We use archival data from the Phoenix Deep Survey to investigate the variable radio source population above 1mJy/beam at 1.4GHz. Given the similarity of this survey to other such surveys we take the opportunity to investigate the conflicting results which have appeared in the literature. Two previous surveys for variability conducted with the Very Large Array (VLA) achieved a sensitivity of 1mJy/beam. However, one survey found an areal density of radio variables on timescales of decades that is a factor of ~4 times greater than a second survey which was conducted on timescales of less than a few years. In the Phoenix deep field we measure the density of variable radio sources to be $\rho =0.98\mathrm{deg}^{-2}$ on timescales of 6 months to 8 years. We make use of WISE infrared cross-ids, and identify all variable sources as an AGN of some description. We suggest that the discrepancy between previous VLA results is due to the different time scales probed by each of the surveys, and that radio variability at 1.4 GHz is greatest on timescales of 2 - 5 years.