Whenever we see the world we rarely observe all the variables of any system of interest at the same time. However, we can create hypotheses of causal relations by combining information of these partially observed scenarios. In this research, we will focus on the question of extracting causal information from scenarios where we don't observe all the variables jointly but only partial combinations thereof. The research will have two orthogonal development axes: on the one hand we will focus on developing methods, based on statistical modelling techniques, for example the Maximum Entropy principle, or Bayesian inference. On the other hand we will prove the empirical utility of the methods in real world applications like meta-analysis, where we merge information from expert opinions, or computer vision, where we always observe a very limited amount of information of the system in each image.