An extensive research effort is being made to develop the next generation of anode materials for lithium ion batteries. A large part of this effort has been related to silicon, primarily due to its considerable theoretical capacity; however, very limited cycling stability has prevented widespread commercial adoption. A potential solution for this is to use convertible sub-stoichiometric silicon nitride (a-SiNx), which goes through an irreversible conversion reaction during the initial lithiation cycle, producing active silicon domains in an inactive, lithium conducting matrix. Relative to pure silicon, the resulting composite material has gained cycle life at the cost of reduced specific capacity. The specifics of the conversion reaction, however, have not yet been determined; hence, the impact of varying nitrogen content remains uncertain. In this work we develop a model reaction which relates the reversible and irreversible capacities of an electrode to the composition of the conversion products. By fitting this model to experimental data from a large number of a-SiNx thin film electrodes with different thickness and composition, we determine with a high probability that the matrix composition is Li2SiN2. From this, the reversible and irreversible capacities of the material can be predicted for a nitride of a given composition.