We experimentally test a game theoretical model of researcher-evaluator interaction à la Di Tillio, Ottaviani, and Sørensen (2017a). Researcher may strategically manipulate sample selection using his private information in order to achieve favourable research outcomes and thereby obtain approval from Evaluator. Our experimental results confirm the theoretical predictions for Researcher’s behaviour but find significant deviations from them about Evaluator’s behaviour. However, comparative statics are mostly consistent with the theoretical predictions. In the welfare analysis, we find that Researcher always benefits from the possibility of manipulation, in contrast to the theoretical prediction that he sometimes is hurt by it. Consistent with theoretical predictions, Evaluator benefits from the possibility of Researcher’s manipulation when she leans towards approval or is approximately neutral but is hurt by that possibility when she leans against approval.