This paper examines the socio-economic implications of scaling back China’s Sloping Land Conversion Programme (SLCP), one of the world’s largest payments for ecosystem services programmes. Using the consolidation phase of SLCP as a natural experiment, a staggered difference-in-differences design is applied to a panel of rural villages (and households) located in poor and ecologically fragile areas. SLCP consolidation, consisting of a 50% subsidy reduction and suspension of a job training programme, leads to larger income declines in treated villages driven by a reduction in the likelihood that SLCP households transition from farm to non-farm work. These harmful socio-economic effects disappear, however, in villages with stronger land rights and more job training prior to consolidation. When these favourable site-specific characteristics exist, SLCP households remain equally engaged in off-farm work even after subsidy payments and training resources are scaled back. Moreover, on aggregate, village income losses are no longer detectable. The findings provide partial support in favour of the behavioural permanence thesis, highlighting the important role of prevailing local characteristics in mitigating harmful development effects otherwise brought about by programme roll-back.