Determining the extent to which citation flows, and hence bibliometric indicators based on them, reflect some intrinsic value of scientific works is an important task made very difficult by endogeneity issues. This paper presents an approach which allows to go beyond the abundant anecdotal evidence by testing whether the citation behavior is free from environmental factors. The hypothesis of independence is strongly rejected, providing causal evidence of a Matthew effect at work: namely, the publication of a new work on behalf of an author increases the flow of citations to previous works. Such result is a step towards the estimation of biases affecting bibliometric indicators, at least when interpreted as measures of scientific productivity. The study is based on a novel framework for the study of endogenous network growth subject to constraints. Constraints can be both positive and negative, and change in time depending on the actions of the agents. The framework is not limited to citation networks, and can be applied to any context in which the formation of a link inhibits or implies the formation of another one.