Impact evaluations, bias, and bias reduction: Non-experimental methods, and their identification strategies

Despite that many scientists believe that experimental designs are the only way to credibly establish a causal relationship, non-experimental designs can also credibly establish a causal relationship. This thesis comprises four studies on impact evaluation using non-experimental designs. They put bias – and bias reduction – back at the centre of the debate on causal inference, emphasizing the need for continued interest and improvement of non-experimental designs as a fundamental alternative to experimental designs. Each chapter of this thesis considers a different setting, focusing on how biases can be minimized. Chapter 2 starts at the macro level, considering the impact of healthcare financing reforms and addressing the self-selection bias that is present when national governments decide to conduct a healthcare financing reform. Next, Chapter 3 looks at the household level, evaluating the impact of a microfinance program in Bolivia where the expansion plans of the microfinance institute is applied to help drawing causal inference. The analysis at the household level continues in Chapter 4 with an evaluation of the impact of a microfinance program in Ghana; a scenario where the evaluation has to be conducted in the absence of a baseline as the project already has started. In an attempt to explain the results of the impact evaluation, the effect of social desirable behaviour on the reported loan use is analysed. The final chapter looks at the individual level, investigating how bias resulting from social desirable and opportunistic behaviour affects the revealed support for Farmers’ Market Organizations in rural Ethiopia.