The heterogeneity of farms and the problem of self-selection are challenging the evaluation of treatments in agriculture. This is particularly the case for rural development measures whit voluntary participation and heterogeneous outcomes. But knowledge about the selection mechanisms for a certain treatment, in combination with econometric methods, can help to overcome these problems. One of these promising methods is the Propensity Score Matching (PSM) approach. In this paper we apply PSM in order to obtain treatment effects from the agricultural investment support programme in Austria on the farm income. We also test the robustness of the results to hidden bias with sensitivity analysis. Furthermore we split the sample in more homogenous subsamples in order to increase the robustness of the results. The results show that treatment effects differ by a large amount for the subsamples and that splitting leads to slightly more robust results.