Estimation of causal effects with small data in the presence of trapdoor variables