Estimation of causal effects of participation in a program or treatment on some outcome of interest can play an important role in many areas including economics. In this paper I consider methods for estimation of average treatment effects that assume that treatment assignment is unconfounded with outcomes conditional on the set of relevant covariates. I present the estimators based on weighting by propensity score, the conditional probability of receiving the treatment and the estimator that combines weighting with the regression approach, called a doublyrobust estimator, because it is consistent even if one of the underlying models used for regression or weighting is misspecified. Along with the outlined estimation methods I describe the theoretical properties of the estimators that lead to the asymptotic standard error estimation. I then apply these methods to estimate the average treatment effect of participation in two types of active labour market programs in Austria, active job-search programs and training programs, on the future employment of individuals after four years following the program start. I use an individual administrative data set from the analysis of Hofer, Sellner and Weber (2007). According to the results of doubly-robust estimator active job-search programs perform better than the training programs particularly for women, where estimated effects are significantly positive for both programs. For men a negative treatment effect is estimated in case of training programs. The results suggest the presence of the stronger lock-in effect of training programs comparing to job-search programs, especially for men, that is due to the initial reduction in the job search effort during the training program participation. Finally I look at the average treatment effect estimated in each quarter up to twenty quarters after the program start what provides a further evidence for the later occurrence of the positive treatment effect for training programs compared to the active job-search programs.