This article examines the effectiveness of a government funded extension programme. Farm-level data are used to assess the economic impact of dairy discussion groups, a common participatory extension method. The evaluation focuses on whether discussion group participants have improved farm profits, which is estimated with an endogenous switching regression model. This method controls for self-selection bias due to unobserved characteristics, such as the farmer’s ability, that may affect both participation and farm profitability. After controlling for this potential bias, we find that the economic returns to discussion group membership are positive, thus supporting government targets to enrol more farmers in discussion groups.