5.5 million households cultivate CGIAR rice varieties in the irrigated Boro season, according to SPIA estimates using DNA fingerprinting. Self-reporting by farmers showed that 3.5 million households cultivate CGIAR rice varieties in the rainfed Aman season.

Generation three or “G3” Rohu (Labeo rohita) has seemingly spread rapidly out to farmers, despite only being widely available from hatcheries from 2022 onwards. While only 8% of the Rohu-growing households that purchased fingerlings in 2024 reported purchasing the G3 strain, this still represents over 150,000 households because Rohu carps are widely cultivated. 

We expect this number to grow significantly in the years ahead despite the longer-term trend in consolidation towards fewer, larger aquaculture ponds that we observe over multiple waves of the BIHS. 

Two CGIAR-related innovations in water management – Alternate Wetting and Drying (AWD) and the use of Axial Flow Pumps – are both adopted by approximately 1.7 million households. In both cases, the link back to CGIAR research and dissemination efforts is weak or indirect, and may be contested. 

Older rice varieties remain dominant in Bangladesh, despite efforts to reduce the average age of varieties in farmers’ fields. The adoption-weighted average age is 20 years for Boro varieties when we study using DNA fingerprinting, and 23 years when relying on self-reported data from farmers. 

Some limited evidence from photos of seed bags, in combination with DNA fingerprinting of their contents, suggests that newer varieties are sometimes sold with the packaging of older varieties, presumably because farmers are looking for varieties that are tried and tested over many years. 

While many rice farmers (1.7 million households) have patterns of wetting and drying in their plots that are consistent with AWD, this is part of a broad distribution we observe in terms of both the number and duration of drying events. We find zero farmers reporting that they use a perforated pipe to observe the water level in the plot. While it is possible for farmers to move beyond using the perforated pipe once they are familiar with observing the visible features of a drying plot, it is certainly surprising to find zero farmers in our sample reporting using the pipe. Awareness of the concept of AWD is limited in our sample and suggests that the innovation has had only a minor contribution to the aggregate number of farmers we observe with alternately wet and dry plots. We cannot observe their management intentions.

Axial flow pumps (AFPs) are reported as being used quite widely but their use is only weakly attributable to the efforts of CGIAR. However, we do observe much faster rates of adoption in divisions where CGIAR actively promoted AFPs, such as Barisal and Chittagong, where their use by agricultural households with irrigated plots increased from 0.3% in 2015 to 28.9% in 2024. 

By contrast, the corresponding figures for all other divisions are 0.7% in 2015 and 9.3% in 2024. 

New technologies are sometimes introduced at times or in places that lack the necessary data to conduct a well-identified impact evaluation. The use of earth observation (EO) data to complement administrative and survey data can offer the possibility to conduct impact evaluations when traditional economic data is missing. 

However, while EO data has been used to answer many research questions about recent or contemporaneous events, its usefulness in answering economic questions about the past has been limited. An evaluated technology may only be effective under a specific set of circumstances. Understanding this is key to validating the use of EO data to conduct impact evaluations in data-scarce environments.

Reconstructing the dissemination of improved crop varieties for long periods of time in Bangladesh may be time-consuming: old records of agricultural data in many developing countries are either not digitalized or inexistent. When information exists, it is usually in remote locations, poorly documented, or inadequately maintained.

This creates gaps in the information needed which can affect the comparability of data across different periods or locations. Planning the collection of this type of data from the start can facilitate its use in future rigorous impact assessments.

The integration of a variety of different component datasets — such as historical agricultural data, well data on groundwater levels, data from lentil seed distributors, survey data on social and economic outcomes — may not result in the desired overlapping. 

This may restrict the ability to evaluate different types of impacts, especially when it cannot offer a suitable exogenous variation, which is key for identifying impacts.