Propensity score matching (PSM) is the most widely used method of program evaluation in situations when only cross-sectional data are available. PSM matches each participant in the program with a group of non-participants who were equally likely to participate in the program as the participant but chose not to participate. A well-acknowledged weakness of PSM is that its matching procedure is carried out solely based on observable variables, and therefore the resultant propensity score is biased if any unobservable variable plays part in people’s decision whether or not to participate. Here, we show through a simulation analysis that the application of a random-parameter probability model reduces the aforementioned bias, and that this method can be used as an easily implementable alternative to the standard PSM procedures. This bias reduction is attributable to the model's capability to assign a separate estimator for each individual, which can partially "absorb" the effect of the individual's unobservable traits on the participation decision.