation
A seminar paper claims: 'Wheat output in a South Asian state rose 60 per cent between 1965 and 1975; therefore the new high-yielding varieties caused the increase.' The author cites state-level production totals, a national fertiliser subsidy figure, and one district-level study of adoption. A reviewer notes that the same decade saw a major canal extension completed in 1968, two consecutive favourable monsoons, and a change in the state's reporting boundaries that added two districts in 1971. The author's district study covers only the canal command area. You must write the reviewer's report.
1. Identify the confounders and explain how each one threatens the causal claim.
Read the worked answer
Four confounders, each fatal to a naive before-and-after inference.
Irrigation. The canal extension completed in 1968 raises yields independently of seed variety, and HYV seed is typically adopted first where water is assured. Irrigation and variety adoption are therefore entangled: the places that got the seed were also the places that got the water.
Weather. Two good monsoons lift output in any technology regime. A decade-long comparison cannot separate a weather effect from a technology effect without year-by-year weather controls.
Boundary change. Adding two districts in 1971 inflates the 1975 total mechanically. Unless the series is rebased to constant boundaries, part of the 60 per cent is an accounting artefact, not a harvest.
Subsidy as national aggregate. A national fertiliser subsidy figure cannot explain state-level output; it is measured at the wrong level and says nothing about how much fertiliser reached this state.
The district study compounds the problem: it samples only the canal command, the one area where the confounders are strongest, so it cannot separate the treatments.
2. Propose a research design that could support a causal claim, and state what data it requires.
Read the worked answer
The design should exploit variation in exposure, not just variation in time.
Core approach: difference-in-differences across districts. Compare output change in districts that received HYV seed early against districts that received it late, before and after the introduction. This nets out statewide weather and price shocks, which hit both groups.
Required data:
- District-level output rebased to constant 1965 boundaries, so the 1971 annexation does not contaminate the series.
- District-level irrigation status and canal command maps, with completion dates.
- District-level fertiliser offtake, not national subsidy totals.
- Annual rainfall or crop-condition data per district.
Strengthen with: a comparison of canal-command and non-command districts within the early-adopting group, which separates the water effect from the seed effect; and a check on whether early adoption was itself driven by pre-existing irrigation, which would violate the parallel-trends assumption.
What to concede: if early-adopting districts were already the irrigated, higher-yield districts, no observational design fully settles causation. The paper should then claim a well-supported association and name the residual ambiguity, which is a stronger paper than an overstated causal claim.