We measure the mean halo-mass of z ≃ 0.5 Mg II absorbers using the cross-correlation (over co-moving scales 0.05-13h −1 Mpc) between 1806 Mg II quasar absorption systems and ∼ 250,000 Luminous Red Galaxies (LRGs), both selected from the Sloan Digital Sky Survey Data Release 3. The Mg II systems have λ2796 rest-frame equivalent widths W λ2796 r 0.3Å. From the ratio of the Mg II-LRG cross-correlation to the LRG-LRG auto-correlation, we find that the bias ratio between Mg II absorbers and LRGs is b Mg II /b LRG = 0.65 ± 0.08, which implies that the absorber host-galaxies have a mean halo-mass ∼ 20-40 times smaller than that of the LRGs; the Mg II absorbers have haloes of mean mass log M h (M ⊙ ) = 11.94 ± 0.31(stat) +0.24 −0.25 (sys). We demonstrate that this statistical technique, which does not require any spectroscopic follow-up, does not suffer from contaminants such as stars or foreground and background galaxies. Finally, we find that the absorber halo-mass is anticorrelated with the equivalent width. If Mg II absorbers were virialized in galaxy haloes a positive M h -W λ2796 r correlation would have been observed since W λ2796 r is a direct measure of the velocity spread of the Mg II sub-components. Thus, our results demonstrate that the individual clouds of a Mg II system are not virialized in the gaseous haloes of the host-galaxies. We review past results in the literature on the statistics of Mg II absorbers and find that they too require an M h -W λ2796 r anti-correlation. When combined with measurements of the equivalent width distribution (d 2 N/dz/dW r ), the M h -W λ2796 r anti-correlation naturally explains why absorbers with W λ2796 r 2Å are not seen at large impact parameters. We interpret the M h -W λ2796 r anti-correlation within the starburst scenario where strong Mg II absorbers are produced by supernovae-driven winds.